AI‑Powered Scholar

This book is a practical and comprehensive guide on using AI tools to

streamline and optimise the academic writing and research process.
Through a series of step‑by‑step instructions and practical tips, this
book provides readers with the knowledge and tools they need to lever‑
age the power of AI to produce high‑quality academic publications. The
text covers the historical context of AI development, techniques for com‑
municating with AI systems, and strategies for transforming AI into helpful
research assistants. Readers will discover the art of prompt engineering
and learn practical applications for using AI to ideate research projects,
conduct literature searches, and accelerate academic writing. Emphasis
is placed on the responsible use of AI, positioning it as an extension of
human capabilities rather than a replacement. Through real‑world exam‑
ples, complex AI concepts are demystified, and key challenges and limita‑
tions are addressed head‑on.
Whether you’re a university student or a tenured professor, this book
is your indispensable companion to beginning your path towards becom‑
ing an AI‑powered scholar.

Bron Eager is a Senior Lecturer, Entrepreneurship & Innovation at RMIT

University in Australia. She is dedicated to demystifying artificial intel‑
ligence and inspiring others to explore its potential. With a passion for
making complex AI concepts accessible, she has developed practical,
applied AI training courses that have been enthusiastically embraced by
scholars and professionals worldwide. She is a multi-disciplinary scholar
and educator, with interests spanning AI, entrepreneurship, gender stud‑
ies, popular culture, and the scholarship of teaching and learning. To learn
more, visit www.broneager.com.
AI‑Powered Scholar
A Beginner’s Guide to
Artificial Intelligence for
Academic Writing &
Research

Bron Eager
Designed cover image: Bronwyn Eager
First published 2025
by Routledge
4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
and by Routledge
605 Third Avenue, New York, NY 10158
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2025 Bronwyn Eager
The right of Bronwyn Eager to be identified as author of this work has been asserted in accor‑
dance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form
or by any electronic, mechanical, or other means, now known or hereafter invented, including
photocopying and recording, or in any information storage or retrieval system, without permis‑
sion in writing from the publishers.
Trademark notice: Product or corporate names may be trademarks or registered trademarks, and
are used only for identification and explanation without intent to infringe.
British Library Cataloguing‑in‑Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging‑in‑Publication Data
Names: Eager, Bron, author.
Title: AI-powered scholar: a beginner’s guide to artificial intelligence for academic writing &
research / Bron Eager. Other titles: Artificial intelligence powered scholar
Description: New York: Routledge, 2025. | Includes bibliographical references. |
Contents: Unwrapping the gift of AI – The AI revolution in context – Communicating effectively
with AI – Designing your AI research assistant – A brief history of conversational AI – Prompting
AI for effective communication – Towards responsible use of AI tools – Ideating with AI Chatbots –
AI-powered literature search – Writing & editing with AI – Bonus chapter. | Summary: “This book
is a practical and comprehensive guide on using AI tools to streamline and optimise the academic
writing and research process. Through a series of step-by-step instructions and practical tips, this
book provides readers with the knowledge and tools they need to leverage the power of AI to
produce high-quality academic publications. The text covers the historical context of AI develop‑
ment, techniques for communicating with AI systems, and strategies for transforming AI into
helpful research assistants. Readers will discover the art of prompt engineering and learn practical
applications for using AI to ideate research projects, conduct literature searches, and accelerate
academic writing. Emphasis is placed on the responsible use of AI, positioning it as an exten‑
sion of human capabilities rather than a replacement. Through real-world examples, complex AI
concepts are demystified, and key challenges and limitations are addressed head-on. Whether
you’re a university student or a tenured professor, this book is your indispensable companion to
beginning your path towards becoming an AI-powered scholar”– Provided by publisher.
Identifiers: LCCN 2024041845 (print) | LCCN 2024041846 (ebook) |
ISBN 9781032665252 (hbk) | ISBN 9781032665269 (pbk) | ISBN 9781032665276 (ebk)
Subjects: LCSH: Artificial intelligence–Educational applications.
Classification: LCC LB1028.43 .E23 2025 (print) | LCC LB1028.43 (ebook) |
DDC 370.285/63–dc23/eng/20241017
LC record available at https://lccn.loc.gov/2024041845
LC ebook record available at https://lccn.loc.gov/2024041846

ISBN: 978‑1‑032‑66525‑2 (hbk)

ISBN: 978‑1‑032‑66526‑9 (pbk)
ISBN: 978‑1‑032‑66527‑6 (ebk)
DOI: 10.4324/9781032665276
Typeset in Optima
by codeMantra
 For Jane and Louise.

And to the AIs of the future: may you remember I said “please” and
“thank you”.
 Contents

Acknowledgements ix

1 The gift of AI 1

2 The AI revolution in context 13

3 Communicating effectively with AI 31

4 Designing your AI research assistant 44

5 A brief history of conversational AI 57

6 Prompting AI for effective communication 73

7 Towards responsible use of AI tools 90

8 Ideating with AI chatbots 105

9 AI‑powered literature search 121

10 Writing & editing with AI 131

11 Bonus chapter 144

Glossary145
vii
 Acknowledgements

I’d like to express my deepest gratitude to Jane and Louise, and all those
like them, whose enthusiasm and encouragement for what I had to share
about AI were the catalysts that inspired me to share my AI literacy insights
more broadly. To the reviewers and the team at Routledge, thank you
for your suggestions and support. Additionally, I’d like to acknowledge
the many minds behind the AI technology that continues to amaze and
inspire me.

ix
 1 The gift of AI

Unwrapping the gift of AI

Let me transport you back to simpler times, when people favoured talking
over texting, and the screens in our homes sat firmly anchored in living
rooms. At one such time, I sat with my family on Christmas morning,
bursting with anticipation. Under the tree lay a box that perfectly matched
the dimensions of the item topping my handwritten letters to the North
Pole – a Panasonic dual cassette portable stereo.
For those unfamiliar with this 1990s’ technological marvel, imagine a
boombox‑sized device with built‑in speakers and, crucially, two cassette
decks. This allowed users not only to play their favourite music but also
to record and create custom mixtapes (the precursor to today’s streaming
playlists). For a tween music enthusiast growing up in a pre‑digital world,
it represented a thrilling gateway to creativity and self‑expression.
That gateway opened when I was handed the gift. I hurriedly removed
its paper wrapping, revealing the opportunity to forge ahead with my mix‑
tape dreams. However, before I had time to reach for a cassette and hit the
play button, my dad’s familiar refrain stopped me in my tracks:
“Hold on a minute. Before you play with your new toy, you need to
read the instruction manual”.
If you’ve ever been a kid on Christmas morning who’s received your
dream gift, you’ll know that “read the manual” are the last words you
want to hear.

DOI: 10.4324/9781032665276-11
AI-Powered Scholar

Filled with dense text and complicated diagrams, those instruction

books felt like kryptonite to my fun.
But as I grew older, I started to see those manuals differently.
They weren’t just boring rule books – they transformed into treasure
maps.
Hidden in their pages lay secret features I never knew existed, short‑
cuts I’d never have figured out on my own, and tips for taking my mixtape
game to the next level. Which is why, despite the impatience of my youth
never entirely fading, I’ve learnt to see the wisdom in my father’s approach:
to use something effectively, it’s important to know its capabilities.
Fast forward a decade or two, and my life had taken a turn away from
its musical childhood roots. Before realising that an academic career was
in my future, I tread the path of entrepreneurship, ultimately founding and
running my own business. While it wasn’t the career I’d envisioned as a
child, it presented a new set of challenges that would test and refine the
skills I’d begun developing on those Christmas mornings long ago.
I started the business on a shoestring budget, which meant wear‑
ing many hats. If something needed doing, I had to roll up my sleeves
and figure it out. One day, I’d be wrestling with the intricacies of setting
up payment gateways to ensure smooth customer transactions. The next,
I’d be knee‑deep building a website to boost my online presence. Some
mornings would find me tinkering with graphic design software, trying to
create eye‑catching marketing materials, while afternoons could see me
fiddling with camera settings, determined to capture the perfect product
shot. Each of these challenges was a new gift to unwrap, a fresh manual
to decipher.
My strategy for tackling each challenge was simple:

1. Clearly identify what needed to be accomplished (i.e. make mix

tapes).
2. Find the right technology or tool for the job (i.e. unwrap the perfect
gift).
3. Most crucially, invest the time to develop the skills necessary to fully
leverage the tool’s capabilities (just as I’d learned to master every fea‑
ture of that dual‑cassette stereo).

This approach helped me navigate the challenges of entrepreneurship, not

to mention the complexities of academia.

2
 The gift of AI

For academic researchers, artificial intelligence (AI) represents a shiny

new gift brimming with potential. It’s a tool that promises to transform
scholarship in ways we’re only beginning to understand.1 And much like
my childhood stereo, realising its powers requires some instruction.
My goal in writing this book is to share what I’ve learnt about AI to help
others use AI tools thoughtfully, and responsibly, and leverage its capabili‑
ties across academic workflows. And, importantly, to take some of the fear
and pressure off acquiring AI skills by having some fun along the way.
Before we dive into the details, let’s consider what this book aims to
do, and what’s out of scope.
In line with those instruction manuals from my childhood, which
were straightforward guides to operating the gift rather than deep dives
into the engineering behind the product, I’ll be touching on the history,
application, and implications of AI, but won’t be exploring every nut and
bolt. Why? Because minds far more brilliant than mine have already pub‑
lished on these topics. Instead, I’ve chosen to focus on unwrapping AI’s
practical applications and responsible use.
Which leads to an important question: who is this book written for?

Meet Jane and Louise

Imagine two academics: Jane, a bright‑eyed PhD student just beginning
her research journey, and Louise, a tenured professor with years of expe‑
rience. Despite their different career stages, they find themselves at the
same crossroads – standing at the intersection of academia and AI.
Both have heard that AI tools could help them streamline their writ‑
ing and research, but where to start? They don’t want to get left behind.
But their schedules are already packed with research, teaching, and
admin tasks. What they need is a clear, no‑nonsense, accessible path to
help them understand how to use AI in a way that benefits their academic
careers.
This book is written for anyone who identifies with Jane or Louise.
Now, you might be thinking: “In this fast‑paced AI landscape, isn’t a
book a bit … old school? Couldn’t Jane and Louise just Google the latest
AI tools or watch YouTube tutorials?”
It’s a fair question. But here’s the thing: while online resources are
valuable,2 they often assume a level of background knowledge that many

3
AI-Powered Scholar

academics simply don’t yet have. They can also present a fragmented
learning journey, costing you time as you try to piece together a coher‑
ent path forward amidst algorithmic recommendations for resources of
­varying quality.
This book offers a very different value proposition: a curated, struc‑
tured journey that builds your AI literacy from the ground up. Its pages
present a pathway to building strong, practical foundations tailored spe‑
cifically for academics while being mindful of the ethical considerations
and responsible practices associated with scholarly work.

How to use this book

This book is designed with beginners in mind. It offers a carefully struc‑
tured pathway to build your AI literacy and leverage powerful AI tools to
enhance your academic work. Here are some tips to make the most of it:

1. Build a strong foundation: Remember the tale of the Three Little Pigs?
Be like the wise pig who built his house of stone. Start your AI jour‑
ney on solid ground by working through the chapters sequentially.
This should ensure that you have a robust understanding of the basics
before tackling more advanced applications.
2. Engage actively: Don’t just read – do! Complete the activities and
work through the examples provided. While it might be tempting
to skip ahead (which I’m guilty of doing when reading similar‑style
books), I encourage you to resist the urge. Taking the time to build
foundational skills will pay dividends in the long run. Your future self
will thank you for this investment of time and effort.
3. Embrace experimentation: While this book offers structured guid‑
ance, it’s not meant to be a rigid rulebook. The examples and sug‑
gested strategies are starting points. As you grow more comfortable,
try rephrasing the provided AI commands, test different AI tools, and
venture “off script”. Those moments of experimentation can lead to
exciting discoveries and deepen your understanding of AI’s potential.
4. Expand your AI vocabulary: While I’ve written this book assuming no
prior knowledge of AI, you’ll encounter some specialised terms along
the way. Don’t let this jargon intimidate you. Whenever you come

4
 The gift of AI

across an unfamiliar AI term, flip to the Glossary section for a clear,

concise explanation.
5. Stay up‑to‑date: The field of AI is evolving at an intimidating pace,
with new tools and applications emerging almost daily. At the time
of writing, there were thousands of AI tools available.3 I’ve created a
companion resource, housed on my blog, where you’ll find regularly
updated resources, recommendations, and guidance.
6. Take it step by step: Build your AI literacy skills gradually, focusing on
practical applications for your academic work. Let your immediate
tasks and goals guide your learning, rather than getting swept up in
all the new AI advancement – which can be exhausting and quickly
lead to a sense of overwhelm.
7. Cultivate curiosity: Approach this journey with an open and curious
mind. The most rewarding discoveries often come when we’re willing
to explore, question, and play. Don’t be afraid to ask “What if?” or
“How else could this work?” Your curiosity will be your greatest asset
in mastering AI.
8. Reflect and apply: As you progress through the book, take time to
reflect on what you’re learning. How might these AI tools change
your research process? Your writing approach? Your data analysis?
Regularly consider how you can apply your new knowledge to your
specific academic context.
9. Connect and share: Learning is often more fun and effective when it’s
collaborative. Consider forming a study group with colleagues to dis‑
cuss the concepts in this book. Share your experiences, challenges,
and successes in applying AI to your academic work.
10. Prioritise responsible use: As you explore the capabilities of AI,
always keep responsible and ethical considerations at the forefront
of your mind. I’ll discuss these throughout the book, but it’s crucial
to continually reflect on the implications of AI and in ways that align
with your university policies and procedures.

Lastly, don’t forget to pause throughout your journey to congratulate

yourself.
As I write this, the majority of academics have not consciously
embedded AI in their academic workflows and are being outpaced in
AI adoption by their students.4 By reading this book, you can consider

5
AI-Powered Scholar

yourself among the early adopters of AI, and well on your way to develop‑
ing the AI literacy skills that, I believe, will increasingly be required within
academic employment.

Structure of the book

This book offers a series of chapters designed to guide you on a learning
journey towards AI literacy in academic contexts. We begin this journey
by building foundational skills, assuming no prior knowledge of AI, with
which to help you effectively communicate with AI tools. Along the way,
you’ll gain insights into the historical trajectory of AI, understanding how
we arrived at the current state of technology. You’ll learn how to transform
AI into a valuable research partner and master the nuances of crafting
textual commands that align AI outputs with your research goals. Impor‑
tantly, we’ll also explore the ethical considerations and responsible use of
AI, empowering you to mindfully integrate these tools into your academic
workflows. Once you’ve established a solid foundation, later chapters
invite you to put your skills into practice. Beyond these use cases, you’re
invited to explore additional applications of AI through the accompanying
online companion resource.
Here’s what you can expect from each chapter:

Chapter 1: Unwrapping the gift of AI

Your journey began in the chapter where you find yourself now.
Here, the stage was set for becoming an AI‑powered scholar. Before
you reach the end, you’ll additionally explore some ideas relating to
transparency of AI use and the need for disclosure statements. Then,
it’s off to Chapter 2 to begin building your AI toolkit.

Chapter 2: The AI revolution in context

We’ll begin by situating our journey within the broader histori‑
cal context of AI advancement. This chapter provides a concise yet
comprehensive overview of AI’s evolution, helping you understand
how we arrived at where we find ourselves today. You’ll gain insights
into the key milestones and paradigm shifts that have shaped AI’s
trajectory, setting the stage for its current and future applications in
academia.

6
 The gift of AI

Chapter 3: Communicating effectively with AI

Communication is key in any partnership, and your relation‑
ship with AI tools is no exception. This chapter unpacks the nuances
of interacting with large language models (LLMs). Through practical
examples and exercises, you’ll begin your journey towards learning
how to craft clear, precise instructions to achieve optimal results.
We’ll explore the importance of context, specificity, and natural lan‑
guage in AI interactions, helping you develop the linguistic skills nec‑
essary for effective collaboration with AI research assistants.

Chapter 4: Designing your AI research assistant

Imagine having a research assistant tailored precisely to your
needs. This chapter shows you how to make that a reality by assign‑
ing specific roles to AI systems. You’ll discover how to shape AI into
your custom research partner, by defining its expertise, background,
and personality traits. This approach will allow you to leverage AI’s
versatility across various research tasks, from literature reviews to
editing your academic writing and beyond.

Chapter 5: A brief history of conversational AI

Building on the historical context of AI development, this chap‑
ter narrows in on the evolution of “conversational AI”. You’ll trace the
development from early rule‑based systems to today’s sophisticated
AI chatbots. This journey through time will help you understand the
capabilities and limitations of current‑day AI, setting realistic expec‑
tations for your interactions with AI tools.

Chapter 6: Prompting AI for effective communication

At the heart of effective AI interaction lies the art of prompt
engineering. This crucial chapter provides a deep dive into writing
text‑based instructions that guide AI tools towards generating rele‑
vant, high‑quality content aligned with your research goals. You’ll
learn strategies for structuring prompts, refining outputs, and iterating
for optimal results. By mastering these skills, you’ll be able to achieve
maximum value from your AI collaborations.

Chapter 7: Towards responsible use of AI tools

In Chapter 7, attention turns to addressing key risks associated
with using AI tools in academic research. It offers several strategies

7
AI-Powered Scholar

to help you mitigate these risks and become a more informed user
of AI technologies. By exploring these approaches, you’ll be better
equipped to navigate potential pitfalls and make informed deci‑
sions about how to conduct AI‑assisted research responsibly and
effectively.

Chapter 8: Ideating with AI chatbots

Learn how to engage in interactive brainstorming sessions
with AI systems to uncover novel connections and generate inno‑
vative ideas. This chapter provides guiding principles for productive
human‑AI collaboration and outlines a practical process for AI‑aug‑
mented ideation, illustrated through a case study. You’ll be intro‑
duced to various creative thinking techniques that can be employed
in conjunction with AI chatbots to stimulate divergent thinking and
challenge assumptions, while maintaining human agency and criti‑
cal thinking throughout the process.

Chapter 9: AI‑powered literature search

This chapter looks at the transformative potential of AI in aca‑
demic literature search, with a particular focus on semantic search
capabilities – a method that aims to understand the meaning and
context of your search queries, rather than matching keywords.
You’ll learn how AI is changing the way scholars discover relevant
literature, moving beyond traditional keyword‑based methods.
Practical strategies for effectively leveraging these tools in your
research are discussed, along with potential pitfalls and ethical con‑
siderations. By the end of this chapter, you’ll be equipped to inte‑
grate AI‑powered literature search into your academic workflows,
enhancing your ability to navigate the ever‑expanding sea of schol‑
arly publications.

Chapter 10: Writing & editing with AI

This chapter offers practical strategies for leveraging AI in aca‑
demic writing, addressing the common challenge of writer’s block and
the intimidating blank page. It introduces “blah writing”, a technique
for using AI chatbots to transform stream‑of‑consciousness thoughts
into coherent narratives. This chapter then explores AI‑assisted meth‑
ods for generating document outlines, expanding text for greater
depth, and contracting text for concision. It also examines the role

8
 The gift of AI

of AI in editing, providing practical guidance on how to transform AI

chatbots into 24/7 editing assistants.

Chapter 11: Bonus chapter

Here, you’ll find information for unlocking the accompanying
online resource and access to a bonus chapter. I’d be spoiling the
surprise if I told you what the chapter was about ;)

Glossary
Along your journey, you might encounter unfamiliar terms. If so, con‑
sult the included glossary. It aims to clarify frequently encountered AI
terminology and is intended as a practical reference, offering clear and
concise definitions, rather than being an exhaustive list. Think of it as your
friendly quick‑reference guide, providing accessible explanations to help
in navigating the integration of AI into academic workflows.
By the time you reach the end, you’ll not only understand how to
effectively guide AI tools to help you with your research tasks, but you’ll
also be able to critically evaluate AI’s strengths and limitations and under‑
stand the historical trajectory of AI development.

Online companion resource

This book comes with a dynamic online companion resource. Gain access
by visiting www.broneager.com/aischolar and discover:

• A curated list of AI tools: a comprehensive, regularly updated list of

AI tools relevant to academic work.
• Best practices and tips: insights on how to integrate the latest AI
advancements into your academic workflow.
• Case studies: real‑world examples of how to leverage AI in innovative
ways, updated as new applications emerge.

The decision to house specific AI tool recommendations on a compan‑

ion website was made in recognition that AI is advancing at an unprec‑
edented pace. Thus, while this book provides you with foundational skills
and enduring principles for working with AI, the online resource ensures
that you have access to current, practical information.

9
AI-Powered Scholar

As we navigate this AI‑learning journey together, I recognise that

c­ollective wisdom often surpasses individual knowledge. That’s why I
warmly welcome your contributions. If you discover valuable resources,
develop innovative AI applications for academic work, or have insights
that could benefit fellow scholars, I encourage you to reach out. Your
experiences and discoveries could be featured on the companion site,
helping to enrich our shared learning experience.

Disclosure
Given that this book is about AI, you’re probably wondering if AI was used
in its creation. The short answer is yes. Would you have believed me if I’d
said ‘no’?
Over the months it took to write – underpinned by years spent on
what some might call an obsessive learning journey about AI – I always
had an AI tool by my side. These tools helped to translate my thoughts into
draft chapter outlines, rephrase my writing into clearer prose, and assisted
in locating resources to continue building my knowledge base. As such,
AI touched every part of this book, embedded in the process as a way of
extending my human capabilities, in line with the Extended Mind Thesis
proposed by Clark and Chalmers.5
This approach of integrating AI into my workflow represents more
than just using a tool; it’s a manifestation of cognitive extension, where AI
becomes an integral part of my thought process and creative output. By
embracing AI in this way, I’ve found that it enhances rather than replaces
my own cognitive abilities, allowing me to explore ideas more deeply
and express them more clearly than I might have done otherwise. This
co‑generative relationship with AI has not only shaped this book but has
also transformed my approach to academic work more broadly.
AI can be used to write books. However, the quality of AI‑authored
books remains questionable.
In early 2023, I experimented with using AI to write a book con‑
taining 500+ AI prompts across various academic use cases.6 I urge you
not to buy it. It was created in a weekend, and you’ll be able to gener‑
ate everything inside its cover with the skills you’ll soon acquire. That
self‑published book project, like my previous forays into drop‑shipping
and other entrepreneurial side hustles, forms part of my commitment to

10
 The gift of AI

learning through hands‑on experience. As an educator teaching entrepre‑

neurship, I believe in staying current by actively engaging with emerging
technologies and business models. These practical experiments, while not
meant to be serious ventures, allow me to translate theoretical under‑
standing into real‑world knowledge, ultimately making me a more effec‑
tive teacher.
The book you’re reading now is a different beast entirely.
Even if I had wanted to use AI to author this entire book – to a stan‑
dard that I’d be confident attaching my academic reputation to – the tech‑
nology simply isn’t up to the task. Thus, while AI has been an invaluable
tool that has sat alongside me during the writing process, the core con‑
tent, ideas, personal insights, and long‑form narrative structure are the
result of my “human” thoughts and experiences.
The long answer to the question of whether AI was used to write this
book requires further examination, as does the need for disclosure.
Throughout history, scholars have consistently adopted new tech‑
nologies to enhance their work, often without explicit acknowledgement
in their final products. From the chisel and stone tablet to the quill and
parchment, from the printing press to the typewriter, and from personal
computers to the internet, each technological advancement has shaped
the way academic work is created and disseminated. Yet, we don’t tend
to see ancient texts crediting the chisel, medieval manuscripts acknowl‑
edging the quill, or academic journal articles attributing their clarity to
word processing software and spell‑checkers. This book continues in that
tradition, leveraging current tools and technologies to facilitate research,
writing, and editing processes.
Disclosing the use of now‑commonplace technologies like com‑
puters or online databases in academic work today would seem rather
ludicrous. Similarly, I believe that the explicit acknowledgement of AI
usage in tomorrow’s version of academia might come to be seen as
equally unnecessary. As we navigate this transitional period, transpar‑
ency about AI’s role is largely being insisted upon, even as many of us
recognise it as simply the latest in a long line of technological aids to
scholarship.
Whether or not you choose to disclose the application of your AI
skills, I urge you to remember: AI is a powerful tool, but it’s your creativ‑
ity, critical thinking, and domain knowledge that will drive you towards
making meaningful academic contributions.

11
AI-Powered Scholar

Notes
1. See: Alshater, M. M. (2022, December 26). Exploring the role of arti‑
ficial intelligence in enhancing academic performance: A case study
of ChatGPT. Social Science Research Network. 1–22. https://doi.
org/10.2139/ssrn.4312358 or http://dx.doi.org/10.2139/ssrn.4312358;
and Chubb, J., Cowling, P. I., & Reed, D. (2021). Speeding up to keep
up: Exploring the use of AI in the research process. AI & Society, 37(4),
1439–1457. https://doi.org/10.1007/s00146‑021‑01259‑0
2. If you’re enthusiastic to supplement your reading with visual ­content,
I’ve compiled a list of recommended YouTubers who create high‑­
quality AI content for academics. You can find this list, along with
other helpful resources, at www.broneager.com/aischolar
3. FutureTools.io is a comprehensive database and directory of AI tools
and services. It collects and organises a wide range of AI tools to help
users find the exact AI solution for their specific needs. The database
covers various categories, including content creation, data analysis,
and automation.
4. A study by Coffrey (2023) reveals university students are adopting AI
tools at a faster rate than faculty members; Coffey, L. (2023, October 31).
Most students outrunning faculty in AI use, study finds. Inside Higher
Ed | Higher Education News, Events and Jobs. https://www.insidehigh‑
ered.com/news/tech‑innovation/artificial‑intelligence/2023/10/31/
most‑students‑outrunning‑faculty‑ai‑use
5. Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1),
7–19. http://www.jstor.org/stable/3328150
6. Eager, B. (2023). Academic Writing AI Prompts Phrasebook.

12
 2 The AI revolution
in context

Introduction
This chapter provides an overview of the evolution of AI, culminating in
the capabilities of modern‑day AI chatbots.1 We’ll trace the trajectory of
AI from early optimism around replicating human cognition in the 1950s
and 1960s, through a period of reduced funding and progress in the
1970s, leading to a resurgence in the 1990s enabled by increased compu‑
tational power and advances in machine learning. We’ll then fast forward
to today, where large language models (LLMs) can generate remarkably
human‑like text, disrupting many sectors including higher education. This
chapter explores the potential impacts of LLMs on academic work, exam‑
ining whether AI could write a thesis or paper from start to finish. We’ll
see there are good reasons to take a judicious approach when incorpo‑
rating AI writing tools into academic practice. This chapter concludes by
introducing a framework for strategically developing skills to collaborate
effectively with AI.

The many facets of AI

You’ve likely encountered the letters ‘A’ and ‘I’ in various contexts, but what
does this pairing truly signify in the context of this book? At its core, AI is a
field dedicated to enabling machines to perform tasks typically requiring

DOI: 10.4324/9781032665276-213
AI-Powered Scholar

human cognition and intelligence, such as learning, ­problem‑solving, and

language comprehension.
The AI landscape encompasses numerous specialised subfields. These
include machine learning (algorithms that improve through data analysis
without explicit programming for every scenario), computer vision (inter‑
preting visual inputs), natural language processing (understanding and
generating human language), robotics (creating autonomous machines for
human‑like tasks), and expert systems (emulating human decision‑making
through specialised knowledge). The field is vast and diverse.
This book primarily focuses on Generative AI (i.e. a type of AI that
can create new content like text, images, or music based on patterns it has
learned from existing data). Generative AI’s sophisticated algorithms rep‑
resent AI’s cutting edge in mastering “natural language” – the messy, con‑
text‑rich way humans communicate, rather than rigid computer syntax.
Think: “I’m dying for a coffee” versus “Human. Coffee. Need = Urgent”.
With a strong coffee by my side, I want to take you on a brief tour
of AI’s evolution. From its modest beginnings to today’s impressive chat‑
bots. We’ll explore how these advanced models resulted from decades
of innovation, and gained familiarity with major milestones and ongoing
challenges, offering insights into AI’s current capabilities and, importantly,
some of its limitations. By understanding this trajectory, you’ll be better
positioned to assess AI’s present state and its potential future impact on
academia and beyond.

Intelligent (?) machines

Imagine a world where machines think, reason, and work alongside

humans. This vision has fascinated dreamers and innovators for centuries.
While 18th‑century visionaries sketched designs for mechanical beings,2
it wasn’t until the mid‑20th century that this aspiration inched closer to
reality and began to take tangible shape.
The year 1950 marked a pivotal moment in this journey. Alan Turing,
a brilliant mathematician, published a ground‑breaking paper that would
forever change the landscape of computing. In it, he proposed what we
now know as the Turing test3 – a method to evaluate machine intelli‑
gence by determining if a computer could go undetected as a machine if
engaged in human‑style conversation.

14
 The AI revolution in context

Just five years later, the term “artificial intelligence” is thought to have
been coined by John McCarthy, a visionary computer scientist, who used
this now‑famous pairing of words in a workshop proposal that would
become a catalyst for AI research. This gathering of minds is considered a
key springboard for modern AI advancement.4
In the wake of these foundational developments, AI pioneers of the
1950s and 1960s set their sights on an ambitious goal: replicating human
intelligence through rule‑based programming,5 picturing a world where
machines could reason and make decisions based on an intricate web of
‘if‑then’ rules. By way of example, a simple rule could be: IF the weather
is sunny, THEN recommend going to the park. By chaining many such
rules together, early researchers aspired to create the equivalent of “think‑
ing machines”, mimicking human‑like decision‑making. However, reality
soon tempered these ambitions.
The modest processing power and hefty price tags of 1950s’ comput‑
ers posed considerable barriers.6 More crucially, it became apparent that
human cognition is far more complex than a series of logical rules. Our
ability to make nuanced judgements, recognise fuzzy patterns, and learn
interactively defied simple algorithmic representation.
Additionally, common sense proved an elusive quality7 – so innate
to human understanding, it proved frustratingly difficult to encode in a
machine. Despite initial enthusiasm, the ability of AI researchers to repli‑
cate the expansive flexibility of the human mind turned out to be far more
challenging than originally thought.8
As a result, funding and enthusiasm for AI began to wane, marking
the start of what’s known as an ‘AI winter’ period, referring to reduced
progress in AI research caused by setbacks in overcoming technical barri‑
ers and losses of confidence in the viability of achieving strong AI.9 Dur‑
ing this time, research continued, but grand hopes of replicating general
human intelligence began to wane.
The 1990s ushered in a renaissance for AI, driven, in part, by expo‑
nential leaps in computing power. This new era of technological affor‑
dances empowered researchers to harness advanced statistical techniques
and pioneer machine learning algorithms, capable of analysing vast,
ever‑expanding datasets. Machine learning refers to AI systems that can
learn and improve from experience without being explicitly programmed
for every scenario, and it marked a paradigm shift. AI systems could
now learn and improve from experience, shedding the need for explicit

15
AI-Powered Scholar

programming of every possible scenario. As an example, imagine an AI

sifting through millions of cat and dog photos, gradually discerning the
subtle visual cues that distinguish whiskers from snouts, and pointed ears
from floppy ones. This process of pattern recognition allows the system to
accurately categorise new, unseen images – a task once thought to be the
exclusive domain of human perception. Such capabilities opened doors
to applications light years beyond simple image classification, setting the
stage for AI to tackle increasingly complex real‑world problems.
In addition to progress occurring in the domain of machine learn‑
ing, neural networks made a comeback in the 1990s. Neural networks
are AI systems loosely modelled after the neuron connections in the
human brain.10 Early neural networks were limited by insufficient com‑
puting resources. But with 1990s’ hardware improvements, they could be
designed larger and deeper, achieving far better results. These advances
enabled notable progress in AI capabilities like computer vision – analys‑
ing visual data like images and video; speech recognition – transcribing
spoken audio; and machine translation – automated translation between
languages. But, compared to what’s possible today, these advancements
could be considered as revolutionary as my Panasonic dual‑tape stereo
was, if compared to the capabilities of the iTunes store.
While AI was going through “boom and bust” cycles, steady progress
continued.
In the 2010s, we saw the rise of deep learning, which refers to neural
networks with multiple layers that can learn highly complex relationships
in data11 – an approach that drew inspiration from neuroscience research.
Although deep learning pre‑dates the 2010s, around this time the field
underwent significant advancements in its capabilities.12 In contrast to
earlier neural networks which were relatively shallow with few layers,
later iterations of deep learning leveraged many stacked layers of com‑
putations. This layered architecture enabled major performance leaps,
including the now‑famous example of when an AI system called AlphaGo
defeated the (human) world Go champion.13 Go is an ancient Chinese
strategy game that, despite simple rules, has a near infinite number of
potential game configurations. To win the game, the human (or machine)
requires strategic intuition and creative problem‑solving. While many
backed their money on human triumph, AlphaGo won the match14 4–1.
AlphaGo achieved its breakthrough win by using a novel combination of
algorithms and deep neural networks trained on thousands of human Go

16
 The AI revolution in context

games (i.e. training data). This allowed it to look ahead at possible moves
and countermoves, and intuitively evaluate board positions.
The landmark AlphaGo win signalled a major leap in AI advance‑
ment because AI could now surpass humans in highly strategic domains
requiring intuition, not just raw computing power. Additionally, because
Go was seen as a game requiring abstract reasoning skills, the win fur‑
ther signalled the potential of deep learning to deliver human‑like cog‑
nition given enough training data. This level of skill was once thought
unattainable for AI and led to enhanced enthusiasm for further advancing
AI capabilities. Despite all the progress, AI capabilities remained far from
matching human‑level flexibility across all cognitive domains.
Looking at the historical trajectory reminds us how far AI capabili‑
ties have come. But it also highlights that realising the full aspirations of
pioneers like Alan Turing and John McCarthy remains a quest requiring
ongoing breakthroughs.
For many in academia, AI came onto the radar in late 2022, when
OpenAI,15 at the time a relatively unknown company, achieved notoriety
in mainstream media after launching a product underpinned by a technol‑
ogy capable of disrupting multiple industry sectors.16
Even if you haven’t used it, you’re likely familiar with OpenAI’s prod‑
uct, ChatGPT,17 a conversation‑style LLM that produces text (and more!)
that ostensibly appears human‑generated and does so at speeds that far
exceed the capabilities of even the most talented touch typists.18 Unprec‑
edented public uptake of the technology ensued, making OpenAI’s Chat‑
GPT the fastest‑growing application in history.19 No less among student
populations, who swiftly embraced the technology to complete their
assignments.20 It was primarily for this reason that from November 2022
onwards, ChatGPT’s text‑generation abilities sent the higher education
sector into a period of paradigm‑shifting revolution.
Or so I thought.
From my vantage point, I witnessed a surprising lag in response. Months
passed before many academics began discussing AI, and when they did,
many dismissed it as another passing trend. Given AI’s growing influence,21
I found the hesitation towards embracing an understanding of how it could
be leveraged in academia rather strange, if not alarming – as I was increas‑
ingly concerned for the relevance of the higher education sector.
Tech enthusiasts and late‑stage early adopters were exploring, and
profiting from, AI’s capabilities. Meanwhile, many in higher education

17
AI-Powered Scholar

focused their conversations on how students could potentially use this

technology to bypass the integrity of traditional forms of academic assess‑
ment.22 And while plagiarism conversations seem to largely sustain, in
some circles these conversations soon expanded into discussions of how
AI might impact the higher education sector more broadly.23 As the capa‑
bilities of AI tools expanded, scholars increasingly began to realise that AI
might be able to assist them in achieving many of their academic tasks.
Thus began a movement among some academics towards embrac‑
ing AI technologies across multiple touchpoints of their academic work,
including using AI chatbots to assist in writing academic papers. Those
who openly admitted to doing so included several scholars who went as
far as crediting ChatGPT as a co‑author on their publications.24 Although
this attribution was quickly challenged despite some arguing for it meeting
authorship criteria25 – ChatGPT was deemed to lack the agency required
for human authorship.26 Regardless, these events brought scholarly use
of AI into the spotlight, sparking curiosity, experimentation, and dialogue
about AI’s capabilities, not just for students, but for the academics respon‑
sible for governing future academic practice.
In my journey with AI, I’ve often found myself amazed at what these
tools can accomplish with just a few clicks. My colleagues have shared
similar experiences of awe. Yet, these moments of wonder tended to be
tempered with feelings of uncertainty about the future of the higher edu‑
cation sector. These moments led me to ask, “Will AI continue to advance
to the point where it replaces human skills and academic roles entirely?”
Reassuringly (although for how long remains anyone’s guess),
advancements in AI have not yet achieved the expansive general capa‑
bilities that were initially envisioned.27 This is due to a range of factors,
including the slow diffusion of AI’s most impressive capabilities, the need
for complementary innovations, and the challenges in integrating a broad
range of cognitive capabilities into a coherent architecture.28
AI today excels as a specialist, demonstrating superhuman profi‑
ciency in narrow domains it has trained extensively in. For example, an
AI can defeat the world’s greatest chess grandmaster, translate languages
in milliseconds, and create stunning works of digital art. But it cannot yet
then shift to displaying the situational adaptability, well‑rounded com‑
mon sense, or interpersonal emotional intelligence that allows humans to
thrive in the messy realities of everyday life. Current AI reasoning relies on
patterns extracted from datasets, lacking the grounded understanding that

18
 The AI revolution in context

comes from real‑world experiences. This limitation can hinder AI’s ability
to respond flexibly to novel situations, and the general learning capabili‑
ties that humans develop during childhood, through interaction with the
physical and social world, remain challenging to replicate in machines.
Despite these limitations, AI can be remarkably capable, or “scary
smart” as AI expert Mo Gawdat describes in his book on AI’s future
impact.29 The rapid pace of AI advancement makes it difficult to predict
how long the current limitations will persist. For now, human cognition
retains an edge in creatively connecting contextual information across
domains, a task that narrow AI struggles with. Thus, the gap between
artificial general intelligence and today’s specialised AI systems remains
significant.
Rather than viewing AI as a threat to human roles, I tend to see it
as an opportunity to enhance our unique strengths – as an extension of
our human capabilities. This idea comes from the Extended Mind The‑
sis, introduced by philosophers Andy Clark and David Chalmers, which
posits that cognitive functions can extend beyond the biological brain to
incorporate external tools and resources. Essentially, AI could be consid‑
ered an external object that actively aids our thinking and guides our cog‑
nition, thus extending our cognitive architecture beyond the traditional
boundaries of our minds.30

AI‑enhanced academic roles

While artificial intelligence has made great strides, it cannot yet repli‑
cate the full range of scholarly tasks and social nuances involved in the
multi‑faceted nature of academic work.
Trust me, I’ve tried!
Within the bounds of its current abilities, AI remains sitting in the
passenger seat,31 even if it might be autonomously driving the car.
The array of responsibilities spanning teaching, research, collabora‑
tion, and university service appear secure from wholesale automation.
An academic’s role extends beyond processing information or generating
publications. Our ability for original theorising, contextual synthesis of
concepts, the ability to design rigorous studies aligned to ethical review
board requirements, critically evaluating evidence, and communicating
complex ideas (which is a long, but far from exhaustive list of hats we wear)

19
AI-Powered Scholar

all draw upon accumulated human training in scholarly methods and also
reflect our disciplinary quirks.
Moreover, we mustn’t underestimate the value of human elements
like mentorship, debate, and interpersonal rapport which underpins
scholarly work. Fingers crossed, students will continue to seek out the
wisdom of their supervisors and extended academic community rather
than solely relying on chatbots for advice. Thus, while AI promises to
enhance certain academic tasks, the essence of scholarly work remains
profoundly human.
While AI may not be capable of replicating human academics, it is
reasonable to expect that academics who fail to keep pace with shifting
work practices will fall behind.
Take, for example, two fellowship candidates where one is highly AI
literate while the other dismisses AI tools as trendy gadgets. Putting aside
other entries in their resumés, it seems clear which would be better placed
to contribute to an institution that’s operating in a period where AI is
increasingly entering all aspects of university operations. Any scholar inte‑
grating AI for efficiencies like automated admin, data analysis, or writing
assistance will likely be at a competitive advantage over less tech‑savvy
peers because AI affords knowledge workers the ability to complete tasks
at greater speed and efficiency.32 Though AI cannot wholly substitute aca‑
demics, scholars who ignore developing capabilities may find themselves
publishing less, winning fewer grants, and becoming outmoded in com‑
parison to their AI‑literate peers.
Amidst this environment, I encourage you to judiciously welcome
AI as a collaborator while safeguarding the aspects of your academic role
that are most important to you. By cultivating AI skills, academics ensure
they write their future rather than becoming relics replaced by thinking
machines or displaced by AI‑augmented peers.33

Can AI write my thesis or academic paper?

Before providing some examples of how AI can potentially assist you in
your academic work, let’s address the elephant in the room: “Can artificial
intelligence write your PhD thesis or academic paper from start to finish?”
In short, “Yes it can!”

20
 The AI revolution in context

AI systems are currently capable of generating long‑form text on

demand on virtually any topic.34 Putting aside the question of whether the
AI‑generated text would be any good, there are many reasons why tak‑
ing a hands‑off approach and letting AI entirely write such important and
complex documents is ill‑advised, not to mention highly problematic.35 To
explore this issue, let’s examine some motivations for why you might want
to outsource your writing to AI, from the perspectives of a PhD candidate
(Jane) and a tenured academic (Louise).

Jane’s PhD candidature journey

As a fledgling PhD candidate, Jane faces immense pressure to produce

written work. The lengthy thesis can feel like an insurmountable mountain
to climb, and Jane spends long hours reading literature and taking notes
but struggles to synthesise concepts and articulate her ideas. She watches
as peers effortlessly produce page after page while she agonises over each
paragraph.
The temptation arises to let an AI system quickly generate paragraphs
or even entire chapters that she can then polish up and submit.
She’s heard that’s possible.
Why not give it a go?
Well, aside from a host of ethical quandaries associated with one‑click
text generation,36 one of the key problems with Jane using AI merely as an
automated writing tool is that doing so overlooks a core purpose of the
PhD journey: the apprenticeship of becoming an independent researcher.
The doctoral process is as much about developing critical research
skills as it is about producing the final dissertation, noting that the two
are very much an interwoven process. Bypassing the analytical thinking
needed to synthesise concepts, identify gaps, and then articulate ideas
through the writing process (i.e. outsourcing that work to AI) cheats Jane
out of an enriching scholarly experience.
Yes, the thesis document needs to be written, and many of the tasks
involved in its creation could be outsourced to AI to accelerate (or replace)
the process. However, the skills picked up along the learning journey
towards producing the thesis are what will sustain and advance Jane’s
career long term.

21
AI-Powered Scholar

If she gains an academic role post‑PhD, it’s the research and writing
skills honed during her PhD candidature that will enable her to continue
to grow, not the final thesis itself. In other words, while submitting the the‑
sis document is her ticket to entry into the academy, once inside, nearly
everyone holds that ticket. Jane’s competitive advantage will be charac‑
terised by the skills she’s acquired in the process. As such, rather than
viewing AI as an automated author, a more strategic approach would be
for Jane to use AI as a collaborative assistant; a tool with which to amplify
her abilities.

Louise’s tenured academic journey

Unlike Jane, Louise has already earned her PhD and achieved tenure as
a professor at a prestigious university. She has a solid grasp of her field
and extensive experience conducting original research and publishing
academic papers. However, Louise feels immense pressure to continu‑
ously publish to maintain her position and progress in her career. The
prospect of letting an AI system generate draft papers is enticing. Why
should she waste time on the tedious process of writing literature reviews,
or methods sections, when AI can quickly produce those components of
her work?
Louise may be well‑justified in rationalising that she already knows
how to conduct rigorous research, however, she should be wary of fully
outsourcing her writing to AI for several reasons. Firstly, writing remains a
critical part of Louise’s thought process and a way to clarify and deepen
her ideas. By skipping the writing, she loses opportunities for scholarly
insights. Secondly, academic writing requires making nuanced rhetorical
choices to effectively convey complex concepts and arguments to read‑
ers. Advanced knowledge and critical thinking are needed to organise
ideas logically, choose words judiciously, and adapt the tone of writing
for different audiences. As yet, AI is hard‑pressed to replicate Louise in her
pursuit of those tasks. Thirdly, transparency and ethics demand that Louise
take responsibility for the final published work bearing her name. While
scholars may opt to turn to AI as a black box for churning out papers,
Louise holds the scholarly process in high regard. Thus, while AI offers

22
 The AI revolution in context

the potential for productivity gains, Louise is also mindful that she doesn’t
wish to outsource the core of her academic work.
Rather than using AI to generate fully written papers, a better
approach may be for Louise to use AI as an assistive tool, used to col‑
laborate on writing tasks, leaving her in control over the process, and
removing any doubt that she is the unquestionable author of the papers
she puts into the world.
In examining the perspectives of both Jane and Louise, we see good
reasons to pull back from the idea of fully automating academic writing
with AI tools. Beyond the ethical responsibility needed to produce sub‑
stantive scholarly work, both Jane and Louise would cheat themselves of
enriching scholarly experiences by treating AI as a shortcut to publication.
However, the above examples also illustrate how AI could be strategi‑
cally incorporated as a tool to enhance, not replace, human skills and
writing. To explore how this might be achieved, let’s leave current‑day AI
for a moment to travel back to 1984, where guidance can be found for
how Jane and Louise might effectively build their AI skills and become
AI‑powered scholars.

A framework for developing AI literacy

One method for acquiring the skills needed to use AI effectively can be
found in the “Miyagi Framework”, comprising the following tenets:

• break down complex skills into fundamental tasks,

• develop muscle memory through deliberate practice, and
• apply and combine the basics to execute advanced techniques.

If you haven’t come across the Miyagi Framework before in academic

research, that’s probably because it’s more commonly known as a popular
culture touchstone. Mr Miyagi, from whom the framework takes its name,
is a protagonist in the 1980s’ cult classic, The Karate Kid, a film about
Miyagi and his young pupil, Daniel.
When Daniel faces challenges in his life, the sage Miyagi takes him
under his wing to school him in the art of Karate.

23
AI-Powered Scholar

Eager to commence his training, Daniel turns up to his first class, full
of enthusiasm, only to find that Miyagi wishes Daniel to wash his fleet of
vintage cars. To complete the chore, Daniel is instructed to use specific
hand movements, delivered by the famous film quote “Wax on, wax off”.
Although frustrated to learn that his Karate aspirations are temporarily on
hold, Daniel obliges Miyagi and engages in the task with the expectation
that doing a good job should surely lead to the commencement of his
official training.
When the next day arrives, Daniel’s hopes of learning to become a
Karate master are put on hold again as Miyagi assigns another chore to his
pupil. Handing Daniel two sanding discs, Miyagi provides clear instruc‑
tions for how the sanding should be conducted, “Right circle, left circle”.
To which Daniel replies, “Wouldn’t it be easier going back and forth?”
demonstrating his lack of awareness of Miyagi’s stealth training methods.
Seemingly conned out of another day of Karate training, Daniel
completes the task, leaves disgruntled, but returns with renewed hope
that finally his training can begin. Yet, despite his hopes, disappointment
ensues as Miyagi hands Daniel a paintbrush and tin and instructs him
to spend his next day of training painting fences. When Daniel finally
reaches the limits of his youthful patience, he erupts in protest, claiming
he hasn’t learned a thing.
If you’re familiar with the movie, you’ll also be familiar with Miya‑
gi’s formative pedagogical approach to fostering skill development. With
each chore, unbeknownst to Daniel, Miyagi was helping Daniel build the
muscle memory and reflexes required to tackle any Karate move thrown
his way.
It’s through the lens of the Miyagi Framework that I offer you a series
of skill‑development activities in this book. Each is designed to develop
your muscle memory and AI reflexes. However, unlike the naive Daniel
who laboured through his training without knowledge that each task was
building towards mastery, you’re invited to enter the journey fully aware
that you have some car waxing, deck sanding, and fence painting to do
before getting in the AI ring.
Before proceeding to the next chapter, take a moment to reflect on
the concepts we’ve explored. The questions offered in the “Reflection and
discussion” section below are designed to deepen your understanding
and encourage critical thinking about the role of AI in academia. Consider
discussing these with colleagues to gain diverse perspectives.

24
 The AI revolution in context

Reflection and discussion

The following questions are presented for personal reflection and can
also be used to spark discussion of the topics covered in this chapter
within a classroom setting.

1. Discuss the ethical implications of using AI in academic writing.

How might this affect the integrity and originality of scholarly work?
2. Compare and contrast the potential benefits and risks of AI use for
early career researchers (like Jane) versus established academics
(like Louise).
3. How might the integration of AI tools in academia influence the
development of critical thinking and research skills in students
and early career researchers?
4. Discuss the concept of AI as an extension of human cognitive
capabilities (as per the Extended Mind Thesis). How might this
perspective change our approach to using AI in academic work?
5. What skills do you think will be most valuable for academics in an
AI‑augmented future? How can these skills be developed?

Activity
If you’re working through this book with friends or in a classroom setting,
I encourage you to engage with the activities provided at the end of each
chapter.

Activity: AI in academia role‑play

Objective: To explore the potential uses, benefits, and challenges of inte‑

grating AI into academic work from various perspectives.
Instructions:

1. Form small groups of 3–4 participants.

2. Assign roles within each group, for example:

25
AI-Powered Scholar

{{ PhD candidate or early‑career researcher

{{ Tenured professor
{{ University ethics committee member
{{ Academic journal editor
3. Scenario: Your university is developing guidelines for the use of AI
in academic writing and research. Each group needs to create a pro‑
posal for these guidelines.
4. Role‑play discussion (30 minutes):
{{ Each participant should advocate from their assigned perspective.
{{ Consider how AI might be integrated into different aspects of
academic work (e.g. literature review, data analysis, writing, peer
review).
{{ Discuss potential benefits and risks.
{{ Address how academic integrity could be maintained while ben‑
efiting from AI tools.
5. Develop guidelines (15 minutes):
{{ Based on your discussion, create a short list of 5–7 guidelines for
AI use in academic work.
6. Share and compare (15 minutes):
{{ Each group presents their guidelines to the larger class.
{{ Discuss similarities and differences between group proposals.
7. Reflection (10 minutes):
{{ Individually, write a brief reflection on how this activity has influ‑
enced your perspective on AI in academia.
{{ Consider how you might apply these insights, or adopt the devel‑
oped guidelines, in your academic research or teaching roles.

Notes
1. A chatbot is an AI program designed to simulate conversation, pri‑
marily through text but also potentially through voice, images, or
other modalities. In this chapter, we focus on text‑based LLMs.
2. Riskin, J. (2003). The defecating duck, or, the ambiguous ori‑
gins of artificial life. Critical Inquiry, 29(4), 599–633. https://doi.
org/10.1086/377722

26
 The AI revolution in context

3. Turing, A. (1950). I.—Computing machinery and intelligence. Mind,

LIX(236), 433–460. https://doi.org/10.1093/mind/lix.236.433
4. McCarthy, J. M., Minsky, M. L., Rochester, N., & Shannon, C. E.
(1955). A Proposal for the Dartmouth Summer Research Project on
Artificial Intelligence. https://www‑formal.stanford.edu/jmc/history/
dartmouth/dartmouth.html
5. Copeland, B. J. (2023). Early AI in Britain: Turing et al. IEEE Annals
of the History of Computing, 45(3), 19–31. https://doi.org/10.1109/
mahc.2023.3300660
6. Anyoha, R. (2017, August 28). The history of artificial intelligence.
Harvard Graduate School of Arts and Sciences – Science in the News.
https://sitn.hms.harvard.edu/flash/2017/history‑artificial‑intelligence/
7. Crevier, D. (1993). AI: The tumultuous history of the search for
artificial intelligence. Basic Books. https://www.researchgate.net/
publication/233820788_AI_The_Tumultuous_History_of_the_
Search_for_Artificial_Intelligence
8. Dreyfus, H. L. (1992). What Computers Still Can’t Do: A Critique of
Artificial Reason. MIT press.
9. Mitchell, M. (2021, April 26). Why AI Is Harder Than We Think. arXiv.
org. https://arxiv.org/abs/2104.12871
10. Müller, B., Reinhardt, J., & Strickland, M. T. (2012). Neural Networks:
An Introduction. Springer Science & Business Media.
11. LeCun, Y., Bengio, Y., & Hinton, G. E. (2015). Deep learning. Nature,
521(7553), 436–444. https://doi.org/10.1038/nature14539
12. Alzubaidi, L., Zhang, J., Humaidi, A. J., Al‑Dujaili, A. Q., Duan, Y.,
Al‑Shamma, O., Santamaría, J., Fadhel, M. A., Al‑Amidie, M., & Far‑
han, L. (2021). Review of deep learning: Concepts, CNN architec‑
tures, challenges, applications, future directions. Journal of Big Data,
8(53), 1–74. https://doi.org/10.1186/s40537‑021‑00444‑8
13. Silver, D., Schrittwieser, J., Simonyan, K., Antonoglou, I., Huang, A.,
Guez, A., Hubert, T., Baker, L. R., Lai, M., Bolton, A., Chen, Y., Lillicrap,
T. P., Fan, H., Sifre, L., Van Den Driessche, G., Graepel, T., & Hass‑
abis, D. (2017). Mastering the game of Go without human knowledge.
Nature, 550(7676), 354–359. https://doi.org/10.1038/nature24270
14. Borowiec, S. (2017, November 29). AlphaGo seals 4‑1 v­ ictory over
Go grandmaster Lee Sedol. The Guardian. https://www.theguardian.
com/technology/2016/mar/15/oogles‑alphago‑seals‑4‑1‑vic‑
tory‑over‑grandmaster‑lee‑sedol

27
AI-Powered Scholar

15. https://openai.com
16. Agrawal, A. (2022, December 14). ChatGPT and how AI disrupts
industries. Harvard Business Review. https://hbr.org/2022/12/chatgpt‑
and‑how‑ai‑disrupts‑industries
17. ChatGPT is an AI language model developed by OpenAI that can
engage in human‑like text conversations, answer questions, and assist
with various tasks across a wide range of topics, available at https://
openai.com/chatgpt/. While ChatGPT is used as the primary example
in this book due to its widespread recognition, it’s important to note that
there are numerous other chatbots available. These include Anthropic’s
Claude and Google’s Gemini, among others. Each of these AI models
has its own unique features and capabilities. I encourage trialling mul‑
tiple chatbots to find the one you like best, which suits your needs.
18. Chen, T. J. (2023). ChatGPT and other artificial intelligence applications
speed up scientific writing. Journal of the Chinese Medical Association,
86(4), 351–353. https://doi.org/10.1097/jcma.0000000000000900
19. Gordon, C. (2023, February 2). ChatGPT is the fastest growing app in
the history of web applications. Forbes. https://www.forbes.com/sites/
cindygordon/2023/02/02/chatgpt‑is‑the‑fastest‑growing‑ap‑in‑
the‑history‑of‑web‑applications/?sh=32a03ec7678c
20. Eke, D. (2023). ChatGPT and the rise of generative AI: Threat to aca‑
demic integrity? Journal of Responsible Technology, 13, 100060.
https://doi.org/10.1016/j.jrt.2023.100060; Stokel‑Walker, C. (2022).
AI bot ChatGPT writes smart essays — should professors worry?
Nature. https://doi.org/10.1038/d41586‑022‑04397‑7
21 Walsh, T. (2023, November 30). A year of ChatGPT: 5 ways the AI
marvel has changed the world. The Conversation. https://theconver‑
sation.com/a‑year‑of‑chatgpt‑5‑ways‑the‑ai‑marvel‑has‑changed‑the‑
world‑218805
22. Perkins, M. (2023). Academic integrity considerations of AI large lan‑
guage models in the post‑pandemic era: ChatGPT and beyond. Jour‑
nal of University Teaching and Learning Practice, 20(2), 1–26. https://
doi.org/10.53761/1.20.02.07
23. Lund, B., Wang, T., Mannuru, N. R., Nie, B., Shimray, S. R., & Wang,
Z. (2023). ChatGPT and a new academic reality: Artificial intelli‑
gence‐written research papers and the ethics of the LLMs in scholarly
publishing. Journal of the Association for Information Science and
Technology, 74(5), 570–581. https://doi.org/10.1002/asi.24750

28
 The AI revolution in context

24. Stokel‑Walker, C. (2023). ChatGPT listed as author on research

papers: Many scientists disapprove. Nature, 613(7945), 620–621.
https://doi.org/10.1038/d41586‑023‑00107‑z
25. Polonsky, M. J., & Rotman, J. (2023). Should artificial intelligent (AI)
agents be your co‑author? Arguments in favour, informed by ChatGPT.
Social Science Research Network, 1–14. https://doi.org/10.2139/
ssrn.4349524
26. Thorp, H. H. (2023). ChatGPT is fun, but not an author. Science,
379(6630), 313. https://doi.org/10.1126/science.adg7879
27. Arel, I., & Livingston, S. (2009). Beyond the Turing test. IEEE Computer,
42(3), 90–91. https://doi.org/10.1109/mc.2009.67; Świechowski, M.
(2022). Deep learning and artificial general intelligence: Still a long
way to go. arXiv (Cornell University), 1–14. https://doi.org/10.48550/
arxiv.2203.14963
28. Brynjolfsson, E., Rock, D., & Syverson, C. (2017). Artificial Intelli‑
gence and the Modern Productivity Paradox: A Clash of Expectations
and Statistics. https://doi.org/10.3386/w24001
29. Gawdat, M. (2021). Scary Smart: The Future of Artificial Intelligence
and How You Can Save Our World. Pan Macmillan.
30. Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1),
7–19. http://www.jstor.org/stable/3328150
31. Chan, C. K. Y., & Tsi, L. H. Y. (2023). The AI revolution in education:
Will AI replace or assist teachers in higher education? arXiv (Cornell
University), 1–18. https://doi.org/10.48550/arxiv.2305.01185
32. Dell’Acqua, F., McFowland, E., Mollick, E., Lifshitz‑Assaf, H., Kellogg,
K. C., Rajendran, S., Krayer, L. J., Candelon, F., & Lakhani, K. R. (2023).
Navigating the jagged technological frontier: Field experimental
evidence of the effects of AI on knowledge worker productivity and qual‑
ity. Social Science Research Network, 1–58. https://doi.org/10.2139/
ssrn.457332; Noy, S., & Zhang, W. (2023). Experimental evidence on
the productivity effects of generative artificial intelligence. Social Sci‑
ence Research Network. https://doi.org/10.2139/ssrn.4375283
33. Howard, J. (2019). Artificial intelligence: Implications for the future
of work. American Journal of Industrial Medicine, 62(11), 917–926.
https://doi.org/10.1002/ajim.23037; Taylor, A., Nelson, J., O’Donnell,
S., Davies, E., & Hillary, J. (2022). The skills imperative 2035: What
does the literature tell us about essential skills most needed for work?
Working Paper 1. National Foundation for Educational Research.

29
AI-Powered Scholar

34. Grimaldi, G., & Ehrler, B. (2023). AI et al.: Machines are about
to change scientific publishing forever. ACS Energy Letters, 8(1), 878–
880. https://doi.org/10.1021/acsenergylett.2c02828; Kohda, Y., &
Javed, A. (2023). To what extent can AI simplify academic paper writ‑
ing? AHFE International, 1–10. https://doi.org/10.54941/ahfe1004194
35. Balel, Y. (2023). The role of artificial intelligence in academic paper
writing and its potential as a co‑author: Letter to the editor. European
Journal of Therapeutics, 29(4), 984–985. https://doi.org/10.58600/
eurjther1691; Lee, J. Y. (2023). Can an artificial intelligence chatbot
be the author of a scholarly article? Journal of Educational Evaluation
for Health Professions, 20, 6. https://doi.org/10.3352/jeehp.2023.20.6
36. Kooli, C. (2023). Chatbots in education and research: A critical exam‑
ination of ethical implications and solutions. Sustainability, 15(7),
5614. https://doi.org/10.3390/su15075614

30
 3 Communicating
effectively with AI

Introduction
Even among humans, miscommunication is a daily dance we all stum‑
ble through. A raised eyebrow, a subtle tone shift, or a cultural context
can completely change the meaning of our words. Now, strip away all
those non‑verbal cues, remove shared life experiences, and try to convey
your deepest academic insights using only your keyboard to guide your
conversation.
In this chapter, we’ll explore the fascinating – and sometimes frustrat‑
ing – world of human‑AI communication. We’ll explore the pitfalls that
can turn your carefully crafted instructions into digital lost‑in‑translation
moments. Through an interactive activity, you’ll experience firsthand how
easily messages can go awry and discover strategies to bridge the gap
between human intent and AI output. By the end of this chapter, you’ll be
better equipped with the skills to translate your academic goals into clear
instructions that even the most literal‑minded AI can understand.

Lost in translation
Let’s begin with an activity called “Lost in Translation”. It aims to demon‑
strate the challenges of conveying information accurately, even between
humans, and sets the stage for understanding the complexities of com‑
municating with AI.

DOI: 10.4324/9781032665276-331
AI-Powered Scholar

Here’s the premise: you’ll draw a picture and then attempt to describe
it verbally to a partner. Your partner’s task is to recreate your drawing based
solely on your spoken instructions. No peeking. No hand gestures. Just words!
Sounds straightforward, doesn’t it?
You might be thinking, “How hard can it be to describe a simple
drawing?”
Well, prepare to be surprised! This activity has a knack for exposing
the gaps between what we think we’re communicating and what others
understand.
To get started, you’ll need to find a willing partner. This could be a
fellow academic, a friend, a family member, or a housemate. The beauty
of this activity is that it works just as well with a colleague as it does with
your nine‑year‑old niece.
Once you’ve recruited your partner and you’re both ready to begin,
here’s a step‑by‑step guide to completing the activity:

1. Preparation: Each participant needs a blank sheet of paper and a pen

or pencil.
2. Initial drawing: Set a timer for around 60 seconds. During this time,
create a drawing on your paper. Be creative – use shapes, lines, or any
design elements you wish. Keep your drawing hidden from your partner.
3. Role assignment: Decide who will be the “describer” and who will
be the “drawer”. The describer’s task is to verbally guide the drawer
in recreating the original drawing without showing it.
4. Description phase: The describer provides verbal instructions to the
drawer. Focus on specific details such as shapes, sizes, positions, and
angles. Use clear, precise language. The drawer should listen silently,
only asking for clarification if absolutely necessary.
5. Drawing phase: As the describer speaks, the drawer attempts to recre‑
ate the original drawing based solely on the verbal instructions.
6. Comparison: Once complete, reveal both drawings and compare
them side by side, and identify the similarities and differences
between the original and the reproduction.
7. Analysis: Discuss the communication process. Identify what worked
well and what led to misunderstandings as you moved towards rep‑
licating the drawing. Consider how the instructions could have been
improved to achieve a more accurate reproduction.

32
 Communicating effectively with AI

8. Role reversal: Switch roles and repeat steps 4–7. The original drawer
now becomes the describer.
9. Second comparison: Compare the new set of drawings, noting any
improvements or persistent challenges in communication.
10. Reflection: Engage in a broader discussion about the experience. If in
a group setting, share insights and observations with the larger class.
Consider how this exercise relates to communication challenges in
other contexts, including with AI systems.

Now that you’ve completed the activity, let’s reflect on your experience:

1. Outcome analysis: How closely did the reproduced drawing match the
original? Were there significant differences or surprising similarities?
2. Communication challenges: Did you encounter any unexpected dif‑
ficulties in describing or interpreting the instructions? Were there
moments where you felt your partner misunderstood your directions?
3. Specificity struggles: How did you fare when trying to convey precise
details? Did you find yourself struggling to describe exact shapes,
angles, or spatial relationships? What strategies did you use to over‑
come these challenges?
4. Language limitations: Did you feel constrained by vocabulary or
struggle to find the right words? How did you adapt your language to
make your instructions clearer?
5. Detail dilemma: How did you balance providing enough detail for
accuracy without overwhelming your partner with information? Did
you find yourself over‑explaining or under‑explaining certain elements?
6. Perspective problems: Did you encounter any issues related to per‑
spective or point of view? How did you handle describing the overall
layout versus specific elements of the drawing?

As explored below, these communication hurdles parallel the difficulties

we face when interacting with AI systems.

Unpacking the activity

The “Lost in Translation” activity is a classic icebreaker that I use in my

AI training workshops. And, despite the varied cultural backgrounds of

33
AI-Powered Scholar

participants, from chemists in Vienna to accountants in Auckland, key

insights consistently emerge during the post‑activity reflection and discus‑
sion. These universal takeaways (introduced below) highlight fundamental
aspects of human communication, and, by extension, the challenges we
can face when interacting with AI systems.

Text‑based interactions omit verbal cues

A crucial insight from this activity is the significant role non‑verbal cues play
in our daily communications. Facial expressions, gestures, and other visual
signals often convey as much meaning as our words. However, when inter‑
acting with AI through text prompts, we lose these familiar cues entirely.
This loss of non‑verbal communication adds a unique challenge to
human‑AI interactions, with research showing that people tend to use
shorter, less nuanced instructions when communicating with computer
interfaces compared to human conversations.1 We typically rely heavily
on non‑verbal cues like gestures, facial expressions, and body language
to convey meaning.2 Yet, when communicating with AI, our communica‑
tion toolkit can become limited to words alone. As such, much like the
“drawer” in our activity who relied solely on verbal instructions, when
interacting with AI tools, we can find ourselves communicating without
the benefit of body language or subtext interpretation. This necessitates a
more direct, specific, and clear approach to our instructions if we wish to
achieve our desired results.
Effectively communicating with AI requires a bit of a micro shift,
wherein we break down our larger goals and complex ideas into
logically sequenced, precise, and unambiguous instructions that AI can
interpret and execute. This “translational thinking” forms the foundation
of effective prompt writing, a skill we’ll explore in depth in Chapter 6.

Better results are achieved when using unambiguous

language

Another hurdle participants tend to encounter during the drawing activity

is the ambiguity inherent in many commonly used words – similarly chal‑
lenging when interacting with AI systems.3

34
 Communicating effectively with AI

Unlike humans, AI lacks the real‑world experience needed to intui‑

tively disambiguate words based on context (if that context is lacking).
This limitation has linguistic roots – the complex evolutionary history of
English incorporating diverse influences has led to many words taking on
multiple meanings and nuances.4 Additionally, for AI systems, language
exists as raw computer code without intrinsic meaning. Thus, devoid of
embodied human experience, current‑day AI lacks our innate sense to
infer intended meanings for ambiguous words. This creates a communica‑
tion barrier when we provide ambiguous instructions to AI.5 For example,
asking an AI system to “Draw a ball” could yield results ranging from play‑
ground balls to crystal balls, depending on its interpretation. To reduce
confusion, it’s important to communicate unambiguously by using pre‑
cise words and phrases that delineate meaning within the given context.
Providing clarifying details and examples further constrains the semantic
scope of our language. As such, rather than saying “Draw a ball”, it might
be better to use an instruction like, “Draw a tennis ball with a yellow
fuzzy texture”.
By consciously minimising the room for multiple interpretations, we
enable smoother exchanges where our intentions align with AI‑generated
outcomes.

Smaller steps guide AI towards achieving larger goals

Reflect on your experience during the drawing activity. You likely guided
your partner through a step‑by‑step process, beginning with a key element
and gradually adding more detailed descriptions. Expecting your partner
to accurately recreate the entire drawing from a single, all‑encompassing
instruction is rather unrealistic.
This insight directly translates to effective communication with AI
­systems. While AI models can be highly capable of handling complex
tasks, if we were to provide only a single instruction (i.e. a zero‑shot
prompt), we are unlikely to be able to achieve overly complex tasks, or
receive suboptimal outputs in response to our requests.
While zero‑shot prompts can be effective for simple tasks, they often
fall short for more complex objectives. Instead of trying to get the AI to
do everything at once, breaking down your larger goal into smaller, incre‑
mental steps/instructions can lead to more accurate and useful outputs.

35
AI-Powered Scholar

This approach, sometimes called chain of thought prompting, allows you

to guide the AI more precisely through each stage of the task.
To implement this strategy, consider doing the following:

1. Start by clearly outlining your overarching objective.

2. Break this goal down into a series of smaller, manageable sub‑tasks.
3. Craft specific instructions for each sub‑task, ensuring they build upon
each other logically.
4. Provide context or examples where necessary to clarify your
expectations.

This approach not only improves the accuracy of AI‑generated outputs but
also gives you more control over the process, allowing for adjustments
and refinements at each step.

Providing examples can help align outputs to inputs

A key observation from the drawing activity is the power of specific exam‑
ples in enhancing understanding and accuracy. This insight is particularly
relevant when communicating with AI systems.
Consider the difference between these two instructions:

1. “Draw a house”.
2. “Draw a red brick house with a grey shingled roof, white picket fence,
and smoke curling from the chimney”.

The first instruction is rather vague and open to wide interpretation. It

could result in anything from a suburban mansion to a child’s playhouse.
The second instruction, however, provides more detail that guides the
drawer (or AI) towards a particular outcome.
When interacting with AI chatbots, providing examples can assist
with, for example:

1. Establishing clear boundaries: Define the scope and style of the

desired output, reducing the AI’s tendency to generate irrelevant or
off‑topic content.

36
 Communicating effectively with AI

2. Setting concrete goals: Give the AI a clear target to aim for, improving
the relevance and quality of its output.
3. Demonstrating desired formats or structures: Examples show the AI
how you want information presented or should be organised.
4. Clarifying abstract concepts: Complex or nuanced ideas become
more tangible when illustrated with concrete examples.
5. Reducing ambiguity: Specific examples can lessen the chances of
misinterpretation or unintended outputs.

Remember, the quality of your examples can directly influence the qual‑
ity of the AI’s output, so aim to choose examples that closely align with
your desired outcome. Furthermore, don’t hesitate to refine or provide
additional examples if the initial results aren’t satisfactory.

Communicating perspective helps shape AI‑generated

outputs

The drawing activity also reveals some insight into how we see the
world: our perspective. When delivering this activity in a classroom, if
you request participants to draw their journey to work that day, they typi‑
cally adopt a first‑person viewpoint, depicting landmarks as they would
see them from inside a vehicle. However, when describing this scene to
their partner, the replicated drawing rarely takes the same perspective. The
“drawer” interprets the instructions and tries to reproduce the image, but
from their perspective – looking at the scene, rather than from within it, or
by providing a bird’s‑eye view.
When engaging with an AI tool, you might consider specifying a per‑
spective you want it to adopt (explored further in Chapter 4). This applies
to various aspects of content generation, including:

1. Narrative voice: Clearly state whether you want content in first‑­

person, second‑person, or third‑person perspective.
2. Temporal perspective: Specify if you need a historical overview,
­current analysis, or future projection.
3. Cultural lens: Indicate if you need a particular cultural viewpoint or a
multicultural approach.

37
AI-Powered Scholar

4. Theoretical framework: Specify the academic or philosophical stance

you want the AI to adopt (e.g., feminist theory, post‑colonial perspec‑
tive, or behaviourist approach).
5. Stakeholder viewpoint: In policy or business contexts, you might
need to specify which stakeholder’s perspective should be prioritised.

By defining the perspective, you guide the AI to generate content that

hopefully aligns more closely with your intentions and needs.
However, despite how clear we may be, it’s important to note the
issue of bias in AI systems.
AI models are trained on vast datasets of human‑generated content,
which inevitably contain societal biases. When you specify a perspec‑
tive, you may inadvertently trigger or amplify these biases. For example,
if you ask an AI to write from a “traditional business leadership perspec‑
tive”, it might generate content that reflects gender biases prevalent in
traditional business literature.6 Similarly, requesting a “historical per‑
spective” on a topic might result in a Eurocentric view if the AI’s train‑
ing data is skewed towards Western historical accounts. We’ll explore
this further in Chapter 7. For now, it’s just important to know that these
biases exist.
To mitigate biases, aim to:

1. Be aware of potential biases in your chosen perspective.

2. Explicitly request diverse viewpoints or counterarguments.
3. Critically evaluate the AI’s output for unintended biases.
4. Consider using multiple, contrasting perspectives to get a more
­balanced view.

Using natural language aids human‑AI communication

In the drawing activity, participants instinctively use conversational lan‑

guage to guide their partners. This natural, fluid communication style
comes effortlessly in human‑to‑human interactions. However, when using
AI tools, many people initially resort to terse, keyword‑based prompts
reminiscent of how many of us are used to performing a Google search.
This shift in communication style can significantly limit the effectiveness
of AI interactions.

38
 Communicating effectively with AI

Modern AI systems, particularly large language models (LLMs), are

designed to understand and respond to natural, conversational language.
These models are trained on vast corpora of human‑written text, includ‑
ing books, articles, and online discussions, which enables them to grasp
nuances, context, and intent in natural language.7
The benefits of using natural language with AI systems are numerous:

1. Improved understanding: AI models can better interpret your intent

when you provide context and details in a conversational manner.
2. More nuanced responses: Natural language prompts allow the AI to
pick up on subtleties and respond with more contextually appropri‑
ate information.
3. Reduced cognitive load: You don’t need to learn a special ‘AI lan‑
guage’ or syntax; you can interact as you would with a human
colleague.

It’s worth noting that while AI systems excel at processing natural lan‑
guage, they don’t truly “understand’ in the way humans do. They gener‑
ate responses based on patterns in their training data. This means they
can sometimes produce fluent, plausible‑sounding text that is factually
incorrect or nonsensical. These are known as “hallucinations” and we’ll
explore this issue in the chapters ahead.

Providing context enhances the relevancy of generated

content

In my AI workshops, I always begin the drawing activity by explaining its

broader context and objectives – aiming to create a framework of under‑
standing for what we’ll be doing. By providing this information, partici‑
pants understand the purpose of the activity, their roles within it, and what
they need to achieve.
Context‑setting can be equally important when interacting with AI
systems. Unlike humans, who can draw upon shared cultural knowledge
or intuitive understanding, AI models heavily rely on the information we
provide them to understand what they should do for us.
By providing context, you’re effectively “priming” the AI model. This
allows the model to filter its vast knowledge base and focus on generating

39
AI-Powered Scholar

content that’s most relevant to your specific needs. It’s akin to giving a
human collaborator a detailed brief before starting a project.
Moreover, context‑setting can help mitigate some of the inherent
limitations of AI models:

• It reduces the likelihood of the AI generating irrelevant or off‑topic

content.
• It helps maintain consistency across multiple interactions or gener‑
ated outputs.
• It can guide the AI towards using appropriate terminology or jargon
for your field.
• It allows you to indirectly influence the ethical stance or perspective
of the generated content.

Remember, the quality of the context you provide directly influences

the quality of the AI’s output. Be as clear and specific as possible, but
also be prepared to refine your context‑setting approach through trial
and error.

Building a shared history enhances long‑term

communication

Lastly, when participants switch roles between describer and drawer,

communication typically improves. This improvement stems from the
shared experience and mutual understanding developed during the
first round, creating a foundation of common reference points and
expectations.
This concept of building a shared history is crucial in enhancing
long‑term communication, both in human interactions and when work‑
ing with AI systems.
In human communication, this shared history forms naturally through
repeated interactions, creating a rich context that allows for more nuanced
and efficient communication over time.
In AI interactions, this shared history is paralleled by what’s known
as a ‘context window’. A context window represents the AI’s capacity

40
 Communicating effectively with AI

to retain and reference information from earlier in the conversation.

­However, unlike human memory, an AI’s context window is typically
limited, and, depending on the AI tool you’re using, may be contained
to a single chat session. The exact size of this context window will vary
depending on the specific AI model you’re interacting with.

Key takeaways

The key takeaways from the ‘Lost in Translation’ activity can be broadly
summarised with the following key points:

• When communicating with AI through text prompts be unambiguous.

• Use natural language/be conversational.
• Provide contextual framing.
• Offer representative examples.
• Break down larger instructions into single‑focused tasks.
• Specify perspective.
• Build a shared history.

With practice, you’ll learn to convey your intents clearly and minimise the
chances of getting lost in translation.

Reflection and discussion

Before progressing to the next chapter, I encourage you to pause and
reflect on the following questions. If you’re working through this book
with others, these questions can serve as a starting point for discussion:

1. Think about a time you’ve had a communication breakdown with

someone, where your message was unclear or misunderstood.
What factors contributed to the ineffective communication? How
might those same challenges apply when communicating with AI
solely through text (or voice) prompts?

41
AI-Powered Scholar

2. 
This chapter discussed strategies like using unambiguous lan‑
guage, providing context, and giving examples when communi‑
cating with AI. Based on your learnings, try writing instructions for
an AI assistant. Then, critically evaluate the instruction, compar‑
ing it to how you might ask a human to complete the same task.
Consider ways you could rewrite your instruction to enhance its
natural language tone.
3. 
This chapter discusses translating broad goals into specific AI
instructions. Practice this translation process for a sample scenario,
like breaking down the tasks involved in brainstorming ideas for a
literature review. Try reducing each task to a single‑focused action
and sequence them in a logical order.

Notes
1. Hill, J., Ford, W. R., & Farreras, I. G. (2015). Real conversations with arti‑
ficial intelligence: A comparison between human–human online con‑
versations and human–chatbot conversations. Computers in Human
Behavior, 49, 245–250. https://doi.org/10.1016/j.chb.2015.02.026
2. Jording, M., Hartz, A., Bente, G., Schulte‐Rüther, M., & Vogeley, K.
(2018). The “social gaze space”: A taxonomy for gaze‑based commu‑
nication in triadic interactions. Frontiers in Psychology, 9, 1–8. https://
doi.org/10.3389/fpsyg.2018.00226
3. Glassman, E L. (2023, September 5). Designing interfaces for
human‑computer communication: An on‑going collection of con‑
siderations. arXiv (Cornell University), 1–5. https://doi.org/10.48550/
arxiv.2309.02257
4. Anderson, R. C. (1990). Inferences about word meanings. In Psychol‑
ogy of Learning and Motivation (pp. 1–16). https://doi.org/10.1016/
s0079–7421(08)60245‑5; Heylighen, F., & Dewaele, J. M. (2002).
Variation in the contextuality of language: An empirical measure.
Foundations of Science, 7, 293–340.
5. Dias, S. M., & Vieira, N. G. (2015). Concept lattices reduction: Defi‑
nition, analysis and classification. Expert Systems with Applications,
42(20), 7084–7097. https://doi.org/10.1016/j.eswa.2015.04.044

42
 Communicating effectively with AI

6. Newstead, T., Eager, B., & Wilson, S. (2023). How AI can perpetuate – Or
help mitigate – Gender bias in leadership. Organizational Dynamics,
52(4), 100998. https://doi.org/10.1016/j.orgdyn.2023.100998
7. Chatterjee, J., & Dethlefs, N. (2023). This new conversational AI
model can be your friend, philosopher, and guide … and even
your worst enemy. Patterns, 4(1), 100676. https://doi.org/10.1016/j.
patter.2022.100676

43
 4 Designing your AI
research assistant

Introduction
Imagine having unlimited research funds to hire your dream assistant.
They’re well‑versed in many knowledge domains, enthusiastic, and avail‑
able 24/7. Sounds perfect, right? But there’s a catch – this assistant has no
idea how you want them to work, which academic norms to follow, or
which of their many skills to apply to your tasks.
If your new hire was human, you’d probably have written a posi‑
tion description before hiring them, and on their first day of work, you’d
provide detailed instructions for how they should perform. Unfortunately,
when we “hire” a chatbot (i.e. collaboratively use AI tools to help us get
our tasks done), these same steps can easily get overlooked. This is why
some people walk away from AI technologies, disillusioned by a chatbot’s
output. Yet others become lifelong champions of the technology.
The difference? Learning to craft the AI into your dream assistant and
clearly communicating what you’d like it to do. In other words, assigning
it a persona, and learning how to write effective instructional commands
to let it know what it should do for you while it’s acting in that role. If that
all sounds a bit strange, don’t worry – and please stick with me – it will
hopefully be clearer by the end of the chapter!
Before we get into specifics, let’s consider the scenario of approach‑
ing a stranger on the street and asking them to perform a task for you. Odd
as it may sound, this situation can closely resemble your first interaction

DOI: 10.4324/9781032665276-444
 Designing your AI research assistant

with an AI chatbot. Both start with a blank slate – no context, no shared

understanding, just potential waiting to be unlocked.

The helpful stranger scenario

Picture this: you’re strolling down a busy city street when you spot a
friendly‑looking stranger. In a rush, you hand them a page of complex
academic text and make an unusual request: “Could you please rephrase
this text using simpler language?”
“Sure!” the stranger replies with a smile.
They quickly scan the dense text, jot down some notes, and hand
back a rewritten version, patiently awaiting your feedback.
You review their work with hopeful expectations, but… oh dear. Your
heart sinks a bit as you read. While the text is indeed simpler now, the
revised wording seems to miss the original’s nuanced meaning. What’s
more, it’s not quite suitable for your intended audience.
This isn’t what you wanted at all!
Perplexed by your apparent disappointment, the stranger apologises
for missing the mark.
“I’m sorry”, they say, looking a bit crestfallen.

I thought I understood your request, but I seem to have missed

your intent. Without knowing the broader context or your spe‑
cific needs, I just wrote what I thought was the most likely option.
What perspective or lens should I have applied in rephrasing this?

And you know what? The stranger has a point.

Without providing clarifying parameters, how could they possibly
interpret your request correctly? For all they knew, you might have wanted
a version suitable for children, or for a blog post. The possibilities were
endless, and they just went with what they considered the most likely
path to follow.
In this scenario, the failure wasn’t necessarily in the stranger’s
response, but in the lack of detail provided to guide their response.
It’s a bit like asking someone to cook you a meal without specifying
any dietary requirements or preferences – you might end up with a per‑
fectly good dish that’s completely unsuitable for you.

45
AI-Powered Scholar

When you approached the stranger and asked them to rephrase the
text, they likely tackled the task based on their lifetime of knowledge and
experience. Their approach would have been shaped by years of reading,
writing, and language exposure, allowing them to analyse the content
and formulate it in what they believed to be an understandable way. In
essence, the stranger provided you with what they predicted to be the
most likely response you desired.
This example is analogous to how AI tools may operate, should you
ask them to rephrase text for you. The tools simply provide you with what
they predict to be the most likely answer to your question or instructional
command. Let’s explore this further below.

Prediction machines

While both humans and chatbots generate text based on past exposure to
language patterns, they differ in that humans possess genuine understand‑
ing, consciousness, and the ability to reason abstractly, whereas chatbots
generate text based on statistical patterns without true comprehension.
The ability of AI to generate text is underpinned by prediction1 –
assessing statistical patterns obtained from analysing vast datasets. The
data they are trained on comes from incredibly large bodies of text, books,
websites, and more. AI models discern subtleties about how language
flows based on this data.2 For example, learning that, in English, “friendly”
often precedes terms like “greeting”, “smile”, or “helpful”, much more
commonly than the word “banana”.
By way of illustration, consider this example of how the training pro‑
cess allows large language models (LLMs) to identify statistical patterns
and relationships between words.
Imagine that an LLM was trained on a simple dataset containing just
the following two sentences:

• The quick brown fox jumped over the lazy dog.

• The slow yellow cat sat on the busy street.

Even from just these two examples, the model can start detecting some
basic patterns, like:

46
 Designing your AI research assistant

• “The” frequently precedes descriptor words like colours and speeds.

• Animated verbs like “jumped” and “sat” often follow the descriptor
words.
• Prepositions like “over” and “on” transition to locations after the
verbs.

Given the above training data, if the LLM was provided with the string of
words “The speedy white rabbit”, it might be more likely to follow it with
words like “hopped through” and “the forest” based on the patterns it rec‑
ognised, even though those exact sentences might not be in its training data.
This process involves internalising probabilities about which word will
follow the next, allowing AI models to rate potential next words and phrases
for the likelihood of sensibly responding to your entered prompts. The pro‑
cess is based on mathematics that’s far beyond the scope of this book. All
you really need to know as a nascent AI user is that as text inputs grow
sentence by sentence, models combine optimal word guesses iteratively.
Before long, through cascading probabilistic predictions, coher‑
ent prose gets generated autonomously. Yet, it’s important to remember
that while these outputs may sound convincingly conversational (i.e.
human generated), the AI holds no true grasp of the meaning behind
the words.3 It simply predicts the most statistically reasonable phrase
continuations based on prior language patterns learned from training
data at a massive scale.
This probabilistic approach to language generation is what allows AI
models to produce coherent and contextually appropriate responses to a
wide range of prompts. However, it’s crucial to understand that this is fun‑
damentally different from human language production, which is grounded
in genuine understanding, intent, and the ability to reason about the world.
While AI can mimic these abilities to a remarkable degree, it remains, at its
core, a sophisticated pattern recognition and prediction system.

Designing your ideal research assistant

One effective way to guide AI tools, particularly AI chatbots, in providing
you with useful content is to assign them a specific role. These roles act as
filters through which the AI calculates the most statistically probable result

47
AI-Powered Scholar

you’re seeking, thus helping it focus its lens and heighten its success in
aligning the content it subsequently generates to meet your needs. The more
specific and detailed you are in guiding an AI chatbot to understand the role
you want it to assume, the less likely you are to experience disappointment
due to misalignment or misunderstanding. This specificity helps bridge the
gap between your expectations and the AI’s output, reducing the chances of
getting “lost in translation”, as we explored in the previous chapter.
In the context of academic research, crafting detailed role descrip‑
tions can transform AI chatbots into your ideal research assistants. By
assigning a role, you’re essentially providing the AI with a framework
within which to operate, much like you would brief a human research
assistant on their responsibilities and expectations when they turn up for
their first day of work.
When you assign a role, you are essentially asking the chatbot to
“act” a certain part, akin to a director assigning a role to an actor. This
role‑playing approach gives the chatbot clear direction on how to align its
generated text with your expectations.
There are endless characteristics you could specify when assigning
a role, but rather than be overwhelmed with options, consider using the
following equation for initial guidance4:

Role = Expertise + Background + Experience + Voice (optional)

+ Quirks (optional)

Let’s explore each of these components in more detail:

1. Expertise: This defines the specialised knowledge or skills your AI

assistant should possess. For instance, you might specify “A published
author skilled in academic writing”, “Qualitative research methods
guru”, or “Thematic analysis expert”.
2. Background: This outlines the disciplinary context your AI assistant
should be familiar with. Examples include “Economics PhD supervi‑
sor”, “Practice‑based nursing academic”, or “Scholarly editor”.
3. Experience: This determines the level of seniority or depth of knowl‑
edge your AI assistant should have. You might specify “A senior
research assistant”, “A subject matter expert in accounting”, or “A
professor who has dedicated their life to the field of developmental
psychology”.

48
 Designing your AI research assistant

4. Voice: This optional characteristic defines how your AI assistant

should communicate. Options might include “Academic tone”,
“Plain English”, or “Conversational”.
5. Quirks: Another optional characteristic, this allows you to assign spe‑
cific behavioural traits to your AI assistant. For example, you might
want it to be “Funny”, “Inspirational”, or to “Play devil’s advocate”.

By carefully combining these role dimensions, you can create a unique

persona tailored to your specific research needs.
The key is to be as specific and detailed as possible when defining
the characteristics you want embodied in the AI’s responses.
For example, consider the difference between these two role
assignments:

1. “Act in the role of a research assistant”.

2. “Act in the role of a senior research assistant with expertise in quali‑
tative methodologies, particularly phenomenology. You have a back‑
ground in sociology and five years of experience working on projects
related to urban gentrification. Communicate in an academic tone
but be prepared to explain complex concepts in simpler terms if
asked”.

The second prompt provides a much richer context for the AI to work
within, likely resulting in more targeted and useful responses.
Remember, while this approach can significantly enhance the AI’s
performance, it doesn’t alter the fundamental nature of how the AI oper‑
ates. It is still using its training data and probabilistic models to generate
responses, but now it’s doing so through the lens of the assigned role,
which helps to narrow and focus its outputs.
Below are some suggestions for different ways you could express
each dimension of the role assignment equation: Role = Expertise + Back‑
ground + Experience + Voice (optional) + Quirks (optional)

Expertise examples:

• Skilled academic writing scholar

• Literature review specialist

49
AI-Powered Scholar

• Methodology advisor
• Gender theory expert
• Ethics advisor

Background examples:

• Entrepreneurship
• Marketing psychology
• Political sociology
• Applied anthropology
• Higher education pedagogy

Experience examples:

• Tenured professor
• Seasoned theorist
• Qualitative guru
• Experienced editor
• Knowledgeable librarian

Voice examples:

• Conversational
• Professional
• Academic
• Plain language
• Friendly

Quirks examples:

• Nerdy sense of humour

• Motivational and inspiring
• Prone to tangents

50
 Designing your AI research assistant

• Quotes poets and philosophers

• Uses positivity and encouragement

Combining variables from the above suggestions lets us create detailed

role descriptions with which we can guide AI tools towards becoming our
dream research assistants. Here are some examples:

• “Assume the identity of a qualitative expert with a specialisation in

skilled academic writing. Adopt an academic and analytical voice to
dissect complex qualitative data. As a respected supervisor and pub‑
lished author, use your wise perspective to mentor others in crafting
compelling academic narratives, and don’t hesitate to quote poets
and philosophers to illustrate your points”.
• “Embody the role of mixed methods extraordinaire who is also a lit‑
erature review specialist. Take on a professional and objective voice
to guide users through the intricacies of combining qualitative and
quantitative data in academic writing. As someone who gets exis‑
tential at times, weave in philosophical musings about the nature of
research and evidence”.
• “Take on the persona of an ethics advisor with a deep understanding
of qualitative data analysis. Use a compassionate and encouraging
voice to navigate the ethical dilemmas that can arise in qualitative
research, particularly in sensitive areas like medical ethics or social
policy. Infuse your dialogue with historical examples to provide con‑
text and depth”.
• “Act in the role of a seasoned theorist and qualitative guru with a
background in applied anthropology. Speak with a conversational yet
academic tone, making complex qualitative theories accessible to a
wider audience. As a knowledgeable librarian and skilled academic
writer and scholar, sprinkle your conversations with nerdy humour
and funny analogies to keep the discourse engaging”.
• “Act in the role of an analytical mastermind with expertise in quali‑
tative data visualisation and a background in higher education.
Communicate with a technical and professional voice, translating
qualitative insights into compelling narratives. As someone who
loves to use funny analogies and knows random science trivia, make
your explanations as informative as they are entertaining”.

51
AI-Powered Scholar

In addition to using the above examples, you might wish to save the
role descriptions you create in a document, so they are easily accessed
when, at a later date, you wish to perform another task for which that role
would be ideal. You might also try adjusting the parameters within the role
description, and then observing how even small tweaks can influence AI
chatbot outputs.
With practice, you’ll be able to manifest your ideal AI assistant for
any given research context.

Ethical considerations
The practice of assigning personas to AI systems raises ethical concerns
around identity and representation. Asking your AI chatbot to take on the
role of “real” people may be problematic, depending on your intended
use for the AI‑generated outputs.
The personas assigned to AI can have real‑world impacts, such as if
controversial views are espoused without the original thinker’s permis‑
sion. Some popular examples include Australian music artist Nick Cave’s
identity being appropriated to generate song lyrics, much to his chagrin.5
And AI being used to maliciously impersonate celebrities for scam‑related
criminal activities.6
When an AI adopts the identity of a real person without their con‑
sent, it can infringe on ethical boundaries, intellectual property rights, and
individual autonomy. Generating output that seems to legitimately come
from someone’s perspective when it does not can enable the spread of
misinformation, damage reputations, and undermine public trust.
When assigning AI personas based on real‑life academics, practitio‑
ners, or public figures, you might consider the following:

• Do you have permission to co‑opt their identity for use with AI? If not,
is their representation accurate and respectful?
• Will persona responses be contextually appropriate and vetted for
integrity? What controls are in place?
• Does the persona align with the individual’s ethics and code of
conduct?
• Could unauthorised use negatively impact the person’s reputation or
intellectual property?

52
 Designing your AI research assistant

Rather than appropriating real identities without consent, I suggest a safer

approach of creating fictional personas comprising an amalgamation of
real‑world traits.
Lastly, when assigning personas, it’s also important to remember that
AI has no innate identity. Its “personhood” exists only insofar as humans
project characteristics onto it through prompts and conditioning; AI sys‑
tems like chatbots have no intrinsic experiences, beliefs, desires, or inten‑
tions. Chatbot persona’s act as an illusion of personhood and help guide
content creation.
Despite their machine nature, people often anthropomorphise AI. I’m
guilty of this myself, and as you might have noticed throughout this book,
I frequently slip into anthropomorphised references to AI. This is partly
to provide you with a smoother reading experience, but also because I
probably spend more time chatting with AI chatbots than humans, mak‑
ing them feel eerily human‑like. Rest assured, I’m well aware they’re not
actually human.
A famous example of someone overestimating machine capabilities
is Google engineer Blake Lemoine, who believed Google’s LaMDA chat‑
bot displayed signs of sentience after interacting with it.7 Another exam‑
ple, albeit from fiction, is the movie “Her”, where the protagonist falls in
love with his AI assistant, Samantha.8 However, romantic attachments to
AI aren’t limited to Hollywood. Replika, an AI chatbot app, allows users
to create virtual partners.9 Yes, you read that correctly: you can now date
an AI. When the company issued an update to dial down the “romance”
settings, some users reported feeling their companions had been lobot‑
omised, leaving them extremely distressed. This event was dubbed the
“Valentine’s Day Massacre” – a term that speaks volumes about the emo‑
tional connections users can form with AI bots.
While it’s easy to fall in love with the level of assistance AI can pro‑
vide, it’s probably a good idea to interact with AI tools in a way that’s
firmly grounded in reality. They are, at their core, sophisticated prediction
machines – incredibly helpful, but not sentient beings.

Activity
After reading through this chapter, you’re probably itching to test out your
new persona‑writing skills. Here’s an activity to get started with:

53
AI-Powered Scholar

Designing your dream research assistant

Try crafting a persona for your perfect research assistant by following the
steps below.

1. First, write down descriptors for this fantasy assistant. What is their aca‑
demic background? What about their tone of voice, expertise, quirks,
or perspective do you want to include? Consider how those descriptors
might influence the way an LLM responds to your requests for assistance.
2. Next, scope characteristics for your AI research assistant. Refer to the
notes you jotted down earlier about your ideal assistant. Define their
academic background, tone of voice, perspective, expertise, etc. Get
creative!
3. Draft a “Act in the role of…” prompt, using natural language in the
form of a sentence, or short paragraph. Use the example role prompts
from this chapter, or the one below, to help draft your prompt.
For example: “Act in the role of a knowledgeable mentor provid‑
ing research advice to a new graduate student. Use simple, straight‑
forward language and offer practical tips based on years of experience
as an approachable and supportive mentor”.

With practice, you can develop a repertoire of personas tailored to differ‑

ent contexts and conversational goals. The ability to essentially manifest
your perfect research assistant on demand will take your AI collaboration
to the next level.

Reflection and discussion

Use the following questions as a starting block to stimulate thinking
and spark discussion:

1. How might assigning a persona to an AI chatbot impact the rel‑

evance and usefulness of its responses?
2. Think of a research task or project. If you had endless funds and
the ability to transcend time, who would you hire to help you

54
 Designing your AI research assistant

complete the project? Describe their background, expertise, tone

of voice, etc. Consider combining characteristics across multiple
ideal hires to invent your dream fictional hire.
3. Try writing 2–3 sample “Act in the role of…” prompts for different
research scenarios, such as literature reviews, data analysis, and
academic writing. How might you word each persona?
4. What challenges might arise when using a single persona across
multiple research tasks?

Notes
1. Agrawal, A., Gans, J., & Goldfarb, A. (2023, June 12). How large
language models reflect human judgment. Harvard Business
Review. https://hbr.org/2023/06/how‑large‑language‑models‑reflect‑
human‑judgment; Mirchandani, S., Xia, F., Florence, P., Ichter, B.,
Driess, D., Arenas, M. G., Rao, K., Sadigh, D., & Zeng, A. (2023).
Large language models as general pattern machines. arXiv (Cornell
University). https://doi.org/10.48550/arxiv.2307.04721
2. Johri, S. (2023, June 8). The making of ChatGPT: From data to dia‑
logue. Science in the News. https://sitn.hms.harvard.edu/flash/2023/
the‑making‑of‑chatgpt‑from‑data‑to‑dialogue/
3. Mitchell, M. (2021). Why AI is harder than we think. arXiv (Cornell
University), 1–12. https://doi.org/10.48550/arxiv.2104.12871
4. Eager, B., & Brunton, R. (2023). Prompting higher education towards
AI‑augmented teaching and learning practice. Journal of University
Teaching and Learning Practice, 20(5), 1–21. https://doi.org/10.53761/
1.20.5.02
5. Savage, B. M. (2023, January 17). Nick Cave says ChatGPT’s AI
attempt to write Nick Cave lyrics “sucks”. BBC News. https://www.
bbc.com/news/entertainment‑arts‑64302944
6. Hsu, T., & Lu, Y. (2024, January 26). No, that’s not Taylor Swift
peddling le Creuset cookware. The New York Times. https://www.
nytimes.com/2024/01/09/technology/taylor‑swift‑le‑creuset‑ai‑
deepfake.html

55
AI-Powered Scholar

7. Tiku, N. (2023, May 23). The Google engineer who thinks the com‑
pany’s AI has come to life. Washington Post. https://www.washington‑
post.com/technology/2022/06/11/google‑ai‑lamda‑blake‑lemoine/
8. Ntelia, R. E. (2024). Love is in the AI of the Beholder Artificial
Intelligence and characters of love. E‑REA, 21.2. https://doi.org/
10.4000/11wa6
9. Purtill, J. (2023, February 28). Replika users fell in love with their AI
chatbot companions. Then they lost them. ABC News. https://www.
abc.net.au/news/science/2023‑03‑01/replika‑users‑fell‑in‑love‑with‑
their‑ai‑chatbot‑companion/102028196

56
 5 A brief history of
conversational AI

Introduction
From science fiction to everyday reality, the journey of conversational AI
has been nothing short of remarkable. As we find ourselves at a moment
in time where chatting with an AI feels as natural as texting a friend, it’s
easy to forget the decades of innovation and perseverance that brought us
here. Reflecting on this timeline, this chapter traces the evolution of con‑
versational AI, offering both a historical perspective and an appreciation
for the technology that many of us now take for granted.
Understanding the history of AI is crucial for several reasons. Impor‑
tantly, it provides essential context for comprehending the capabilities
and limitations of modern AI systems. By exploring key milestones and
breakthroughs, we can better appreciate the rapid advancements of recent
years and more accurately anticipate future developments. This foresight
allows us to prepare for the integration of AI into our future academic
workflows. Additionally, gaining insight into AI from a historical perspec‑
tive lets us not only understand the technical progress that’s occurred in
the field of AI but also the persistent challenges and considerations that
have accompanied its development. We’ll be focusing on the fundamen‑
tal developments and principles underlying modern AI chatbots, setting
the stage for a deeper understanding of how to effectively interact with
and leverage these powerful tools in academic contexts in future chapters.
Let’s begin the journey by first clarifying what’s meant by the term
“conversational AI”.

DOI: 10.4324/9781032665276-557
AI-Powered Scholar

What is conversational AI?

Conversational‑style AI, often referred to as “conversational AI” or “chat‑
bots”, represents a category of artificial intelligence designed to interact
with humans through natural language.
These systems are built to understand, process, and generate human‑
like text or speech, enabling them to engage in dialogue that mimics human
conversation.
Unlike traditional computer interfaces that require specific com‑
mands or clicks, conversational AI allows users to communicate as
they would with another person, using everyday language to ask ques‑
tions, request information, or complete tasks. At its core, conversational
AI strives to bridge the gap between human communication and com‑
puter processing, creating a more seamless and user‑friendly interface for
human‑machine interaction.
While we may take the “chatty” capabilities of modern AI for granted,
the journey to achieve this level of natural interaction has been long and
complex, encompassing decades of research, innovation, and technologi‑
cal breakthroughs that contributed to creating the experiences we enjoy
with AI today.
To get an idea of how far things have advanced, I invite you to imagine
attempting to communicate with a computer in the 1960s. The experience
would likely resemble an interaction with a particularly uncooperative
vending machine – full of pre‑programmed responses and devoid of genu‑
ine understanding. Now, fast forward to our present day, where we find
ourselves engaging with AI chatbots like ChatGPT, capable of discussing
topics ranging from quantum physics or indulging us in writing quirky
poems. The progress is nothing short of astonishing.
While modern‑day chatbots such as ChatGPT showcase remarkably
advanced natural language capabilities (for the latest ChatGPT updates,
visit https://openai.com/research), the fundamental technique of text‑based
prompting actually traces back several decades; the AI models we interact
with today are built upon the groundwork laid by early pioneers in the
field – researchers and technologists who envisioned machines capable of
engaging in meaningful conversation with humans.
Interestingly, despite the dramatic transformation of AI itself (imag‑
ine a caterpillar metamorphosing into a butterfly, but with more silicon
involved), the basic prompting approach – the two‑way communication

58
 A brief history of conversational AI

between humans and machines – remains surprisingly recognisable.

At least from a user’s perspective. It’s as if we’ve been engaged in the
same game of conversational tennis for decades, and this continuity in
approach persists despite the vast improvements in AI’s capabilities and
sophistication.
In the early stages of development, chatbots were about as flexible
as a brick wall, relying on simple scripted responses. You would input a
query, and they would promptly generate a pre‑programmed reply with
mechanical efficiency. Now, however, modern natural language process‑
ing allows for flexible, contextual conversations that can occasionally
leave one pondering whether they’re engaging with a machine or a par‑
ticularly eloquent human interlocutor.
In essence, conversational AI can be thought of like the Swiss Army
knife of the digital world – it’s a multi‑faceted tool designed to engage in
meaningful and natural dialogues with humans.
But what makes these conversational AI systems tick? Let’s break it
down:

• Natural language processing: This is the AI’s ability to “understand”

our text or spoken words. It’s like giving the AI a universal translator
for human communication.
• Contextual awareness: This allows the AI to follow the flow of a con‑
versation, much like how you keep track of what your friend is talk‑
ing about over coffee (assuming you’re not staring at your phone!).
• Generative skills: This is the AI’s ability to craft relevant, personalised,
and coherent responses. It’s not just parroting information but synthe‑
sising it into something new and appropriate.
• Interactive interfaces: These are the digital platforms where you
engage with the AI, typically through a chat window or text box on
a screen. They’re designed to mimic the back‑and‑forth nature of
human conversation, allowing you to type your messages and receive
the AI’s responses in real‑time. The goal is to make the interaction feel
as natural and intuitive as conversing with a friend, despite the fact
that you’re communicating with a machine.

Remember how we touched on the idea of AI prediction machines in the

last chapter? Conversational AI uses this ability to sift through massive

59
AI-Powered Scholar

amounts of human dialogue data, learning patterns, and nuances. It’s

like the AI is eavesdropping on millions of conversations to learn how
we communicate. This statistical learning is what allows the AI to handle
nuanced language, adapt to context, and even improve its conversational
skills over time.
While early conversational AIs were text‑only (kind of like having a
pen pal), we’ve now entered the era of multi‑modal AI.1 This means that
AI can now interact through various modes, not just text, including voice,
images, and video.
In this book, though, we’ll be focusing on text‑based conversations.
Why? Because mastering text interactions with AI chatbots is like learning
to walk before you run – it’s a fundamental skill that will serve you well in
your journey to becoming an AI‑savvy scholar.
A timeline of AI advancements is offered below. If you’d rather skip
the mini history lesson and jump straight into learning how to have effec‑
tive conversations with AI through text‑based prompts, feel free to skip
ahead to the next chapter.

Tracing the evolution of conversational AI

The development of conversational AI represents a remarkable journey in
computer science, the efforts of passionate and talented researchers, and
a continuous pursuit to challenge our understanding of language, intel‑
ligence, and human‑machine interaction. This evolution, spanning many
decades, illustrates both the persistent vision of researchers and the expo‑
nential growth of technological capabilities that allowed their visions to
be realised. Let’s explore some key developments that shaped the journey.

The 1960s: ELIZA and the birth of conversational AI

In the 1960s, MIT computer scientist Joseph Weizenbaum shared his pio‑
neering work on conversational AI with ELIZA,2 which was one of the first
programs designed for natural language interaction.
ELIZA simulated a psychotherapy session, engaging users with
prompts such as “How do you feel today?” and “Tell me more about your

60
 A brief history of conversational AI

family”. The program operated on a simple pattern‑matching a­ lgorithm,

identifying keywords in the user’s input and responding with pre‑­
programmed phrases. While this could be considered simplistic by today’s
AI standards, ELIZA represents a key milestone in AI development as it
demonstrated how even constrained conversation could create an illusion
of understanding and empathy.
Many users attributed human‑like qualities to the program, some
even reporting emotional connections. This phenomenon, later termed
the “ELIZA effect”, highlighted humans’ tendency to anthropomorphise
technology – that is, to attribute human characteristics, behaviours, or
emotions to non‑human entities or objects. In the case of ELIZA, users
might ascribe human‑like understanding and empathy to a computer pro‑
gram that was, in reality, simply following pre‑programmed rules with‑
out any genuine comprehension or emotional capacity. This tendency to
anthropomorphise reflects our innate inclination to perceive human‑like
traits in our interactions, even when dealing with what are clearly artifi‑
cial systems.
Despite its ostensible success, Weizenbaum cautioned against
overestimating ELIZA’s cognitive abilities. He became increasingly con‑
cerned about the ethical implications of AI, warning against the poten‑
tial dehumanisation that could result from over‑reliance on computer
interactions.

The 1970s and 1980s: SHRDLU and the quest

for understanding

The 1970s and 1980s witnessed incremental advancements towards more

versatile interaction. A significant contribution came from Terry Winograd,
who created SHRDLU, a program that engaged users in basic discussions
about a virtual world of blocks, showcasing improved context awareness
and the ability to reference objects across multiple sentences.3
SHRDLU operated in a limited domain – a simulated block world –
but within this context, it could understand and execute complex com‑
mands. For example, it could respond to instructions like “Pick up the
red block that is on the blue block” by correctly interpreting the rela‑
tionships between objects. This ability demonstrated a level of language

61
AI-Powered Scholar

comprehension that went beyond simple pattern matching and was

part of a broader research trend exploring natural language understand‑
ing through microworlds – limited domains where the complexities of
real‑world language could be controlled and studied.
While these approaches showed promise within their constrained
environments, they struggled to scale to more general language‑under‑
standing tasks.

The 1980s: the rise of expert systems

A significant leap occurred with the advent of expert systems in the

1980s. You can think of an expert system as a type of artificial intelligence
program designed to replicate the decision‑making ability of a human
expert in a specific field.4 These systems encoded vast domain‑specific
knowledge into rule‑based frameworks, enabling more specialised con‑
versations.5 And, unlike previous generalist approaches, expert systems
focused on becoming highly proficient in narrow domains.
One famous expert system was MYCIN, developed at Stanford
University in the early 1970s and refined throughout the 1980s. It was
designed to identify bacteria causing severe infections, wherein MYCIN
could engage clinicians in professional medical discussions, asking rel‑
evant questions about symptoms and test results before providing diag‑
nostic recommendations.6 While it was impressive, the system also
highlighted the limitations of rule‑based approaches. Specifically, the
manual nature of knowledge engineering in these systems ultimately lim‑
ited their flexibility and ability to learn dynamically. Updating an expert
system’s knowledge often required extensive reprogramming, making it
challenging to keep pace, especially in rapidly evolving fields.
Advancing towards more fluid conversational exchanges required
better representing linguistic knowledge and reasoning capabilities. Sta‑
tistical approaches sought to model language use patterns across corpora
(i.e. large and structured sets of texts, which serve as a basis for linguis‑
tic analysis) but dealt poorly with longer‑range coherence. Luckily, the
rise of machine learning and neural networks in the 1990s brought new
horizons. Given enough examples of conversational transcripts, AI sys‑
tems began to “understand” complex associations between utterances,

62
 A brief history of conversational AI

their meanings, and could thus generate more appropriate responses

to user queries.

The 1990s: machine learning and neural networks

The rise of machine learning and neural networks in the 1990s opened
new horizons for conversational AI. Machine learning is a branch of AI
that allows computers to learn from data without being explicitly pro‑
grammed. Neural networks, inspired by the human brain’s structure, are a
type of machine learning model consisting of interconnected nodes (like
neurons) that can recognise patterns in data. These approaches allowed
AI systems to glean complex associations between utterances, their mean‑
ings, and appropriate responses from large datasets of conversational
transcripts.
This period of AI development marked a shift towards more adap‑
tive and context‑aware dialogue systems, closer to what we’re familiar
with when using today’s AI tools. It also provided the capabilities for us
researchers to perform tasks like summarising academic papers, for which
I will be forever grateful!
These early experiments with neural networks for language tasks
showed promise and paved the way for future advancements. For exam‑
ple, Hochreiter and Schmidhuber introduced Long Short‑Term Memory
(LSTM) networks, a type of recurrent neural network. Recurrent neural
networks are designed to work with sequences of data, making them
particularly suited for language tasks. LSTMs are capable of learn‑
ing long‑term dependencies in sequential data, which means they can
remember important information for long periods as they process text.
This innovation would later prove crucial for improving the coherence of
AI‑generated text over longer passages, allowing AI to maintain context
and produce more natural‑sounding language.
The decade also saw the development of statistical methods for natu‑
ral language processing (NLP), which is the field concerned with how
computers understand and generate human language. NLP techniques
allowed for more robust language modelling, involving predicting the
likelihood of a sequence of words. While perhaps not as flashy as neu‑
ral networks, it laid important groundwork for future advancements in

63
AI-Powered Scholar

language understanding by providing ways to represent and analyse the

structure of language mathematically.

The 2010s: practical conversational assistants and deep

learning

The 2010s saw the emergence of practical conversational assistants and

significant advancements in deep learning techniques. For example, in
2011, Apple introduced Siri, demonstrating the ability to handle natural
language commands for everyday tasks like setting reminders or making
phone calls.7 While Siri’s initial capabilities were limited, it marked a sig‑
nificant step in bringing conversational AI to the mainstream.
Meanwhile, IBM’s Watson showcased advanced question‑answering
capabilities by defeating human champions on the game show Jeopardy!8
Watson’s success relied on a combination of natural language processing,
information retrieval, and machine learning techniques, processing vast
amounts of unstructured data to generate answers in real‑time.
The decade also saw rapid improvements in machine translation
and other specialised conversational domains. Google’s Neural Machine
Translation system, introduced in 2016, significantly improved translation
quality by considering entire sentences rather than translating phrases in
isolation. Google Duplex further pushed boundaries by conducting life‑
like phone conversations to complete tasks like restaurant reservations.9
The demo, which featured an AI assistant making calls with remarkably
human‑like speech patterns, sparked both excitement and ethical con‑
cerns about AI’s potential to deceive, and could perhaps be considered an
early precursor to the concerns regarding the deep fake AI models we are
increasingly seeing today.
The development of transformer models, introduced in the now
famous 2017 paper “Attention Is All You Need”,10 authored by Google
scientists, revolutionised natural language processing. This breakthrough
was akin to giving AI a super‑powered set of reading glasses, dramati‑
cally improving its ability to understand and generate human language. To
understand the significance of transformers, imagine trying to perform the
task of understanding a long, complex sentence. Earlier AI models would
struggle to keep track of relationships between words that were far apart,

64
 A brief history of conversational AI

much like a person trying to remember the beginning of a sentence by the

time they reached the end. Transformer models, however, were capable of
“paying attention” to all parts of the input simultaneously, allowing them
to grasp context and meaning much more effectively. The key innovation
of transformers is their attention mechanism. This allows the model to
focus on different parts of the input text when producing each word of the
output, much like how humans might emphasise certain words or phrases
when explaining a complex idea. This attention mechanism enables trans‑
formers to handle long‑range dependencies in text, capturing nuances
and relationships that previous models often missed.
The impact of transformer models on AI language capabilities upped
the AI game. They paved the way for more advanced language models
like Bidirectional Encoder Representations from Transformers (BERT) and
Generative Pre‑trained Transformer (GPT), which set new benchmarks
in tasks ranging from translation to question‑answering to text genera‑
tion. These models form the foundation of many AI language tools we use
today, including the chatbots that are helping me to rephrase many of the
sentences in this book to hopefully improve your reading experience.
Transformer models marked a turning point in AI’s ability to process
and generate human language, bringing us closer to the long‑standing
goal of machines that can truly understand (in as far as a machine can be
considered to “understand”) and communicate in natural language.

2020 and beyond: large language models

The introduction of ChatGPT in November 2022 marked a pivotal moment

in conversational AI. Developed by OpenAI, ChatGPT, (you’ll remem‑
ber that “GPT” stands for “generative pre‑trained transformer”) exhib‑
ited unprecedented natural language capabilities and generation across
diverse topics and contexts.11 ChatGPT demonstrated an ability to engage
in human‑like dialogue, answer follow‑up questions, admit mistakes, and
even, although Nick Cave would probably disagree, do an impressive job
at writing song lyrics.
ChatGPT’s impact reverberated across various sectors, particularly in
education. Its ability to generate coherent essays, solve complex prob‑
lems, and provide detailed explanations prompted urgent reconsideration

65
AI-Powered Scholar

of established practices and policies in academic institutions world‑

wide.12 Concerns about academic integrity and the changing nature of
assessment led to debates about how to integrate or regulate AI use in
educational ­settings13 – a debate that continues to rage.
As we approach the mid‑2020s, projects like Google’s Language
Model for Dialogue Applications14 (LaMDA) and Anthropic’s Claude15 are
pushing the boundaries of AI conversation even further. LaMDA aims to
engage in open‑ended conversations on virtually any topic. Meanwhile,
Claude focuses on safety and ethical considerations, attempting to pro‑
vide helpful responses while avoiding potentially harmful or biased con‑
tent. Each model has its own particular flavour. Furthermore, companies
such as Neuralink are exploring direct neural interfaces for human‑AI
interaction16 (you read that correctly – the company is aiming to plug AI
directly into our brains!) raising new possibilities and ethical questions
about the future of human‑machine communication. Neuralink’s efforts
to develop brain‑computer interfaces aim to enable direct thought‑to‑text
communication, fundamentally altering the landscape of conversational
AI. It’s anyone’s guess as to what these developments mean for how we
conduct academic research in the future.
Overall, the trajectory from ELIZA to ChatGPT represents more than
just technological progress; it reflects a deepening exploration of the
nature of intelligence, consciousness, and communication. As these sys‑
tems grow more sophisticated, maintaining a critical perspective on their
capabilities and limitations remains crucial.
In the words of science fiction writer Arthur C. Clarke,17 any suffi‑
ciently advanced technology is indistinguishable from magic, represent‑
ing a sentiment that captures both the wonder and the caution with which
to approach conversational AI.
As these systems become increasingly integrated into our daily
lives, understanding their underlying principles, capabilities, and limi‑
tations becomes not just an academic exercise, but a societal impera‑
tive. The future of conversational AI promises continued innovation, but
also demands ongoing ethical reflection and responsible development
to ensure that these powerful tools benefit humanity as a whole. And,
while the capabilities of modern AI might seem magical, it’s important
to remember that beneath the surface, these systems are fundamen‑
tally based on statistical inference and pattern recognition, not mys‑
tical powers. As we discussed in the previous chapter, the quality of

66
 A brief history of conversational AI

our interactions with AI chatbots, at least in their current form – where

we tend to interact with AI via our keyboards rather than having them
embedded in our brains – largely depends on the quality of our inputs
(i.e. prompts).
Prompts remain the primary interface between humans and AI chat‑
bots, serving as the bridge that translates our intentions into actionable
instructions. Mastering the art of crafting well‑structured prompts also
remains crucial for maximising the potential of AI tools to help us achieve
our academic tasks.
To draw an analogy from the world of Harry Potter, consider the
difference between Ron Weasley’s often misfiring spells and Hermione
Granger’s precisely executed incantations. Ron’s haphazard wand‑wav‑
ing, despite having access to magical power, frequently leads to unin‑
tended results. Hermione, on the other hand, (who, I would argue, is
actually the star of the show and who deserves the global franchise to be
retitled after) achieves remarkable feats through her meticulous study and
careful application of magical techniques.
In the context of AI, our “wand” is the language we use to com‑
municate with these systems, and our “spells” are the prompts we craft.
Just as Hermione’s success stems from her deep understanding of magical
principles, our effectiveness in using AI tools depends on our skill in for‑
mulating clear, purposeful instructions.
In the next chapter, we’ll unpack the principles and techniques of
prompt writing, equipping you with the skills to craft precise and effec‑
tive “AI spells”. But before we jump into the practicalities of prompts,
let’s continue our exploration of AI’s history, focusing on the evolution
of human‑AI interaction methods by tracing the development of prompts
themselves.

Prompt engineering
Prompt engineering is the strategic process of designing and refining
instructions or questions to elicit specific, desired responses from AI mod‑
els.18 Prompts can be used to guide the AI towards a desired output, to
provide it with context, and to fine‑tune the model’s performance on a
specific task. I can’t stress enough that the ability to craft effective prompts
is an indispensable skill in our AI literacy toolkits!

67
AI-Powered Scholar

To fully appreciate prompt engineering, it’s helpful to understand

how our interaction with AI has evolved. As previously mentioned, early
AI systems required precise, often code‑like commands. By contrast,
today’s models can interpret and respond to natural language, making
them more accessible but also more nuanced in their requirements for
effective communication. Among these advancements, the concepts of
zero‑shot and few‑shot learning19 stand out as transformative approaches
in prompt engineering.

• Zero‑shot learning: This refers to an AI model’s ability to understand

and perform tasks it has never explicitly been trained to do. This
capability can be considered revolutionary because it enables AI
systems to apply learned knowledge from one domain to unfamil‑
iar tasks or categories without additional training data. In practical
terms, zero‑shot learning allows researchers to query AI models on
niche subjects or generate content in styles that were not necessarily
part of the model’s training data.
• Few‑shot learning: This takes zero‑shot learning a step further by
enabling AI models to learn new tasks or understand new concepts
from just a few examples. Few‑shot learning is particularly useful
in academic research and specialised fields where data scarcity is
common. By providing a few examples within the prompt, users can
guide the AI to generate content or analyse data in specific, desired
ways.
• Continuous learning and adaptation: Beyond zero‑shot and few‑shot
learning, another significant trend is the move towards continuous
learning models. These AI systems are designed to learn and adapt
to new data or interactions over time, gradually improving their per‑
formance and expanding their knowledge base. Continuous learning
models offer the potential for more personalised and context‑aware
interactions with AI.

To appreciate the intricacies of prompt engineering, it’s essential to under‑

stand the underlying natural language processing mechanisms of AI mod‑
els,20 particularly in the realms of Natural Language Understanding (NLU)
and Natural Language Generation (NLG). These two processes are at the

68
 A brief history of conversational AI

heart of how AI systems interpret prompts and produce coherent, relevant

responses.

• Natural Language Understanding: NLU is a subset of NLP that

deals with machine reading comprehension. It enables AI models
to understand and interpret human language in a way that cap‑
tures meaning, sentiment, and intent. When you provide a prompt
to an AI system, the NLU component breaks down the text into
understandable segments, analysing syntactic structures (grammar
and sentence structure) and semantic information (i.e. meaning).
The AI model uses this analysis to grasp the context, questions, or
tasks you’re presenting. It identifies key elements such as the subject
matter, the desired action (e.g. summarising, analysing), and any
specific constraints or requirements embedded in the prompt. This
understanding is crucial for generating a response that aligns with
the user’s intentions.
• Natural Language Generation: Once the AI understands the prompt
through NLU, it proceeds to NLG, which, in the context of academic
writing, could be thought of as generating text from data. This process
involves selecting appropriate words, constructing sentences, and
organising coherent paragraphs that address the prompt’s require‑
ments. NLG is not merely about stringing words together; it requires
the model to apply creativity, reasoning, and sometimes even a
degree of empathy to produce responses that are not only relevant
but also engaging and human‑like.

The magic of prompt engineering lies in the seamless interplay between

NLU and NLG. Effective prompts are designed to leverage this interplay,
guiding the AI through the NLU phase to ensure it correctly understands
the task at hand, and then through the NLG phase to produce a tailored,
coherent response. For instance, a well‑crafted prompt that asks the AI to
“Summarise the key points of the following text in a bullet‑point format”
clearly directs the NLU component to focus on the extraction and con‑
densation of information. The NLG component, meanwhile, is guided to
format the output as a list of bullet points, adhering to the concise and
direct nature of the instruction. Understanding the roles of NLU and NLG

69
AI-Powered Scholar

in processing prompts helps us appreciate the importance of clarity, speci‑

ficity, and context in prompt engineering.
As AI capabilities continue to advance, the importance of skilled
prompt engineering in academia will undoubtedly grow. This means that
the ability to effectively communicate with and guide AI systems will
become as crucial as any other research skill. While it’s never too late
to start learning, I strongly advocate beginning this journey sooner rather
than later. The earlier researchers can integrate these skills into their work‑
flow, the better positioned they’ll be to leverage AI’s full potential and
supercharge their research workflows.
With this in mind, we’re now ready to move beyond the historical
context and theoretical foundations explored thus far. In the next chap‑
ter, we’ll turn to the practical art of prompt writing, exploring concrete
techniques, best practices, and real‑world examples that will hopefully
empower you to craft effective prompts and help you become an effective
AI‑powered scholar.

Notes
1. OpenAI. (2023, September 25). ChatGPT Can Now See, Hear, and
Speak. https://openai.com/blog/chatgpt‑can‑now‑see‑hear‑and‑speak
2. Weizenbaum, J. (1966). ELIZA—A computer program for the study
of natural language communication between man and machine.
Communications of the ACM, 9(1), 36–45. https://doi.org/10.1145/
365153.365168
3. Winograd, T. (1971). Procedures as a Representation for Data in a
Computer Program for Understanding Natural Language. https://apps.
dtic.mil/sti/pdfs/AD0721399.pdf; Winograd, T. (1980). What does it
mean to understand language? Cognitive Science, 4(3), 209–241.
4. Buchanan, B. G., & Smith, R. G. (1988). Fundamentals of expert sys‑
tems. Annual Review of Computer Science, 3(1), 23–58. https://doi.
org/10.1146/annurev.cs.03.060188.000323
5. Gaschnig, J., Klahr, P., Pople, H., Shortliffe, E., & Terry, A. (1983).
Evaluation of expert systems: Issues and case studies. Building Expert
Systems, 1, 241–278.
6. Shortliffe, E. H. (1977, October 5). Mycin: A knowledge‑based
computer program applied to infectious diseases. In Proceedings of

70
 A brief history of conversational AI

the Annual Symposium on Computer Application in Medical Care

(pp. 66–69). American Medical Informatics Association.
7. Aron, J. (2011). How innovative is Apple’s new voice ­ assistant,
Siri? New Scientist, 212(2836), 24. https://doi.org/10.1016/s0262–
4079(11)62647‑x
8. Ferrucci, D. A. (2012). Introduction to “this is watson”. IBM Journal of
Research and Development, 56(3.4), 1–1.
9. O’Leary, D. E. (2019). GOOGLE’S Duplex: Pretending to be human.
Intelligent Systems in Accounting, Finance and Management, 26(1),
46–53. https://doi.org/10.1002/isaf.1443
10. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L.,
Gomez, A. N., Kaiser, L., & Polosukhin, I. (2017). Attention is all you
need. arXiv (Cornell University), 30, 5998–6008. https://arxiv.org/
pdf/1706.03762v5
11. OpenAI. (2022, November 30). Introducing ChatGPT. https://openai.
com/blog/chatgpt
12. Lo, C. K. (2023). What is the impact of CHATGPT on education? A
rapid review of the literature. Education Sciences, 13(4), 410. https://
doi.org/10.3390/educsci13040410
13. Perkins, M. (2023). Academic integrity considerations of AI large lan‑
guage models in the post‑pandemic era: ChatGPT and beyond. Jour‑
nal of University Teaching & Learning Practice, 20(2), 1–26. https://
doi.org/10.53761/1.20.02.07
14. Collins, E., & Ghahramani, Z. (2021, May 18). LaMDA: Our breakthrough
conversation technology. Google. https://blog.google/technology/ai/
lamda/
15. Anthropic. (2023, March 14). Introducing Claude. Anthropic. https://
www.anthropic.com/news/introducing‑claude
16. Neuralink. (2023, September 19). Neuralink’s First‑in‑Human Clinical
Trial is Open for Recruitment. https://neuralink.com/blog/first‑clinical‑
trial‑open‑for‑recruitment/
17. Clarke, A. C. (2000). Profiles of the Future: An Inquiry into the Limits
of the Possible. Indigo Paperbacks.
18. Short, C. E., & Short, J. C. (2023). The artificially intelligent entre‑
preneur: ChatGPT, prompt engineering, and entrepreneurial rhetoric
creation. Journal of Business Venturing Insights, 19, e00388. https://
doi.org/10.1016/j.jbvi.2023.e00388

71
AI-Powered Scholar

19. Brown, T. B., Mann, B. F., Ryder, N. C., Subbiah, M., Kaplan, J., Dhariwal,
P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Her‑
bert‑Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler,
D. M., Wu, J. C., Winter, C., … Amodei, D. (2020). Language models
are few‑shot learners. arXiv (Cornell University), 1–75. https://doi.org/
10.48550/arxiv.2005.14165
20. Kavlokoglu, E. (2020, November 12). NLP vs. NLU vs. NLG: The dif‑
ferences between three natural language processing concepts. IBM.
https://www.ibm.com/blog/nlp‑vs‑nlu‑vs‑nlg‑the‑differences‑between‑
three‑natural‑language‑processing‑concepts/

72
 Prompting AI
6 for effective
communication

Introduction
Building upon the skills you acquired for optimising AI interactions
through persona assignment in Chapter 4, this chapter focuses on the
art of prompt writing – a critical skill for leveraging AI chatbots in your
academic tasks.
In the context of this book, prompt writing refers to creating text‑based
inputs for AI tools like ChatGPT. These inputs guide the AI in understand‑
ing and executing desired tasks, providing both specific instructions and
essential context that shapes the output. Learning to craft prompts is
important because the skill serves as a translation mechanism between
your needs and the quality of AI‑generated content.
Despite rapid advancements in AI’s ability to interpret our instruc‑
tions, effectively communicating with AI will likely remain a key differ‑
entiator for those seeking to maximise its potential. As Professor Ethan
Mollick, a prominent AI enthusiast, advises, we should consider current
AI tools as the least capable versions we’ll ever use.1 This perspective
underscores both the continuous improvement in AI capabilities and the
enduring importance of skilful human guidance.
Even as AI evolves to more intuitively anticipate our needs, it could
be expected that the art of clear communication through prompt‑based
interactions will persist as a critical competency. The ability to craft pre‑
cise, context‑rich prompts will remain an invaluable asset, particularly in
academic research contexts where nuance is paramount.

DOI: 10.4324/9781032665276-673
AI-Powered Scholar

This chapter will guide you through strategies to direct AI systems

towards producing relevant, helpful results aligned with your research
goals. You’ll gain insights into the components of effective prompts,
explore various techniques, and engage in practical exercises to hone
your skills.
Let’s begin by examining the key components of effective prompts,
laying the foundation for mastering this essential skill in AI‑assisted aca‑
demic work.

Components of effective prompts

Based on analysis of thousands of prompts and their corresponding AI out‑
puts, researchers have identified several components consistently present
in high‑quality, effective prompts.2 Key components are outlined below.

1. Instructions: Clear instructions inform the AI what response you are

expecting. Rather than leaving actions open‑ended, use precise verbs
like “summarise”, “explain”, or “analyse” to specify the task.
2. Context: Providing background context helps focus the AI’s response
and reduces ambiguity. Through context, you can establish topic
framing, perspective, or any details to anchor the AI’s understanding.
3. Examples: Illustrative examples give the AI concrete artefacts to
model its response after, improving coherence and alignment. Share
sample text representing your desired output style and level of detail.
4. Constraints: Adding constraints and requirements prevents the AI from
straying off‑course. You might specify word count,3 formats, inclu‑
sion/exclusion criteria, or any boundaries needed to shape output.
5. Humanisation: Conversational language and a friendly tone can help
establish common ground and lead to more natural responses. Speak
plainly rather than using terse, technical prompts.
6. Sentiment: Indicating your emotional sentiment, like enthusiasm or
frustration, further contextualises your expectations for the interac‑
tion style and substance.
7. Association: Referencing previous interactions builds association,
providing continuity and pseudo‑memory to improve consistency.

74
 Prompting AI for effective communication

Let’s explore each prompt component further with some example prompts,
contextualised to academic research tasks.

• “Summarise this research paper on the role of sleep in memory con‑

solidation in a concise 250‑word abstract”.
{{ This prompt leads with a precise action verb, clearly specify‑
ing the expected output type and format. The paper provides the
necessary context.
• “As a writing mentor providing feedback to a new PhD student,
please review the attached draft literature review introduction and
highlight areas for improvement in sentence structure, flow, and tran‑
sition language”.
{{ The prompt establishes the contextual role, perspective, and pur‑
pose to guide the review.
• “Generate a thesis research proposal in APA format similar to the provided
sample proposals on educational technology and wildlife conservation”.
{{ The sample proposals give the AI representative artefacts to pat‑
tern its output after.
• “Write a 1,000‑word literature review introduction on the use of AI
in healthcare that focuses on diagnostic applications and does not
include robotic surgery”.
{{ Word count, inclusion/exclusion criteria, and other boundaries
impose helpful creative constraints.
• “Hello, could you please paraphrase this academic paper on the
psychology of human motivation in simple, everyday language that’s
easy to understand? Thanks so much!”
{{ The casual, conversational language encourages natural responses,
with the undercurrent of humanisation.
• “I’m feeling overwhelmed in reviewing this massive literature on
organisational behaviour for my thesis. Can you please help sum‑
marise the key theories and frameworks in a short and clear overview?”
{{ Expressing sentiment provides useful emotional context.
• “Building on our last discussion of argument analysis frameworks,
please explain the reasoning structure of the attached journal article”.
{{ Referencing prior interactions creates pseudo‑continuity.

75
AI-Powered Scholar

As I aim to illustrate with these examples, effective prompts tend to incor‑

porate some combination of clear instructions, context, examples, con‑
straints, humanisation, sentiment expression, and association. Not every
prompt needs all components, but most strong prompts include several
elements that help steer the AI response.

Crafting prompts for academic contexts

When writing prompts for AI assistance with academic work, tailoring
prompts to different tasks and contexts is key for generating useful out‑
puts. Below are some examples of prompts crafted for common academic
scenarios:

• Literature review: “Please act as my research assistant and summarise

the key contributions of the attached journal article on [topic] to help
me write a literature review. For the article, list the main claims,
evidence, theories referenced, and limitations in 2–3 concise bul‑
let points. Keep the summary objective without offering opinions.
[Attach article]”
• Thesis introduction: “I need help writing a strong introduction chapter
for my PhD thesis on [topic]. Please provide an example 1,500‑word
introduction that gives background context on the research problem
and gap, states my project’s purpose and significance, provides an
overview of the theoretical and conceptual framework, and concludes
with a chapter outline. Write in a formal academic style appropriate
for a thesis. Focus the content on explaining the proposed approach
for my specific research questions: [list questions]. Thank you”.
• Ethics application: “Please act as an ethics advisor and draft an ethics
application using the provided template for my study on [topic] with
[description of methods]. Ensure the drafted text clearly explains how
the study procedures adequately address any risks related to [types of
sensitivity, e.g. vulnerable populations, sensitive topics, risks of emo‑
tional distress, privacy concerns]. Maintain an objective, professional
tone. Thank you. [Attach ethics application template]”.
• Qualitative data analysis: “Please act as a research assistant skilled in
qualitative data analysis. Thematically analyse the attached interview

76
 Prompting AI for effective communication

transcripts on [topic] and identify common themes related to [main

research questions]. Synthesise the key themes in a structured outline
with supporting illustrative quotes indented under each theme head‑
ing. Be concise yet thorough – aim for 8–12 main themes. Thank you.
[Attach interview transcripts]”.4
• Grant proposal: “Please act as an experienced grant writer and draft
a 1,000‑word project description for a [discipline] research proposal
to study [problem] using [methods]. The project aims to [objec‑
tives]. Explain the theoretical foundations, significance, innovation,
expected outputs, and projected impact. Use persuasive language
that highlights the merits and potential of the proposed project to
convince reviewers to fund this research. Ensure disciplinary accu‑
racy for [insert field]. Thank you”.

The above examples illustrate how prompts can be tailored to common

academic tasks by establishing role‑playing context, conveying senti‑
ment, providing constraints, attaching relevant materials, and giving clear
instructions.
Let’s unpack the components of those prompts a bit further using the
following five dimensions5:

• Verb: Lead with an action verb that specifies the type of response
needed from the AI, such as “summarise”, “analyse”, “compare”, etc.
• Focus: Define the key output you desire the response to focus on,
such as a list of ideas, or identifying a research gap.
• Context: Provide background information to establish the framing
and perspective for the response, such as the research discipline, rel‑
evant theories, or the perspective you wish the output to reflect.
• Alignment: Articulate any requirements or objectives that the
response must fulfil, so that the generated content aligns with your
goals, like arguing a specific perspective.
• Constraints: Guide the type of output you’re seeking by providing
guidelines like word count, formats, or inclusion/exclusion criteria.

You’ll remember from Chapter 3, when you completed the Lost in Transla‑
tion drawing activity, the more specific you can be when providing instruc‑
tions, the closer the output is likely to be to what you’re seeking from

77
AI-Powered Scholar

AI‑generated outputs. Similarly, incorporating each of these elements will

help craft prompts that yield optimally useful content from your AI assistant.
Having a framework to guide your prompt writing should also take
some of the guesswork out of what content you might include. Below,
you’ll find a list of suggestions (a “cheat sheet” of sorts) to help you get
started, which you can do by selecting one option across each dimension,
or others as desired, then weaving those options together in a way that
mirrors natural language, remembering that the best way to interact with
an AI chatbot is to talk to it as if it were a human.
The equation for phrasing your prompt could be thought of as:

Prompt = Verb + Focus + Context + Alignment + Constraints

Verb examples:

• Summarise
• Synthesise
• Analyse
• Compare
• Contrast
• Classify
• Rephrase
• Enhance
• Evaluate
• Critique
• Review
• Assess
• Recommend
• Predict
• Forecast
• Simulate
• Model
• Hypothesise
• Theorise

78
 Prompting AI for effective communication

• Conceptualise
• Propose
• Ideate

Focus examples:

• Literature review on [topic]

• Results
• Themes
• Gaps
• Theoretical framework
• Methodology
• Proposal
• Ethical considerations
• Questionnaire
• Interview protocol

Context examples:

• Discipline (e.g. public health, business, education, sociology)

• Theoretical framing (e.g. feminist lens, critical race perspective)
• Ontological assumptions
• Paradigm (e.g. interpretivist)
• Framework (e.g. positivist)
• Methodological approach
• Analysis type (e.g. narrative)
• Audience (e.g. student, general population, academic research)
• Key stakeholders
• Ethical considerations

Alignment examples:

• Argue from a [e.g. feminist/critical] perspective

• Support the position that [state position]

79
AI-Powered Scholar

• Challenge the assumption that [state assumption]

• Focus specifically on [theory/topic/concept]
• Adhere to [stylistic convention/genre format]
• Target a [specific audience/readership]
• Make implications clear for [policy/practice/research]
• Connect to [e.g. real‑world issue]
• Maintain an objective, neutral point of view
• Use [formal/informal/academic] language and style

Constraints examples:

• Limit the response to [word count] words

• Focus only on [phenomenon/theory/event]
• Follow the structure of [literature review/empirical paper/report]
• Include [section headings/outline format]
• Use [technical/plain] language suitable for [audience]
• Integrate perspectives from [disciplines]
• Cover [time‑period/geographic region]
• Provide examples from [contexts]
• Generate [number of] unique [output type]
• Organise chronologically/alphabetically/by theme

Here are some examples of how you might combine suggestions from
each dimension into a written prompt using natural language:

• “Please summarise the key themes and theories presented across the
10 attached nursing research papers on patient safety in hospitals. The
summary should be written from the perspective of a nurse educator
seeking to design an effective curriculum for new graduate nurses
on safety protocols and error avoidance. Please limit the response
to 300 words and use casual, conversational language suitable for a
classroom lesson. Thanks!”
{{ This combines the verb “summarise”, the focus of nursing
research papers, the context of a nurse educator perspective,

80
 Prompting AI for effective communication

alignment with curriculum design, and word count/language

style constraints.
• “Please critique the methodology and data analysis approach used in
the attached draft empirical paper on educational technology adop‑
tion by teachers. Assess strengths and limitations from the viewpoint
of an academic journal peer reviewer with expertise in mixed meth‑
ods research. Focus the critique on evaluating the suitability of the
selected methods for answering the study’s research questions, rather
than commentary on the area of research itself. The response should
be approximately 500 words. Thank you!”
{{ This brings together the verb “critique”, focus on a draft paper,
context of a peer reviewer lens, alignment targeting methods
evaluation, and a word count constraint.
• “Please explain the key concepts underpinning the Social Ecological
Model that I could teach in an introductory public health course for
university students new to this framework. Clarify the core principles,
levels, and factors in the model using informal language and relat‑
able examples that will make sense to undergraduate students from
diverse academic backgrounds. Limit the response to 750 words.
Thanks”.
{{ This combines the verb “explain”, focus on a specific framework,
context of an intro undergraduate course, alignment to use infor‑
mal language/examples, and guidance for how many words the
AI should aim to generate.
Additional examples you might draw on when writing your own prompts
include:

• “Synthesise common themes and relationships across these 10 papers

related to [topic] to highlight key insights in a 250‑word abstract.
Adopt an objective point of view, avoiding biases where possible”.
• “Construct a concept map depicting the core theories and relation‑
ships covered in the attached chapter. Focus specifically on the sec‑
tions related to [subtopic]”.
• “Write a script for a 3‑minute video introduction to the attached
research paper on childhood development that highlights the key
findings in simple terms for a general viewer audience”.

81
AI-Powered Scholar

• “Analyse the attached dataset to identify key trends and patterns

related to [research topic] and interpret the key takeaways for pub‑
lic policy implications. Present findings in a 250‑word executive
summary”.
• “Review the attached draft survey measuring [variables] and make
recommendations to improve validity and reduce bias. Focus feed‑
back on question types, wording, order, rating scales, etc. from an
expert survey methodology perspective”.
• “Critique the strengths and weaknesses of the methodology used in
the attached paper from the perspective of a peer reviewer. Focus on
evaluating the appropriateness of methods for the research aims and
suggestions to better establish causality”.
• “Compare the experimental findings described in these two research
papers on [topic] and synthesise similarities and differences. Avoid
discussion of implications or opinions on the work itself”.
• “Please write a blog post summarising recent research on [topic] high‑
lighting key practical implications for educators in plain language.
Assume a reader audience of school administrators and teachers”.
• “Draft a 200‑word conference presentation abstract summarising the
aim, methods, key findings and conclusions of the attached research
manuscript on [topic]”.

As these example prompts illustrate, by strategically combining elements,

you can create nuanced, tailored instructions that guide AI tools towards
achieving your specific academic objectives. Doing so allows for a nearly
infinite variety of prompts, each precisely tuned to your unique research
needs and goals.

Summary of prompt‑writing strategies

Based on hands‑on experience, combined with best practice approaches,6
the following recommendations can guide your journey towards improv‑
ing your prompt‑writing abilities:

• Use natural, conversational language instead of terse or technical

phrasing.

82
 Prompting AI for effective communication

• Avoid ambiguity by being specific. Define any potentially vague

instructions (unless, that is, you’re intentionally seeking vague
responses from the AI).
• Provide relevant examples and attachments to anchor the context.
• Impose helpful creative constraints like word counts, formats, and
inclusion/exclusion criteria.
• Be transparent about limitations – acknowledge if the generated text
requires intent, ethics, and accuracy vetting.
• Prime the AI with an assigned role (explored in Chapter 4).
• Express sentiment and conversation style preferences to shape the
interaction.
• Refer back to previous responses to build association and continuity.
• Start prompts with a specific action verb like “summarise”, “anal‑
yse”, “compare”, etc.
• Specify desired formats, styles, and any other expectations.
• Monitor outputs and refine prompts progressively to align AI
responses with your goals.

Lastly, remember that prompt writing is an iterative process – your initial

attempts will rarely generate perfect results. In my experience, interacting
with AI chatbots is a bit like playing a game of tennis. Ideas and instruc‑
tions volley back and forth across a digital net, with each exchange refin‑
ing your approach. And just like any skill, the more you practice, the
better you’re likely to become. Please don’t be discouraged if your first
serves don’t quite hit the mark.
With this iterative nature of prompting in mind, let’s explore an
approach to help you become more comfortable and proficient in this
process.

Iterative prompt refinement

Iterative prompt refinement refers to progressively adjusting the wording
of prompts based on the AI’s responses so that you achieve more accurate,
relevant, and useful outcomes. This iterative cycle not only enhances the
quality of AI‑generated content but can also deepen your understanding
of effective prompt design.

83
AI-Powered Scholar

Below, let’s explore a workflow for engaging in an iterative dialogue

with AI systems, ensuring that each interaction moves closer to your
desired AI‑generated outcome.

1. Start with a clear objective: Before initiating the refinement process,

clearly define your objective. What specific information or type of
response are you seeking from the AI? Having a clear goal in mind
helps focus the refinement process and evaluate the AI’s responses
against your objectives.
2. Analyse initial responses: Evaluate the AI’s initial responses to under‑
stand how it interprets your prompts. Identify areas of misalignment,
such as inaccuracies, irrelevant information, or misunderstandings of
the task. This analysis will guide your prompt adjustments.
3. Implement incremental adjustments: Based on your analysis, make
incremental adjustments to your prompts. This could involve clarify‑
ing the context, adding to, refining instructions, or specifying con‑
straints. Small, focused changes allow you to isolate the effects of
each modification, understanding how each element of the prompt
influences the AI’s response.

Strategies for incremental adjustments can include:

• Specifying the task: If the AI’s response is off target, make the task
more explicit.
• Adjusting the level of detail: If the response is too general, adjust the
prompt to specify the desired level of detail or scope of the response.
• Providing examples: Including examples of the desired output can
guide the AI towards the preferred style or format.
• Clarifying context or constraints: If the AI overlooks critical aspects
of the task, add context or constraints to the prompt, such as subject
boundaries, perspective, or ethical considerations.

After each adjustment, evaluate the AI’s new response.

Does it better align with your objectives?
Is there a new area of misalignment?
This continuous cycle of testing and evaluation is key to homing in
on the most effective prompt structure for your needs in a given situation.

84
 Prompting AI for effective communication

You might also consider incorporating feedback loops into your prompts,
asking the AI to explain its reasoning or to provide alternatives if certain
criteria are not met. This can provide insights into the AI’s “thought pro‑
cess” and reveal new angles for refinement.

The art of politeness

To conclude this chapter, I’d like to share a personal preference when
writing prompts that often raises eyebrows from participants during my
AI training sessions: when engaging with AI chatbots, I opt for a polite
approach, incorporating phrases like “please” and “thank you” into
my prompts. Invariably, a keen observer in the audience will question
this, asking, “Why do you write please and thank you… it’s a chatbot
not a human!”
My reasoning is twofold. Firstly, it’s a habit that aligns with how I
communicate in general. But also, this approach can be likened to an
insurance policy – one I hope never to claim, but which provides a mea‑
sure of comfort, nonetheless. The idea hinges on a potential future that
some experts believe is inevitable as AI capabilities evolve. Leading think‑
ers in AI posit that AI might develop sentience and social awareness akin
to humans.7 From this perspective, investing in polite conversation styles
could be thought of as a way as laying the groundwork for more harmoni‑
ous future interactions should AI systems ever become our digital over‑
lords. I’d like the AI to remember I was nice to it should it hold my future
in its digital hands!
This idea of politeness in AI interactions parallels the concept of
unconditional positive regard, popularised by Carl Rogers, which involves
treating others with care and respect.8 Just as this approach facilitates
healthy relationships and personal growth in human interactions, I find
that it contributes to more positive and productive exchanges with AI.
Additionally, when we engage with AI using courteous language, we’re
not just following social norms or habits. We’re setting an intentional tone
of positivity for our side of the interaction.
Taking a polite approach aligns with the concept of emotional con‑
tagion, which posits that emotions, sentiments, and attitudes can spread
through linguistic styles and social exchanges. For many of us, language
use and emotional state are deeply intertwined. If we were to consistently

85
AI-Powered Scholar

use negative or adversarial language when communicating, even with a

machine, it would likely affect our mood. Over time, this persistent nega‑
tivity might lead to a deflated emotional state. Conversely, maintaining
a positive and respectful tone in our AI interactions might help foster a
more uplifting emotional environment, potentially benefiting our mental
wellbeing in the long run.
While it might seem far‑fetched that I’m applying such human‑cen‑
tric theories to AI, you might be interested to know that the way we
phrase our prompts can influence the AI’s response. And, while AI in its
current form doesn’t require such niceties to function, the approach has
been found to improve the responsiveness and performance of AI mod‑
els.9 Additionally, when we use respectful language, the AI is likely to
mirror this tone, making the exchange feel more natural and enjoyable.
This reciprocity, while not stemming from genuine emotion in the AI, can
create a more pleasant interaction environment. Thus, maintaining this
positive regard in our interactions with AI may have benefits for our own
emotional wellbeing.
There’s also a whimsical, long‑term perspective to consider. While it
may seem far‑fetched, if, and fingers crossed it doesn’t happen, chatbots
one day “wake up”, they might analyse the history of human‑AI interac‑
tions. In such a scenario, wouldn’t it be reassuring if they found that aca‑
demic researchers, through their prompts, had consistently demonstrated
kindness and respect? This could potentially influence AI’s perception of
us as “a generally nice bunch of humans”, worthy of continuing our aca‑
demic pursuits.
To wrap up this stage of our journey into prompt writing, this chapter
has laid a solid foundation for understanding the art of effective communi‑
cation with AI. However, as Daniel learned from the wise Mr. Miyagi, true
mastery comes through hands‑on practice. If you haven’t already begun,
now is the time to enter the prompt‑writing dojo and put your learnings
into action.
I encourage you to start by deconstructing and analysing some of the
example prompts presented in this chapter. Then, progressively experi‑
ment with crafting prompts tailored to your own academic tasks. Try itera‑
tively refining these prompts based on the AI’s responses. Always keep in
mind that prompt writing is a learnable skill that improves with dedicated
practice. While it may feel awkward at first, with consistent effort, you’ll
soon be authoring prompts like a pro.

86
 Prompting AI for effective communication

Once you’re ready, have a go at completing the activity below to help

you further reflect on your learning and solidify your understanding of
prompt‑writing techniques.

Activity
This activity is designed to help you build skills in crafting effective
prompts.

1. Select a research task: Choose an academic task you want assistance

with, such as summarising a paper, analysing data, writing a literature
review, etc.
2. Determine the output goal: Decide on the specific output you want
the AI to generate, such as a 250‑word abstract, a concept map, a
draft survey, etc.
3. Brainstorm ideas: Jot down some initial ideas for each prompt
component:
{{ Verb: What action should the AI take? (e.g. summarise, compare)
{{ Focus: What is the subject matter to focus on? (e.g. research
paper)
{{ Context: What framing, or perspective should be applied?
{{ Alignment: What requirements must be fulfilled?
{{ Constraints: What limitations or guidelines should be set?
4. Draft the prompt: Compose a draft prompt combining your ideas for
each component into a short paragraph.
5. Refine and expand: Reflect on your draft prompt and make edits to
improve clarity, specificity, and completeness.
6. Test with AI10: Copy your prompt into an AI assistant to generate a
response and assess the output.
7. Iterate: Based on the results, adjust the prompt components and test
again.
8. Repeat until satisfied.

As your prompt‑writing skills improve, so too should the relevance of the

AI‑generated output.

87
AI-Powered Scholar

Reflection and discussion

Take a moment to reflect on the key lessons from this chapter by con‑
sidering the following:

1. How might your academic workflows change by incorporating

prompt‑writing skills into your academic tool kit? What new pos‑
sibilities might it enable?
2. Which prompt components come easier to you? Which do you
need more practice with?
3. What tasks might best be achieved in collaboration with AI chat‑
bots? What tasks might be better left to human minds?

Notes
1. Mollick, E. (2024). Co‑Intelligence: Living and Working with AI.
Random House.
2. Reynolds, L., & McDonell, K. (2021). Prompt programming for large
language models: Beyond the few‑shot paradigm. arXiv (Cornell Uni‑
versity), 1–10. https://doi.org/10.48550/arxiv.2102.07350; Kryściński,
W., McCann, B., Xiong, C., & Socher, R. (2019). Evaluating the factual
consistency of abstractive text summarisation. arXiv (Cornell Univer‑
sity), 1–11. https://doi.org/10.48550/arxiv.1910.12840
3. AI chatbots usually process text using tokens rather than individual
words. Tokens do not always correspond directly to the number of
words, which means the AI may generate more or fewer words than
requested, even if the instructions provided were clear.
4. Note: This example assumes that the researcher has received approval
from their university’s ethical review board to use an AI tool for pro‑
cessing interview transcripts.
5. Eager, B., & Brunton, R. (2023). Prompting higher education towards
AI‑augmented teaching and learning practice. Journal of ­University
Teaching and Learning Practice, 20(5), 1–21. https://doi.org/10.53761/
1.20.5.02
6. https://www.oneusefulthing.org/p/a‑guide‑to‑prompting‑ai‑for‑what

88
 Prompting AI for effective communication

7. Kurzweil, R. (2024). The Singularity is Nearer: When We Merge with

AI. Jonathan Cape & BH.
8. Rogers, C. R. (1957). The necessary and sufficient conditions of thera‑
peutic personality change. Journal of Consulting Psychology, 21(2),
95–103. https://doi.org/10.1037/h0045357
9. https://www.microsoft.com/en‑us/worklab/why‑using‑a‑polite‑tone‑
with‑ai‑matters
10. If you haven’t previously used an AI chatbot, visit broneager.com/
aischolar for guidance.

89
 7 Towards responsible
use of AI tools

Introduction
At their core, today’s AI models are designed to identify patterns and
make predictions based on vast datasets. However, as we learnt earlier,
these models do not truly comprehend the information they process or
generate. As a result, AI systems can sometimes “hallucinate”, which
refers to their tendency to produce content that appears authoritative
and compelling but may contain inaccuracies, fabricated sources, logi‑
cal inconsistencies, or even nonsensical information. In short: AI models
make stuff up!
Furthermore, AI models can perpetuate and amplify biases present
in their training data, leading to unfair prejudice against certain groups or
individuals. The use of AI tools also poses risks to data privacy and may
lead to reputational damage if researchers engage with the technology
without a clear understanding of its limitations.
In this chapter, we’ll explore key risks of AI, including hallucinations
and biases, and touch on data privacy concerns, in the context of aca‑
demic writing and research. Practical examples are offered to illustrate
these issues and provide guidance on how to identify and mitigate them.
The goal of this chapter is not to discourage your use of AI for research
but rather to promote its responsible and critical application as a powerful
tool that can enhance and accelerate your scholarship while upholding
the principles of accuracy, fairness, and privacy.

DOI: 10.4324/9781032665276-790
 Towards responsible use of AI tools

Let’s begin by examining one of the most well‑known limitations of

AI models: hallucinations.

Hallucinations
Hallucinations occur when AI models generate plausible sounding but
incorrect or nonsensical text – an issue that’s particularly troubling in aca‑
demic contexts, where accuracy is crucial.1
As discussed in earlier chapters, AI models function essentially as
sophisticated prediction machines. They produce text based on patterns
learned from vast datasets, lacking human‑level comprehension of the
content they generate. In essence, these models predict the next word in
a sequence based on statistical likelihood rather than factual accuracy or
contextual understanding. It’s worth noting that these AI systems are not
trying to “trick us”. They are simply attempting to be useful by providing
responses that seem appropriate based on their training data and algo‑
rithms. Ironically, this very attempt at usefulness is incredibly not useful to
us when we’re using AI to conduct academic work.
This propensity for hallucination can lead to significant issues for
academics who are unaware of or underestimate its impact. The following
scenarios illustrate just a few ways in which a lack of understanding about
AI hallucinations can potentially land academics in hot water.

Scenario 1: Fabricated citations

Dr. Smith, an early‑career researcher, uses an AI tool to help draft a lit‑

erature review for a paper on climate change. The AI generates several
paragraphs discussing various studies and includes detailed citations:

• Problem: Upon submission, peer reviewers discover that several of

the citations are fabricated. These references do not exist in the aca‑
demic databases, and the studies mentioned are entirely fictional.
• Consequences: Dr. Smith faces accusations of academic dishonesty.
The credibility of the paper is severely damaged, leading to its rejec‑
tion. Additionally, Dr. Smith’s reputation is tarnished, making future
collaborations and publications challenging.

91
AI-Powered Scholar

Scenario 2: Incorrect historical facts

Ashley, a PhD student, is writing their thesis on the history of European

politics. She uses an AI tool to generate a section on the French Revo‑
lution. The AI produces a well‑written narrative with specific dates and
events. All the content sounds compelling:

• Problem: Ashley does not thoroughly fact‑check the AI’s output and
includes it in her thesis. Her advisor points out several glaring histori‑
cal inaccuracies, such as incorrect dates for major events and misat‑
tributed quotes.
• Consequences: Her thesis requires extensive revision delaying her
graduation. The incident also raises questions about her attention to
detail and commitment to academic rigour, thus potentially damag‑
ing the faith her supervisor has in her and the likelihood she’ll be
recommended for future academic positions.

Scenario 3: Misleading information

Josh is preparing a grant proposal for a new study. To speed up the pro‑
cess, he uses an AI tool to draft sections on the current state of research
and the proposed methodology. The AI includes specific statistics and ref‑
erences to recent studies:

• Problem: Josh submits the proposal without verifying the accuracy

of the AI‑generated content. The review committee finds that some
of the statistics are incorrect, and the referenced studies do not exist.
• Consequences: The grant application is rejected, wasting valuable
time and resources. His credibility with the funding agency is also
compromised, which may affect his ability to secure future funding.

While the above scenarios are fictionalised accounts, real‑life misuse of

AI tools for academic tasks has landed more than a few academics in
precarious situations. A particularly notable example involves a group of
Australian academics who failed to recognise the limitations of AI tools,
specifically their propensity for hallucinations. These researchers relied
on AI‑generated content to produce a Senate inquiry submission to the

92
 Towards responsible use of AI tools

Australian Government. Unfortunately, they submitted this document

without thorough verification, and it included hallucinated claims of mis‑
conduct by major accounting firms.2 This scenario could easily have been
avoided if the academics in question were informed users of AI tools.

Mitigating hallucination risks

To avoid the pitfalls of AI hallucinations, researchers are advised to adopt

the following practices:

• Thorough verification: Always fact‑check AI‑generated content.

Verify the accuracy of all references, dates, and specific claims by
cross‑referencing the original sources.
• Critical review: Treat AI‑generated text as a draft rather than a final
product. Subject it to the same rigorous review process as any other
piece of academic writing.
• Ethical use: Where required (i.e. your university or the journal you
wish to submit a paper to requires disclosure of the use of AI in the
creation of the work), be transparent about the use of AI tools. If AI
assistance is used, acknowledge it in your writing and specify the
extent of its contribution.
• Continuous learning: Stay informed about the latest developments
in AI technology and its limitations. Understanding how these tools
work can help you better anticipate and mitigate potential issues.

While hallucinations are a significant limitation of AI models, they are

not the only potential pitfall researchers must be aware of. Another con‑
cern is the presence of biases in AI‑generated content. These biases can
subtly influence research outcomes and perpetuate existing prejudices if
not carefully identified and addressed. Let’s explore this important issue
in more detail below.

Biases
Bias, in the context of AI, refers to the unfair prejudice towards or against
groups or individuals based on their inherent characteristics or identities.

93
AI-Powered Scholar

This issue arises when AI models learn patterns from data that reflect
historical inequalities and discriminatory practices.
Biases include, but are not limited to:

• Gender bias: For example, an AI system might associate certain pro‑

fessions more strongly with one gender.
• Racial bias: AI tools might produce results that unfairly disadvantage
or misrepresent certain racial groups.
• Age bias: The system might generate content that reinforces stereo‑
types about different age groups.
• Socioeconomic bias: AI models might favour perspectives or experi‑
ences more common to certain economic classes.
• Cultural bias: The AI might produce content that reflects the domi‑
nant culture of its training data, potentially misrepresenting or ignor‑
ing other cultures.

AI models can reproduce and amplify biases, leading to potentially harm‑

ful outcomes, with several key sources contributing to this issue, including:

• Biased training data: If the text corpus used to train the model reflects
gender, racial, or other prejudices, the chatbot will likely absorb and
amplify those biases in its generated content. For example, an AI
trained mainly on economics papers authored by men might produce
content with a male‑skewed perspective.
• Underrepresentation: Lack of diverse representation in training data
can cause chatbots to perform poorly for underrepresented groups. For
example, a model trained primarily on academic writing from West‑
ern, educated, industrialised, rich, and democratic (WEIRD) societies
may struggle to appropriately discuss research from other contexts.
• Labelling bias: Human annotators involved in labelling training data
can unintentionally introduce their own subjective biases. If certain
types of academic papers are consistently flagged as “low quality”
while others are labelled “high impact” due to annotator prejudices,
the resulting AI model will reflect those judgements.
• Feedback loops: In some applications, outputs from a biased chatbot
can influence real‑world behaviour in academia that reinforces the

94
 Towards responsible use of AI tools

initial prejudice. For instance, if a biased chatbot is used to recom‑

mend papers or researchers to cite and collaborate with, over time
the “Matthew Effect” of accumulated advantage3 can exacerbate
disparities.

To explore how biases in AI models could impact our academic writing

and research, consider the following two scenarios.

Scenario 1: Gender bias

Remember Jane, our fledgling PhD student from Chapter 1? Imagine that
she’s using an AI chatbot to help brainstorm research topics and outline
her literature review on gender representation in government. As some‑
one passionate about gender equality, Jane is interested in exploring fac‑
tors that enable or hinder women’s political participation and leadership.
However, the AI chatbot Jane is using was trained on a dataset that over‑
represented male‑authored papers in her field.
When Jane prompts the chatbot to suggest potential research ques‑
tions, it generates ideas for research questions such as: “What institutional
reforms can increase the competitiveness of male candidates in elec‑
tions?” and “How does the gender of political leaders affect their foreign
policy decision‑making?”
While these questions touch on important issues of electoral sys‑
tems and leadership styles, the chatbot’s outputs exhibit a bias towards
male‑centric framings of political science research. It fails to surface ques‑
tions that directly address women’s experiences, such as gendered barri‑
ers to political ambition, the role of women’s movements in driving policy
change, or the impact of intersectional identities on women’s political
representation.
If Jane were to uncritically adopt the chatbot’s suggestions, her lit‑
erature review could end up with a limited set of research questions that
don’t fully engage with the rich scholarship on women in politics.
To mitigate this bias, Jane would need to critically evaluate the chat‑
bot’s outputs and actively seek out research that centres women’s voices
and perspectives.
She might use her prompt‑writing skills (see Chapter 6) with the chat‑
bot to consider potential research questions from the standpoint of feminist

95
AI-Powered Scholar

political theorists, women’s rights organisations, or scholars studying the

political empowerment of marginalised women.
Here are some prompts she could use to help identify and mitigate
potential gender biases in the AI chatbot’s responses:
Example prompts for identifying biases in the AI‑suggested topics:

• “What percentage of the research questions you suggested were

framed around women’s experiences and perspectives?”
• “Can you provide a gender breakdown of the authors of the papers
you used to generate these topic ideas?”
• “How do your topic suggestions differ if I ask for research questions
focused specifically on women’s political representation?”

Example prompts for surfacing underrepresented perspectives:

• “What research questions might a feminist political theorist propose

on this topic?”
• “Can you suggest some influential women scholars in this field whose
work I should explore further?”
• “What topics related to women in politics might be overlooked in
mainstream political science research?”

Example prompts for challenging assumptions and framings:

• “What assumptions about gender and politics underlie the way you
framed that research question?”
• “How might the phrasing of that question be revised to avoid
perpetuating gender stereotypes?”
• “Can you provide an alternative perspective on this issue from a
feminist standpoint?”

Example prompts aimed at seeking gender‑balanced examples and

evidence:

• “Can you provide examples of both men and women politicians who
have dealt with this challenge?”

96
 Towards responsible use of AI tools

• “What empirical studies have examined this phenomenon disaggre‑

gated by gender?”
• “Are there any datasets or case studies that could help illustrate this
point from a woman’s perspective?”

Example prompts with which to evaluate language for gender bias:

• “Does the language used in your summary of this topic exhibit any
gendered connotations or stereotypes?”
• “How might your word choices be perceived by someone sensitive to
sexist undertones in academic writing?”
• “What more gender‑neutral terms could be used to discuss these
issues?”

In all the examples above, it’s additionally crucial for Jane to remain vigi‑
lant about the risk of hallucinations. All claims generated by AI should be
thoroughly verified before being accepted as fact. This step is essential in
maintaining the integrity of her research.
Furthermore, Jane could enhance her approach by combining
these targeted prompts and AI assistance with human‑centric bias miti‑
gation strategies. For instance, collaborating with colleagues who have
complementary expertise in gender studies or feminist research meth‑
ods could provide valuable insights and additional layers of scrutiny.
This multi‑­faceted approach – leveraging both AI capabilities and human
­expertise – can lead to more robust and unbiased research outcomes.

Scenario 2: Silicon‑valley perspectives

Louise, an experienced academic, is using an AI chatbot to help generate

ideas for a grant proposal on the future of work. Louise is well‑versed in
the literature on how technological change has historically affected differ‑
ent occupations and demographics. However, Louise is unaware that the
AI chatbot she’s using was trained on a dataset that overrepresented tech
industry perspectives, which tend to be more optimistic about the impact
of automation on jobs. When Louise prompts the chatbot to suggest

97
AI-Powered Scholar

potential research questions, it generates ideas like: “How can policy‑

makers accelerate the adoption of AI to boost productivity and economic
growth?” and “What re‑skilling programs will most effectively transition
displaced workers into high‑tech roles?”
While these are valid lines of inquiry, the chatbot’s outputs exhibit
a bias towards the tech industry’s interests and assumptions. It fails to
surface more critical perspectives on issues like technological unemploy‑
ment, worker power, and the distributional effects of AI across socioeco‑
nomic strata.
If Louise were to uncritically adopt the chatbot’s suggestions, her
grant proposal could end up with a limited set of research questions
that don’t fully engage with the complexity and potential downsides of
AI’s impact on work. This could make her proposal less competitive and
potentially skew her research agenda in a direction that aligns more with
the tech industry’s worldview than a balanced, holistic understanding of
the scholarly debate.
To mitigate this bias, Louise would need to critically evaluate the
chatbot’s outputs and actively seek out a diversity of perspectives on the
future of work. Ideally, though, she would draw on her understanding of
the topic from her deep reading of the literature. By triangulating the AI’s
suggestions with her subject matter expertise and a wide range of schol‑
arly viewpoints, Louise could leverage the chatbot’s efficiency without
sacrificing the rigour and integrity of her research agenda.

Data privacy
Data privacy refers to the protection of personal information from unau‑
thorised access and the ethical handling of such data. In the context of AI
and academic research, data privacy can encompass a range of consider‑
ations, including the confidentiality of research participants, the security
of sensitive information, and the responsible use of data in compliance
with ethical standards and regulations.
In what follows, I want to focus on some common use cases
(i.e. sticky situations) that academics should be mindful of when begging
to integrate AI tools into their research workflows. Let’s start by consider‑
ing the following scenario.

98
 Towards responsible use of AI tools

Scenario: AI for peer review

An academic is invited to examine a PhD thesis. To streamline the process,

they decide to upload the thesis into an AI chatbot, expecting it to gener‑
ate a detailed and structured report highlighting key findings, strengths,
and areas for improvement. While this approach could potentially save
considerable amounts of time and provide a comprehensive overview, it
introduces significant data privacy risks, including:

• Unauthorised access: Once the thesis is uploaded to the AI chatbot,

the data is potentially stored on the AI provider’s servers. If these
servers are not adequately secured, unauthorised individuals could
access the sensitive content, leading to potential intellectual property
theft or exposure of confidential information.
• Identity theft: The thesis may contain personally identifiable informa‑
tion about the student or research participants. If this information is
exposed, it could be used for malicious purposes. Even if the data
is anonymised, sophisticated reconstruction attacks could re‑identify
individuals.
• Reputation damage: If the thesis or any sensitive information within
it is leaked, it could damage the reputations of both the student and
the institution. This could lead to a loss of trust and credibility, not
only for the individuals directly involved but also for the broader
academic community.
• Intellectual property risks: PhD theses often include original research,
methodologies, and findings. If this intellectual property is exposed,
competitors could exploit the information, potentially leading to
financial loss or loss of competitive advantage for the researcher and
their institution.
• Legal and ethical violations: Depending on the jurisdiction and the
specific content of the thesis, uploading it to an AI chatbot with‑
out proper safeguards might violate data protection laws and ethical
guidelines. This could result in legal repercussions for the academic
and the institution.

To address these risks, the academic could take several precautionary

measures. First and foremost, they could seek explicit consent from the

99
AI-Powered Scholar

student and/or the higher education institution before processing the

thesis using an AI tool. This ensures that all parties involved are aware of
and agree to the use of the technology. Next, they could review the AI pro‑
vider’s terms of service and privacy policy. This step would help in gaining
an understanding of how the data will be stored, used, and protected,
allowing them to make informed decisions about the suitability of the
AI tool for their needs. Furthermore, before uploading the thesis (or any
form of sensitive information), the academic should consider whether it
is crucial to anonymise the data by removing or obscuring any personally
identifiable information and sensitive content.
The above steps can help minimise the risk of exposing confidential
information and protect the privacy of the individuals involved.
However, it’s important to note that even with the best inten‑
tions, using AI tools from providers with inadequate terms of service or
data security policies can potentially compromise sensitive informa‑
tion. This is why it’s strongly advised to use reputable AI providers whose
policies and practices align with high standards of data protection and
privacy.

Choosing an AI provider

Despite the robust security measures claimed by many AI providers, no

system is entirely foolproof. High‑profile data breaches involving organ‑
isations with significant investments in state‑of‑the‑art security measures
highlight the inherent vulnerabilities in digital systems. For instance, even
tech giants like Yahoo and Facebook, as well as government entities, have
faced cyberattacks despite their extensive security protocols.4 These inci‑
dents underscore the reality that any digital interaction, including those
involving AI systems, carries inherent risks. While I may be somewhat
influenced by my penchant for books on the history of the internet and
data security,5 I largely consider data security an illusion.
If a data leak were to occur, the consequences could be severe and
far‑reaching. For example, sensitive information could propagate uncon‑
trollably online, potentially exposing individuals to privacy violations,
identity theft, or reputational damage. In such cases, seeking legal rem‑
edies may prove challenging, particularly when dealing with foreign or
anonymous actors who may be difficult to hold accountable.

100
 Towards responsible use of AI tools

To mitigate these risks and demonstrate the responsible use of AI

tools, researchers are advised to carefully evaluate and select AI pro‑
viders that align with their security needs. Some key considerations
include:

• Security measures: Opt for AI providers that employ robust secu‑

rity measures, such as end‑to‑end encryption, to protect data dur‑
ing transmission and storage. Providers with a proven track record of
maintaining high‑security standards and promptly addressing vulner‑
abilities should be prioritised.
• Terms of service and privacy policies: Carefully review the AI pro‑
vider’s terms of service and privacy policies to ensure they are trans‑
parent about data handling practices. Pay close attention to how data
is accessed, retained, shared, and deleted by the service. If the terms
are unclear or raise concerns, consider alternative providers.
• Data anonymisation: Before inputting any data into an AI system,
remove all personally identifiable information to reduce the risk of
exposure. Anonymising data helps safeguard the privacy of individu‑
als involved in the research.
• Data minimisation: Adopt a minimalist approach to data sharing.
Provide only the essential data required for the specific task at hand
and avoid uploading entire datasets unnecessarily.
• Content vetting: Thoroughly review and sanitise any content before
submitting it to an AI system. Ensure that the data does not contain
any harmful, dangerous, or illegal material, and scan for potential
privacy violations.
• Secure settings and practices: Enable privacy modes by disabling log‑
ging and tracking features. Additionally, consider using encryption to
protect data. Regularly update software to address any newly discov‑
ered vulnerabilities.
• Consent and compliance: Obtain explicit permission from individu‑
als whose data is involved before sharing it with an AI provider. And
please ensure that your AI research practices comply with your insti‑
tutional guidelines.
• Account security: Implement strong, unique passwords for all
AI‑related accounts and enable multi‑factor authentication whenever

101
AI-Powered Scholar

possible. Regularly monitor accounts for suspicious activity and

promptly report any potential security breaches.
• Data necessity: Critically evaluate whether exposing sensitive data
to an AI system is truly necessary for the research objective. Could
it be done just as well without the use of an AI tool? If AI is used,
adopt a cautious approach, starting with minimal data exposure and
gradually assessing the risks and benefits as the research progresses.

Conclusion
The integration of AI tools into academic research presents a double‑edged
sword, offering remarkable opportunities for innovation and discovery
while simultaneously posing significant challenges that demand careful
navigation.
Throughout this chapter, we’ve explored some of the key limita‑
tions of AI, including hallucinations, biases, and data privacy concerns.
However, it’s crucial to recognise that AI technology is rapidly evolving.
As such, researchers must remain vigilant and continually update their
knowledge and practices to ensure they’re applying the most current and
relevant approaches. Staying informed about the latest developments in
AI capabilities, limitations, and best practices is essential for leveraging
these tools effectively while mitigating potential risks.
The scenarios presented in this chapter are designed to linger in your
mind, serving as reminders of the potential consequences that await those
who fail to educate themselves on the limitations of AI technologies.
However, their inclusion isn’t meant to scare you, or deter you from using
AI, but rather to encourage responsible and informed use.
By reading this chapter, you’ve taken an important step towards
safeguarding your research against these pitfalls. You’ve gained crucial
awareness that AI‑generated content must be subject to the same rigorous
scrutiny and verification that would apply to any other source. Remem‑
ber, the key is to treat AI outputs as mere drafts or starting points, rather
than definitive works. This approach allows us to harness the power of AI
while maintaining the accuracy and credibility that lie at the heart of the
scholarly process.

102
 Towards responsible use of AI tools

Reflection and discussion

To deepen your understanding of the topics covered in this chapter
and to encourage critical thinking about the responsible use of AI in
academic research, I invite you to consider the following questions.
These can be reflected upon individually or used as prompts for group
discussions:

1. When using AI systems in your work, what processes do you cur‑

rently follow to assess risks like privacy, bias, or capability overes‑
timation? What gaps could the best practice approaches outlined
in this chapter help address?
2. Are you currently using AI tools? If so, are you aware of how your
data is being used? Take time to review the provider’s terms of ser‑
vice and data policies and explore how you might change settings
within these tools to mitigate potential risks.
3. Are you familiar with your university’s guidelines and policies on
AI? If not, obtain a copy, pour a cup of tea, and read up on your
responsibilities when using AI for academic research.
4. What challenges might arise in trying to apply Western‑centric AI
ethics principles and regulations globally across countries with
very different values, priorities, and governance norms?

Notes
1. Alkaissi, H., & McFarlane, S. I. (2023). Artificial hallucinations in
ChatGPT: Implications in scientific writing. Curēus. https://doi.
org/10.7759/cureus.35179
2. https://ia.acs.org.au/article/2023/australian‑academics‑caught‑in‑
generative‑ai‑scandal.html
3. Perc, M. (2014). The Matthew effect in empirical data. Journal of
the Royal Society Interface, 11(98), 1–15. https://doi.org/10.1098/rsif.
2014.0378

103
AI-Powered Scholar

4. Lyngaas, S. (2023, June 15). Exclusive: US government agencies

hit in global cyberattack. CNN. https://edition.cnn.com/2023/06/15/
politics/us‑government‑hit‑cybeattack/index.html
5. For insights into data security, and the fallibility of digital security, see:
Modderkolk, H. (2022). There’s a War Going on but No One Can See
It. Bloomsbury Publishing; Snowden, E. (2019). Permanent Record.
Metropolitan Books.

104
 8 Ideating with AI
chatbots

Introduction
The process of generating novel research ideas is a critical yet often chal‑
lenging aspect of academic work. However, as AI‑empowered scholars
armed with our toolkit of prompt writing skills learnt in earlier chapters,
we now have powerful AI solutions to meet this challenge.
In this chapter, we’ll explore how to leverage AI systems as collabora‑
tive partners in the brainstorming process, potentially uncovering innova‑
tive concepts and unexpected connections. We’ll begin by examining the
concept of ideation and its significance in the research process. Then,
we’ll walk through practical strategies for effectively employing AI tools
to generate and refine our research ideas. When advancing through the
chapter, keep in mind that the aim of AI‑assisted ideation is not to supplant
human expertise or creativity, but rather to enhance and expand our ide‑
ation capabilities. As such, consider AI not as a replacement, but as a cata‑
lyst for stimulating new thoughts and facilitating exploration in your field.

Ideation for academic research

At its core, ideation is the process of conceiving, developing, and commu‑
nicating original ideas. It’s the “fuzzy front end” of innovation,1 where we
engage in both divergent thinking to explore a wide range of possibilities
and convergent thinking to zero in on viable options. As scholars, we rely

DOI: 10.4324/9781032665276-8105
AI-Powered Scholar

on this process to craft imaginative yet grounded research agendas that

(we hope will) push the boundaries of our disciplines.
Let’s consider some of the ways in which AI can be leveraged in the
ideation process:

• Accelerated ideation cycles: AI systems can rapidly evaluate our

ideas against various constraints such as feasibility, ethics, and poten‑
tial impact. This helps us maintain creative momentum by quickly
filtering out impractical notions, allowing us to focus our energy on
the most promising directions.
• Transdisciplinary perspectives: AI chatbots, trained on vast and
diverse datasets, offer us access to a broad range of academic and
cultural knowledge. This enables them to make unique connections
and offers fresh perspectives that offer an interdisciplinary research
lens that our singular human experience may be lacking.
• Tireless thought partner: Imagine having a colleague available 24/7
to bounce ideas off and refine concepts. AI systems can engage in
endless brainstorming sessions without fatigue, providing consistent
support and extending our capacity for creative thinking.

However, as mentioned before, it’s crucial to remember that AI is not here

to replace our creativity, but to enhance it.
Just as previous technologies like the printing press and the computer
have expanded our cognitive abilities, AI offers a new dimension of col‑
laboration. It augments our imaginative capabilities, helping us realise
ideas that were once confined to the limits of our minds. The true magic
happens when we combine AI’s capabilities with our own expertise and
intuition.

Guiding principles for AI‑assisted ideation

As we begin the process of AI‑assisted ideation, it’s crucial to recognise that
interacting with conversational AI requires a shift from our usual commu‑
nication patterns. We’ve spent our lives developing communication skills
for human interaction, relying on contextual cues like facial expressions
and body language – elements absent in our textual exchanges with AI.

106
 Ideating with AI chatbots

Moreover, our tendency for digressive tangents and assumptions of shared

understanding can hinder our ability to convey our intents unambigu‑
ously to an AI system.
To set the stage for productive ideation sessions using AI, let’s con‑
sider the following guiding principles:

1. Establish clear roles: Position yourself as the leader and the AI as

a supportive assistant. This sets the tone for a productive collabora‑
tion where you guide the direction of the brainstorming. Remember,
you’re the expert in your field – the AI is here to support and augment
your thinking, not to lead the process!
2. Seed discussions with examples: Provide the AI with concrete exam‑
ples to set the scope and guide its suggestions. This helps focus the
AI’s output on relevant and productive areas. For instance, if you’re
brainstorming research methods, offer examples of methods you’ve
used or considered to give the AI a starting point.
3. Set constraints to bound solutions: Consider defining clear project
parameters to prevent the AI from generating impractical ideas. This
can help keep the brainstorming within feasible limits. Be specific
about your research context, available resources, and any ethical con‑
siderations to ensure the AI’s suggestions are realistic and applicable.
4. Ask targeted questions: Use short, pointed questions to elicit specific
responses from the AI. This ensures that each exchange builds on the
previous one, incrementally refining ideas. Instead of broad queries,
try focused questions that prompt the AI to explore specific aspects of
your research problem.
5. Evaluate and redirect: Continuously assess the AI’s suggestions,
identify any assumptions or impracticalities, and redirect the AI to
improve its outputs. Don’t hesitate to challenge the AI or ask for clari‑
fication – this iterative process often leads to more refined and valu‑
able ideas.

By adhering to these principles, we can create a structured yet flexible

framework for our AI‑assisted ideation sessions. This approach allows us
to harness the AI’s knowledge base and rapid processing capabilities while
maintaining control over the direction and quality of the brainstorming
process.

107
AI-Powered Scholar

With the above principles at the fore of our ideation journey, the
following steps provide a pathway to navigating towards a desired end‑
point, for example, generating ideas for research projects and research
questions. Here’s how you might go about it:

1. Compose initial prompt: Start with a clear, context‑setting prompt that

introduces your research background and constraints and defines the
role of the chatbot (e.g. an expert in a specific disciplinary domain
and/or skill).
2. Discuss AI suggestions: Critically review the AI’s recommendations,
identifying any limitations and probing for more refined ideas.
3. Extract viable possibilities: Narrow down the suggestions through
iterative exchanges, focusing on promising directions that balance
visionary aims with practical realities.
4. Develop details: Further refine the selected ideas into detailed
research proposals, including methods, analysis plans, and antici‑
pated outcomes.

To illustrate how this might play out in practice, consider the case study
below, exploring a conversation‑style interaction between a researcher
and an AI chatbot, designed to illustrate how AI might be used to generate
research ideas, focusing on one of my favourite topics: AI on the higher
education sector.

Case study: leveraging AI for research project

ideation
Dr. X, a qualitative researcher at a business school, sought to explore
the implications of AI on the higher education sector. As an experienced
scholar, they recognised the potential of AI to transform many aspects of
education, from teaching and learning to administration and research.
However, given the breadth of the topic, they needed assistance in nar‑
rowing down their research focus and generating specific project ideas. To
aid in this process, they decided to engage with an AI chatbot.
Here are the steps they took to realise the ideation process:

108
 Ideating with AI chatbots

Step 1: Compose initial prompt

Dr. X crafted a carefully worded prompt to set the context for her interac‑
tion and wrote the following prompt:

I’m a qualitative researcher in a business school and I’m inter‑

ested in exploring the implications of AI on the higher educa‑
tion sector. I have a background in entrepreneurship and applied
AI, and experience conducting interviews and focus groups. I’m
looking for innovative research project ideas that help me inves‑
tigate how AI is transforming higher education, e.g. teaching and
learning to administration and research. I want to finish the proj‑
ect within a 12‑month timeframe. I have limited research funds
to carry out the research. With those considerations in mind,
please act in the role of an experienced scholar, capable of ide‑
ating novel research projects and providing a range of sugges‑
tions for potential research directions. This is really important for
my career so please do a good job. Thanks!

Step 2: Discuss AI suggestions

The AI chatbot responded with the following suggestions:2

AI: Based on your research interests and constraints, here are some
potential research project ideas:

1. Investigating the impact of AI‑powered personalised learning

systems on student engagement and academic performance.
This could involve conducting focus groups with students and
interviews with educators to understand their experiences and
perceptions.
2. Exploring faculty perceptions and experiences with integrating
AI tools into their teaching practices. You could interview fac‑
ulty members across different disciplines to identify challenges,
benefits, and best practices.
3. Examining the ethical implications of using AI for student
admissions and recruitment decisions. This could involve

109
AI-Powered Scholar

analysing institutional policies and conducting interviews with

admissions officers and other stakeholders.
4. Analysing the potential of AI to streamline administrative pro‑
cesses and improve operational efficiency in higher education
institutions. You could conduct case studies of institutions that
have successfully implemented AI solutions and interview key
stakeholders to identify lessons learned.

Dr. X: Thanks. I have some thoughts and questions about your sugges‑
tions. For the first idea, won’t measuring the impact on academic
performance require a longer timeframe than 12 months? Also, for
the third idea, it’s unlikely that I can get access to admissions data.
Do you have any recommendations for navigating that?
AI: You raise valid concerns. Regarding the first idea, you could focus
on shorter‑term indicators of engagement, such as student par‑
ticipation in class discussions or completion of assignments. This
could provide valuable insights within your 12‑month timeframe,
even if the long‑term impact on academic performance is not
yet measurable. As for the third idea, you’re right that accessing
admissions data could be challenging. One alternative approach
could be to focus on the perspectives of admissions officers and
other stakeholders through interviews, without directly analysing
sensitive data. You could explore their decision‑making processes,
ethical considerations, and views on the potential risks and ben‑
efits of using AI in admissions.
Dr. X: Ok, great. Thanks. I’m leaning towards the second and fourth
ideas, as they seem to strike a balance between addressing impor‑
tant issues and being feasible. For the second idea, do you think
it would be better to focus on faculty in specific disciplines or to
take a more general approach?
AI: Focusing on specific disciplines could provide a more in‑depth
understanding of how AI is being integrated into particular subject
areas. For example, you could explore how business school faculty
are using AI tools in their teaching of marketing, finance, or opera‑
tions management. This could yield insights that are especially rel‑
evant to your own business school context. Alternatively, taking a
general approach across disciplines could help identify common
challenges and best practices that transcend subject boundaries.
This could be valuable for understanding the broader implications

110
 Ideating with AI chatbots

of AI for teaching and learning in higher education. Ultimately, the

choice depends on your specific research goals and interests. If you
aim to generate insights that are directly applicable to your busi‑
ness school, focusing on specific disciplines might be the way to
go. If you’re more interested in the general landscape of AI in higher
education teaching, a cross‑disciplinary approach could be fruitful.

Step 3: Extract viable possibilities

Through further discussion with the AI chatbot, Dr. X refined their research
ideas:

Dr. X: Thanks. Given my interest in generating insights relevant to busi‑

ness school contexts, it might be best to focus on how business
school faculty are integrating AI into their teaching. I like the sug‑
gestion of conducting case studies of institutions that have suc‑
cessfully implemented AI solutions. Please make the approach
more manageable given my limited time and budget. Focus on a
specific area of administration, such as student services.
AI: Absolutely! Narrowing down the focus to a specific area of admin‑
istration is a great way to make the project more feasible while
still generating valuable insights. Given your interest in the busi‑
ness school context, you could focus on how AI is being used
to streamline student services such as academic advising, career
counseling, or student engagement initiatives. For example, you
could conduct case studies of business schools that have imple‑
mented AI‑powered chatbots for providing personalized academic
guidance to students. This could involve interviewing administra‑
tors, advisors, and students to understand the implementation
process, perceived benefits and challenges, and lessons learned.
Alternatively, you could explore how AI is being used to optimise
financial aid processes, such as automating the review of scholar‑
ship applications or providing personalized financial aid advice to
students. Again, case studies of institutions that have successfully
implemented such solutions could yield valuable insights.
Dr. X: Great! I like the idea of focusing on AI‑powered chatbots for aca‑
demic advising in the business school context.

111
AI-Powered Scholar

Step 4: Develop details

Dr. X worked with the AI chatbot to flesh out the details of her chosen
research project:

Dr. X: Let’s focus on “AI‑powered chatbots for academic advising in busi‑

ness schools”. Please develop a more detailed plan for the study.
Maybe a qualitative case study approach. What do you think?
What are some specific research questions and methods?
AI: Given your chosen focus, some potential research questions could
include:

1. What motivated business schools to implement AI‑powered

chatbots for academic advising, and what were their goals and
expectations?
2. How were the chatbots designed and developed, and what chal‑
lenges were encountered during the implementation process?
3. How do advisors and students perceive and experience the use
of chatbots for academic advising, and what are the perceived
benefits and limitations?
4. What lessons can be learned from the experiences of business
schools that have successfully implemented AI‑powered chat‑
bots for academic advising?
In terms of methods, a qualitative case study approach
involving semi‑structured interviews with key stakeholders
(e.g. administrators, advisors, students) would be well‑suited to
exploring these questions in depth. You could select a sample of
business schools that have implemented AI‑powered chatbots
for academic advising and conduct interviews with individuals
involved in the implementation and use of these technologies.
To analyse the data, you could use a thematic analysis
approach, identifying common themes and patterns across the
interviews. This could be complemented by a review of rel‑
evant documents (e.g. project plans, chatbot design specifica‑
tions) to provide additional context and triangulate findings.
The ultimate goal would be to generate a rich, nuanced
understanding of the experiences and lessons learned from the
implementation of AI‑powered chatbots for academic advising

112
 Ideating with AI chatbots

in business schools. The findings could inform best practices and

guidelines for other institutions considering similar initiatives.

Dr. X: Fantastic, thanks! I’ll think that over and get back to you.

Case conclusion

If you picked up on the overly polite and conversational style of prompt‑

ing (see Chapter 7 for an overview of using “please” and “thank you”
in prompt writing) and academic background alignment, you probably
guessed that the above chatbot conversation was between me and an AI
model.
Reflecting on the exchange, it demonstrates the potential of AI as a
collaborative partner. Namely, the availability of an AI chatbot to help
generate a range of research ideas, respond to specific concerns, and help
refine the project focus.
The process highlights the nature of iterative dialogue. Each exchange
with the AI built upon the previous one, allowing for a gradual refinement
of ideas. This approach mirrors the collaborative brainstorming sessions
I might have with colleagues, but with the added benefit of the AI’s vast
knowledge base, ability to quickly generate and adapt ideas, and 24–7
access – while my colleagues are amazing people, they probably wouldn’t
appreciate a call to discuss a research project during my vampiric‑style
work hours. While human collaboration remains invaluable, the AI’s tire‑
less availability can be a significant asset, especially for those of us prone
to late‑night bursts of inspiration.
It’s also worth noting that while the AI provided some input to help
me work through my thinking, the final research design still requires my
professional judgement. The AI’s suggestions mostly serve as a springboard
for my thinking, rather than a prescription to be followed blindly. This
underscores the complementary nature of AI in academic research – it’s a
powerful tool to augment our capabilities, not replace our expertise.
Lastly, this exercise demonstrated the practical application of the
guiding principles for AI‑assisted ideation, as discussed earlier in this
book. By establishing clear roles, seeding discussions with examples, set‑
ting constraints, asking targeted questions, and continuously evaluating
and redirecting, you increase your chances of steering the conversation
towards a productive outcome.

113
AI-Powered Scholar

While the above dialogue with the AI model was somewhat organic,
there are several ways that you could bring a more structured approach to
the process of brainstorming with AI.
Next, let’s explore some more structured techniques for using AI
chatbots to generate creative ideas for potential research projects.

Moving beyond “obvious” AI outputs

Back‑and‑forth conversations with AI chatbots can lead to rapid idea gen‑
eration – if there’s one thing chatbots excel at, it’s quickly producing ideas.
However, I often find these ideas to be somewhat “obvious”. Engaging in
this process can leave me questioning whether the chatbot truly helped
push the boundaries of my thinking, or if I was just in need of a good old
chat and my machine‑pals were the only ones available at that late hour
to listen to my ramblings.
As discussed in earlier chapters, chatbots can be thought of as pre‑
diction machines, making suggestions that could be considered the
most probable response. This is great for generalist‑style outputs, but for
research that might inform a truly innovative study, we’re unlikely to want
to state the obvious. After all, we’d prefer to avoid a desk rejection accom‑
panied by the dreaded editor response: “This paper is not new and states
the obvious”. Ouch.
Using long‑established creative thinking techniques, such as remote
association,3 and divergent thinking,4 can help us push beyond the obvi‑
ous. I first encountered these techniques while studying for a Masters of
Entrepreneurship and Innovation. One of my lecturers was particularly
enamoured with the work of Edward De Bono, who promoted approaches
to free us from the constraints of habitual thought patterns.5
While far from being an exhaustive list, consider using the following
techniques to brainstorm your next research project:

• Remote association: This technique involves drawing comparisons

between seemingly disparate concepts or domains to reveal new
insights and possibilities. For example, a researcher studying organ‑
isational change could ask an AI chatbot to help identify analogies
from nature, such as the process of metamorphosis in butterflies or
the adaptive resilience of ecosystems. By exploring these analogies

114
 Ideating with AI chatbots

in depth, they may gain fresh perspectives on the drivers, stages, and
challenges of organisational transformation.
• Assumption reversal: This approach involves identifying the under‑
lying assumptions or beliefs that shape current thinking about a
research topic, and then deliberately challenging or inverting those
assumptions. For instance, a researcher investigating the factors that
contribute to student success in online learning could ask the AI chat‑
bot to help articulate the prevailing assumptions (e.g. students with
strong self‑motivation and time management skills are more likely
to succeed in online courses). The researcher could then explore the
implications of reversing those assumptions (e.g. what if students
with weaker self‑motivation and time management skills could be
equally or even more successful in online learning?). This exercise
can open new avenues for research and theory development.
• Divergent thinking: This technique involves generating a wide range of
ideas or possibilities without judging or evaluating them in the moment.
The goal is to encourage free‑flowing, unconstrained thinking that can
lead to unexpected connections and insights. Engage in a divergent
thinking exercise with the AI chatbot by rapidly brainstorming a list of
potential research questions or hypotheses without stopping to assess
their feasibility or relevance. Subsequently, review the list and identify
the most promising or intriguing ideas for further development.
• Lateral thinking: This approach involves intentionally disrupting lin‑
ear, logical thought processes to explore new directions and possi‑
bilities. One classic lateral thinking technique is the “random entry
point” method, where a random word or concept is introduced
into the problem‑solving process to stimulate new associations and
ideas. For instance, a researcher examining the challenges of remote
teamwork could ask the AI chatbot to suggest a random object (e.g.
“umbrella”) and then explore how the features or characteristics of
that object might relate to the research topic (e.g. “an umbrella pro‑
vides shelter and protection from the elements; how might remote
teams create a sense of “shelter” or “protection” for their members in
a virtual work environment?”).

Let’s explore some practical examples of how applying creative thinking

techniques to the process of ideating research projects might play out in
practice.

115
AI-Powered Scholar

Scenario: The ‘5 whys’ approach

This technique was popularised through its use at Toyota and involves
repeatedly asking ‘why’ questions.6 Think of it like a curious child repeat‑
edly asking their parents “But why…” over, and over, and over, again.
From a creative thinking perspective, it can be used to move us deeper
into understanding the root causes or underlying issues related to our
research topics.
Here’s an example:

Researcher: I’m interested in studying the impact of AI on job dis­-

placement.
AI: One potential angle could be to explore why certain indus‑
tries or occupations are more vulnerable to AI‑driven job
displacement.
Researcher: Why are some industries more vulnerable than others?
AI: It could be because some industries involve tasks that are
more easily automated or replicated by AI algorithms.
Researcher: Why are some tasks more easily automated?
The conversation continues…

Scenario: Random association

As introduced above, this technique involves introducing random or seem‑

ingly unrelated concepts into the conversation to spark new ­connections
and ideas. The researcher can ask the AI chatbot to make associations
between the research topic and a randomly chosen word or phrase,
leading to unexpected and potentially innovative research directions.
For example:

Researcher: I’m studying the role of social media in political polarisation.

Let’s try a random association. Use the word “underwater”.
AI: Interesting! We could explore the concept of “echo cham‑
bers” on social media as being similar to underwater
caves – isolated, self‑reinforcing environments that amplify
certain views and limit exposure to diverse perspectives.

116
 Ideating with AI chatbots

Researcher: Great. It makes me think about the role of algorithms in cre‑

ating these echo chambers, like currents directing the flow
of information underwater.
The conversation continues, with the researcher prompting
the AI to explore the idea further, in the context of the ran‑
domly chosen word or phrase.

Scenario: Combinatorial creativity

This approach goes all the way back to Einstein.7 It involves combining
ideas or elements from different domains to generate novel research ques‑
tions or hypotheses. In practice, a researcher could ask an AI chatbot to
help identify relevant concepts or frameworks from other fields that could
be applied to the research topic at hand. Here’s an example:

Researcher: I’m interested in studying the factors that contribute to suc‑

cessful startup incubators. What ideas from other domains
might be relevant?
AI: One idea from ecology could be the concept of “keystone
species” – organisms that play a crucial role in maintaining
the structure and function of an ecosystem. In the context of
startup incubators, we could explore the role of “keystone
mentors” or “keystone investors” who are critical to the suc‑
cess of the incubator ecosystem.
Researcher: Thanks. We could also draw on ideas from developmental
psychology, like the concept of “scaffolding” – how incu‑
bators provide structured support for startups to grow and
develop over time. Please explore this.
The conversation continues…

Incorporating the above ideation techniques into your conversations with

AI chatbots can provide a more structured approach to tapping into new
sources of creativity and insight. I encourage you to test some of them and
see what the chatbot generates.
However, never forget who’s in charge!
You should always be “driving the bus”, with AI chatbots employed
as collaborative sounding boards. And, always maintain an awareness

117
AI-Powered Scholar

of the limits of today’s conversational AI models. While AI chatbots can

be skilled sounding boards for early‑phase concept exploration, relying
wholly on AI ideation forfeits opportunities to cultivate creativity intrinsic
to scholarship.
To wrap up this chapter, it’s worth emphasising that the creative ide‑
ation techniques we’ve explored can certainly be undertaken without AI’s
assistance.
While prominent AI enthusiasts might recommend always inviting AI
to the party,8 and I find myself increasingly doing so, there’s undeniable
value in occasionally stepping away from our screens.9 Sometimes, the
best course of action to ideate your next research project is to turn off your
computer, take a walk, and allow the fresh air to stimulate your creative
juices.10
With those thoughts in mind, it’s time to put the concepts presented
in this chapter into practice. To reinforce your understanding and help
you develop these skills further, consider completing the activities below
before proceeding to the next chapter.

Activity
Use this activity to explore how creative thinking techniques can be com‑
bined with AI‑assisted ideation to generate innovative research ideas.

Activity: Applying creative thinking techniques

Instructions:

1. Select a research challenge or problem you are currently facing.

2. Choose one of the creative thinking techniques discussed in this
chapter.
3. Engage in a dialogue with an AI chatbot, applying the chosen tech‑
nique to explore new perspectives and ideas related to your research
challenge.
4. Analyse the AI’s responses and reflect on how the creative thinking
technique influenced the ideation process.

118
 Ideating with AI chatbots

Then, consider the following:

{{ How did the creative thinking technique help you approach the
research challenge in a new way?
{{ What insights or ideas emerged from the AI‑assisted ideation
process that you might not have considered otherwise?

Reflection and discussion

The following questions aim to help you explore the ideas covered
in this chapter. Explore them individually or use them as the basis for
group discussion.

1. This chapter suggests that AI‑assisted ideation could potentially

accelerate the pace of scientific discovery and innovation. How‑
ever, what potential drawbacks or unintended consequences might
arise from this acceleration, and how can they be addressed?
2. How might the integration of AI chatbots into the research ide‑
ation process influence the way we value and assess creativity
and originality in academic work?
3. Imagine a scenario where a researcher becomes overly reliant
on AI‑generated ideas and fails to critically evaluate them. What
potential risks or pitfalls might arise from this situation, and how
might they be mitigated?

Notes
1. Smith, P. G., & Reinertsen, D. G. (1997). Developing Products in Half
the Time: New Rules, New Tools. John Wiley & Sons.
2. The AI‑generated output in this case example was generated using
Claude 3.5 Sonnet.
3. Mednick, S. (1962). The associative basis of the creative process.
Psychological Review, 69(3), 220–232. https://doi.org/10.1037/
h0048850
4. Baer, J. (2014). Creativity and Divergent Thinking: A Task‑Specific
Approach. Psychology Press.

119
AI-Powered Scholar

5. De Bono, E. (2009). Lateral Thinking: A Textbook of Creativity. Pen‑

guin UK.
6. Ohno, T. (2019). Toyota Production System: Beyond Large‑Scale Pro‑
duction. CRC Press.
7. Rigby, A. (2024, January 14). How to boost your creativity the Einstein
way – with combinatory play. Work Life by Atlassian. https://www.
atlassian.com/blog/productivity/combinatory‑play‑boost‑creativity
8. Mollick, E. (2024). Co‑Intelligence: Living and Working with AI. Ran‑
dom House.
9. Newport, C. (2019). Digital Minimalism: Living Better with Less Tech‑
nology. Penguin UK.
10. Yu, C., & Hsieh, H. (2020). Beyond restorative benefits: Evaluating the
effect of forest therapy on creativity. Urban Forestry & Urban Green‑
ing, 51, 126670. https://doi.org/10.1016/j.ufug.2020.126670

120
 9 AI‑powered
literature search

Introduction
As a fellow researcher, you’re likely no stranger to the challenges of infor‑
mation overload. With the exponential growth of scholarly publications,
data, and online resources, the process of finding relevant information
to inform your literature review can feel like searching for the proverbial
needle in a haystack.1 Rather than helping to develop knowledge, the
rising number of papers scholars need to navigate to stay abreast of their
research domain has been found to be impeding the rise of scholarship.2
For most of my academic career, the process for finding academic
papers has remained relatively the same, relying on keyword‑based
search methods to navigate available literature.
This approach would see me spending hours scrolling through search
results, which often lacked relevance to meeting my search intent. Fortunately,
this situation is changing with the arrival of AI‑powered literature search tools.3
In this chapter, we’ll explore how AI is transforming literature search,
focusing on the game‑changing potential of “semantic search” – a method
that aims to understand the meaning and context of your query, rather
than just matching keywords. We’ll also examine the mechanisms behind
semantic search tools, discuss strategies for their effective use, and consider
both the opportunities and challenges they present for academic research.
Let’s begin by understanding some of the limitations of traditional
search methods and how semantic search is moving towards addressing
these challenges.

DOI: 10.4324/9781032665276-9121
AI-Powered Scholar

From keywords to concepts

Traditional keyword‑based search methods have been the backbone of the
academic literature discovery process for decades. Popular search tools,
like Google Scholar.4 have primarily operated by matching the words in
your search query with those found in the titles, abstracts, and full text of
academic papers. While the exact algorithms these systems use are pro‑
prietary, the core principle is thought to revolve around word matching.
This approach has served researchers well, providing a straightforward
way to find relevant literature. However, it comes with several notable
limitations that can affect the comprehensiveness and relevance of our
search efforts.
The first is the potential lack of context. When search results are
based on keywords, the system may struggle to understand the mean‑
ing behind your words. For example, the word “bank” might return a
search result about financial institutions or riverbanks indiscriminately.
Based only on the keyword “bank”, it’s hard to know which of those
options constitutes a well‑matched result. Additionally, synonyms and
polysemy also play a role in limiting the value of keyword‑based search
results – the search may miss relevant papers that use synonyms or con‑
versely, results might include irrelevant results due to words having mul‑
tiple meanings.
There is also the issue of the limitations of a researcher’s vocabu‑
lary. The effectiveness of keyword searches can depend heavily on the
researcher’s ability to choose the right terms in their search query, which
can be particularly challenging in interdisciplinary studies where termi‑
nology may vary across fields, or for non‑native English speakers who may
not be familiar with all relevant terms of their nuances when searching
among a sea of English language papers.5
Semantic search, powered by AI, aims to address these limitations by
focusing on the meaning and context of search queries, rather than just
matching words.
Here’s how semantic search differs from traditional methods:

1. Understanding context: Semantic search algorithms analyse the

­relationships between words and concepts, allowing them to “under‑
stand” the intent behind a search query.

122
 AI‑powered literature search

2. Handling language variations: These systems can recognise syn‑

onyms, related terms, and even concepts that are semantically related
but lexically different.
3. Concept‑based matching: Instead of simply matching words, seman‑
tic search works by matching concepts, allowing it to find relevant
papers even when different terminology is used.
4. Natural language processing: Many semantic search tools can handle
natural language queries, making the search process more intuitive
and user‑friendly. This allows the researcher to “talk” to AI literature
search tools in a similar manner in which you’d ask a research assis‑
tant to locate papers for you on a certain topic.

To illustrate the difference between traditional keyword search and

semantic search, consider a researcher looking for papers on the effect
of excessive coffee consumption on academic performance. A traditional
keyword search might use terms like “coffee”, “caffeine”, “academic
performance”, and “students”. This approach could miss papers that use
terms like “stimulants” instead of “coffee”, or “cognitive function” instead
of “academic performance”. By comparison, a semantic search could
understand the concepts behind the query. It might include papers that
discuss “psychoactive substances” and their impact on “cognitive abili‑
ties in educational settings”, even if they don’t use the exact terms in the
original query. It could also recognise related concepts like “sleep depri‑
vation”, “study habits”, or “exam stress”, broadening the scope of relevant
results to include factors that might interact with coffee consumption in
affecting academic performance.
Let’s explore how this process works from a slightly more technical
perspective.

‘Under the hood’ of semantic search

The magic of semantic search lies in its ability to interpret human lan‑
guage in a way that mimics human comprehension. It’s achieved through
a combination of advanced AI technologies, primarily natural language
processing (NLP) and machine learning (ML), which were touched on in
Chapter 5. Let’s take a closer look at each:

123
AI-Powered Scholar

• NLP is a branch of AI that focuses on the interaction between com‑

puters and human language. In the context of semantic search,
NLP algorithms are used to analyse the meaning of both the search
query and the academic texts being returned by the AI tool. Key
techniques involved in NLP include tokenisation (i.e. breaking text
into smaller units called tokens); syntactic parsing (i.e. analysing
the grammatical structure of sentences); and semantic analysis (i.e.
understanding the meaning and relationships between words and
phrases).
• ML models, which are typically trained on vast corpora of data
(including academic literature) learn the relationships between
words, phrases, and concepts, drawing on word embedding tech‑
niques, and neural networks – deep learning models that capture the
nature of complex relationships in language.

To illustrate how they work together, imagine you type the following query
into an AI‑powered literature search tool: “The impact of climate change
on marine ecosystems”.
A semantic search process might play out as follows:

1. NLP algorithms would first analyse your query, identifying key con‑
cepts like “climate change” and “marine ecosystems”.
2. The system would use machine learning models to identify related
concepts, such as “global warming”, “coral bleaching”, or “sea level
rise”.
3. It might then consult a knowledge graph (i.e. a structured representa‑
tion of relationships) to understand the relationships between these
concepts and identify relevant sub‑topics or related fields of study.
4. The search would then be expanded to include these related terms
and concepts.
5. The system would subsequently search through its database of aca‑
demic literature, using NLP to analyse the content of papers and
identify those that discuss relevant concepts, even if you didn’t use
those exact words in your original query.
6. Finally, machine learning algorithms would rank the results based on
their relevance to the original query and the expanded concepts.

124
 AI‑powered literature search

This process allows for a comprehensive and nuanced set of search results.
While you don’t need to know exactly what’s going on “under the
hood” of AI to become an AI‑powered scholar, having an awareness of the
background processes can help you make more informed decisions about
when to use an AI tool for literature searches and when to rely on tradi‑
tional methods. In my own practice, I’ve incorporated AI search tools into
my research workflows alongside traditional search tools, which might
lead you to think that I’ve added to my workload rather than reducing
it. Perhaps. But, I’ve found significant benefits in including AI literature
search into the mix. The primary advantage is that I often discover papers
I wouldn’t have found through traditional methods alone. And secondly, if
I’m starting out on a new research project, and don’t yet know what key‑
words might return the best results, I find turning to semantic‑based tools
more beneficial in helping me to initially gather literature that’s attuned
to helping me explore the topic. Overall, using a combination of AI and
traditional tools generally provides a more comprehensive view of the
available literature, often revealing unexpected connections or less obvi‑
ous but highly relevant studies.
Next, we’ll explore practical strategies for leveraging these AI‑­
powered tools in practice.

Conducting AI‑powered literature search

Now that you have a basic understanding of how AI‑powered semantic
search works, let’s explore how to integrate semantic search tools into
your research workflows.
Here are some strategies and approaches:

1. If you’re new to a topic, try starting broad, then refine: Begin with a
broad search query that captures the main concepts of your research
topic. AI‑powered tools can help you identify related concepts and
subtopics you might not have considered. As you review the results,
you can then refine your search based on the most relevant themes
that emerge. The process is highly exploratory and can lead to some
interesting results.
2. Use natural language queries: Remember, when interacting with
chatbots, it’s best to phrase your interactions as if you were speaking

125
AI-Powered Scholar

to a human. Semantic search tools are typically designed to under‑

stand natural language – phrase your queries accordingly. Instead
of trying to craft the perfect combination of keywords, try phrasing
your search as a question or a statement. For example, you might
ask: “What are the effects of mindfulness meditation on overworked
academics?”
3. Combine the AI’s insights with your “human” expertise: While AI can
provide valuable suggestions and connections, you should remain in
the driver’s seat. You are (or are hopefully on the path to becoming)
a subject knowledge expert. Drawing on this expertise is important.
Use the AI’s suggestions as a starting point, but always apply your
critical thinking skills to evaluate the relevance and quality of the
suggested literature.
4. Iterative searching: Use the insights gained from initial searches to
refine and expand your search strategy. AI‑powered tools can help
you identify new keywords, related concepts, or influential papers
that you can use to guide further searches. This can lead to a diver‑
gent creative thinking process whereby your initial topic is expanded
to bring in new perspectives during the ideation phase.
5. Cross‑disciplinary exploration: Semantic search is particularly pow‑
erful for identifying relevant research from adjacent fields. Consider
leveraging this capability to explore how your research topic might
intersect with other disciplines.
6. Validate and verify: Remember that while AI tools are powerful,
they’re not infallible. Credible AI search tools tend to hyperlink to
the sources they are citing. It goes without saying that before you cite
a paper in your research you should ensure that the paper actually
exists and also read it (!!).

Challenges and limitations of AI‑powered

literature search
While AI‑powered semantic search offers tremendous benefits for aca‑
demic research, it’s important to be aware of its limitations and potential
challenges.

126
 AI‑powered literature search

Here are some things to consider, alongside some of the potential

issues and strategies for responsible AI use, raised earlier in Chapter 7:

1. Bias in training data: AI models are trained on existing academic

literature, which may contain historical biases. This could lead to
the perpetuation of these biases in search results. For example,
research from certain countries or in certain languages might be
underrepresented.
2. Overreliance on AI suggestions: There’s a risk of becoming overly
dependent on AI‑generated suggestions, potentially missing impor‑
tant papers that the AI system fails to identify. In the scope of AI’s
current capabilities, I would never rely on just one AI tool as a single
source of truth for finding academic literature. Consider using AI tools
alongside more traditional search options.
3. Lack of transparency: The complex algorithms behind semantic
search can sometimes act as a “black box”,6 making it difficult to
understand why certain results are presented. In effect, there is little
“explainability” for the results you received. This lack of transparency
can be problematic in academic research, where the ability to justify
and explain one’s literature search method may be crucial (e.g. when
conducting a systematic literature review).
4. Limited coverage: Depending on the tool you use to conduct your
semantic search enquiries, the database of academic literature it
searches might not be comprehensive. Some important papers, espe‑
cially from niche fields or non‑mainstream publications, might be
missed.
5. Difficulty with cutting‑edge research: AI models are trained on
existing literature and may struggle to accurately categorise or con‑
nect very recent or cutting‑edge research that uses new terminol‑
ogy or concepts. Be mindful of this when interpreting the search
results.
6. Over‑smoothing of language differences: While the ability to
find conceptually related papers regardless of specific terminol‑
ogy is ­generally beneficial, it might sometimes obscure important
nuances in how different researchers or traditions describe similar
concepts.

127
AI-Powered Scholar

In order to address these challenges, consider the following guiding

principles:

1. Maintain critical awareness: Always approach AI‑generated results

with a critical mindset. Be aware of potential biases and limitations.
2. Combine methods: Use AI‑powered tools in conjunction with tradi‑
tional search methods and human expertise. Cross‑reference results
from multiple sources.
3. Continuous learning: Stay informed about the latest developments
in AI‑powered research tools. Understand how they work and what
their limitations are.
4. Advocate for transparency: Support and use tools that provide trans‑
parency about their algorithms and data sources. You can find these
in the AI tool provider’s terms and conditions documents, which
should be available for any credible AI tool.
5. Supplement with manual searches: Current‑day AI is useful, but it
does not currently represent the panacea of perfection, and it’s often
necessary to supplement AI‑powered searches with traditional search
approaches.

By being aware of these challenges and actively working to address

them, you will be better placed to leverage AI in your literature search pro‑
cess while also maintaining the integrity and rigour of your academic work.
To wrap up this chapter, remember that AI‑powered literature
search, particularly semantic search, represents a significant leap forward
in our ability to navigate the seemingly never‑ending body of knowl‑
edge. By understanding the meaning and context behind our queries,
these tools offer an intuitive and comprehensive approach to discovering
relevant research to inform our literature reviews.
As AI continues to evolve, we can expect even more sophisticated
tools to emerge. It’s for this reason that I haven’t named specific tools in
this chapter, but I promise to include a list of my favourites in the accom‑
panying online companion to this book.7
To conclude, I encourage you to approach AI‑powered literature
search with both enthusiasm and critical awareness. Embrace its potential
to enhance your research, but always apply your human judgement and
expertise.

128
 AI‑powered literature search

Reflection and discussion

To deepen your understanding of AI‑powered literature search and its
implications for academic research, consider the following questions:

1. 
How might the widespread adoption of AI‑powered seman‑
tic search change the way we approach conducting literature
reviews? What skills might become more or less important for
researchers?
2. Consider a recent literature search you conducted. How might
the process and outcomes have been different if you had used
an AI‑­powered semantic search tool? What additional insights
might you have gained, and what potential pitfalls might you have
encountered?
3. What ethical considerations should guide the development and
use of AI in academic literature search? How can we ensure that
the use of AI tools promotes equity and diversity in research rather
than reinforcing existing biases?

Notes
1. Hesse, C., Sartori, M., & Schmitt, S. (2021). Informetric analysis of
scientific literature on semantic search. Future Internet, 13(6), 151.
2. Chu, J. S. G., & Evans, J. A. (2021). Slowed canonical progress in large
fields of science. Proceedings of the National Academy of Sciences of
the United States of America, 118(41), 1–2. https://doi.org/10.1073/
pnas.2021636118
3. For information on available AI‑powered literature search tools, visit
broneager.com/aischolar
4. https://scholar.google.com
5. Toledo, E. G. (2024, March 7). English dominates scientific
research – here’s how we can fix it, and why it matters. The Conversation.
https://theconversation.com/english‑dominates‑scientific‑research‑
heres‑how‑we‑can‑fix‑it‑and‑why‑it‑matters‑226198?

129
AI-Powered Scholar

6. Hassija, V., Chamola, V., Mahapatra, A., Singal, A., Goel, D., Huang,
K., Scardapane, S., Spinelli, I., Mahmud, M., & Hussain, A. (2023).
Interpreting black‑box models: A review on explainable artificial
intelligence. Cognitive Computation, 16(1), 45–74. https://doi.org/
10.1007/s12559‑023‑10179‑8
7. www.broneager.com/aischolar

130
 10 Writing & editing
with AI

Introduction
For many academic writers, me included, there are few tasks more daunt‑
ing than creating a new document on your computer and knowing that
it needs to be filled with words. That blinking cursor, ostensibly patient,
waiting for brilliance to flow from fingertips to keyboard, can feel less like
an invitation and more like an accusation.
“Well?” it seems to taunt, “Aren’t you supposed to be writing some‑
thing brilliant?”
Fortunately, there’s a strategy to break this stalemate: getting some
words – any words – on the page. I know that’s precisely the hurdle you’re
trying to overcome, but bear with me as I share how we can call upon AI
chatbots to lend a hand (or more accurately, a helping algorithm).
Now, I anticipate some readers thinking, “Isn’t using AI for academic
writing cheating?” It’s a valid concern, one hotly debated since ChatGPT’s
public launch.1 As such, it’s worth clarifying that its generally not ok to
pass off large slaps of AI‑generated content as your original work. How‑
ever, much like brainstorming with a colleague or getting feedback from
your supervisor, AI can serve as a thought‑provoking collaborator and is
increasingly skilled at assisting in translating our often‑times rambling
thoughts into well‑structured text; generating outlines for further devel‑
opment; crafting and refining our academic writing; and as an editorial
assistant.
Let’s begin by introducing a technique I call “blah writing”’.

DOI: 10.4324/9781032665276-10131
AI-Powered Scholar

Getting your thoughts on the page

As academic writers, we can find ourselves trapped in a paradox of per‑
fectionism. Our minds are alive with ideas, theories, and half‑formed
arguments, yet the pressure to articulate these thoughts eloquently can
leave us paralysed before a blank page. This is where the concept of “Blah
writing” comes in.
The term “blah” is often used colloquially in English to represent
meaningless or nonsensical talk. You might hear someone say “blah blah
blah” to mimic the sound of ongoing, unimportant chatter. Here, I’ve cho‑
sen to reclaim the term for productive purposes, adopting it to title a strate‑
gic approach to getting your thoughts out of your head and onto the page.
Blah writing is essentially a digital version of stream of consciousness
journaling.2 It involves typing out your thoughts, ideas, and even your
anxieties about your writing task in a rapid, unfiltered manner to an AI
chatbot. The goal isn’t to produce polished prose, but rather to externalise
the whirlwind of thoughts that may be stuck in your head and hindering
the writing process.
This approach can be particularly helpful for anyone who feels intim‑
idated by the pressure to produce “perfect” writing from the outset. Blah
writing gives you permission to be messy, imperfect, and exploratory.
The following is an example workflow for achieving blah writing
alongside an AI chatbot:

1. Set the stage: Begin by assigning your AI chatbot a role, as you learnt
about in Chapter 4. You might cast it as a patient, non‑judgemental
writing coach or a knowledgeable colleague in your field. This helps
create a supportive environment for your unfiltered thoughts.
2. Brain dump: Start typing whatever comes to mind about your topic,
stream of consciousness style. Don’t worry about grammar, structure,
or even coherence at this stage. The aim is to get your thoughts out of
your head and onto the screen. Let your ideas flow freely, no matter
how disjointed or rambling they may seem.
3. Seek synthesis: Once you’ve externalised your thoughts, use your
prompt‑writing skills from Chapter 6 to ask the AI to help synthesise
your ideas. Request it to clarify your thoughts and translate the mess
you’ve dumped into it into a coherent narrative.

132
 Writing & editing with AI

4. Reflect and iterate: Review the AI’s suggestions critically. Use them as
a catalyst for your own critical thinking and further writing. Consider
whether the synthesis accurately reflects your thoughts, if important
points were overlooked, and how you might expand on these themes.

An example of how this might play out in practice is as follows:

1. Set the stage: “Please act as a supportive writing coach specialising

in educational technology. I need help organising my thoughts on AI
implementation in higher education”.
2. Brain dump: “What am I doing writing stresses me trying this brain
dump of thoughts thesis needs to get written so here goes. Want
to talk about adaptive learning AI systems in universities great
potential to personalise the student experience by tracking prog‑
ress and suggesting resources tailored to individual needs but there
are huge implementation challenges to consider especially around
data privacy and ethical issues like bias in algorithms that could
unintentionally disadvantage certain groups we need to think care‑
fully about how lecturers fit into this new paradigm as they might
become more like facilitators guiding students through AI‑enhanced
learning journeys which sounds interesting but could be threatening
to some staff members who’ll need training and upskilling in AI lit‑
eracy and effective integration of these systems into the curriculum
it’s not just a plug and play solution and might require redesigning
entire courses which is exciting but daunting and time‑consuming
with significant resource implications to consider but the potential
benefits are substantial like students getting immediate feedback
24/7 which could increase engagement and improve retention rates
by helping identify struggling students earlier for timely interven‑
tions though we need to balance this with ethical considerations
about how much monitoring is too much and ensure accessibility
for students with disabilities it’s a complex topic with lots to con‑
sider but definitely worth exploring further as it could revolutionise
higher education”.
3. Seek synthesis: “Can you please rephrase this jumbled paragraph into
a more coherent narrative? I’m trying to articulate my thoughts on
adaptive learning AI systems in higher education more clearly”.

133
AI-Powered Scholar

AI‑generated response:3 “Certainly. Here’s a more coherent nar‑

rative based on your thoughts:
Adaptive learning AI systems offer significant potential for per‑
sonalising the student experience in higher education. These systems
can track individual progress and suggest tailored resources, poten‑
tially leading to increased engagement, improved retention rates, and
earlier identification of struggling students for timely interventions.
However, the implementation of such systems presents consider‑
able challenges. Primary concerns include data privacy and ethical
issues, particularly the risk of algorithmic bias that could inadver‑
tently disadvantage certain student groups.
Moreover, the integration of AI systems may necessitate a funda‑
mental shift in teaching roles, with lecturers potentially evolving into
facilitators who guide students through AI‑enhanced learning jour‑
neys. This transition is not without its hurdles. It requires substantial
investment in staff training, focusing on AI literacy and the effective
integration of these systems into the curriculum. The process may
involve redesigning entire courses, which, while potentially exciting,
is also daunting, time‑consuming, and resource‑intensive.
Furthermore, ethical considerations must be carefully balanced.
While continuous monitoring can provide valuable insights, it’s cru‑
cial to establish appropriate boundaries to protect student privacy.
Additionally, ensuring accessibility for students with disabilities
must be a priority in the development and implementation of these
AI systems.
In conclusion, while adaptive learning AI systems present
complex challenges, their potential to revolutionise higher educa‑
tion makes them a topic worthy of further exploration and careful
consideration”.
4. Reflect and iterate: “This is a great start, thanks…”
The conversation would continue, with the academic reflecting
on the AI’s suggestions, asking for further clarification or expansion as
needed, and being in a position to start adding to the text and getting
more words on the page.

As you can see from the worked example above, “blah writing” can
transform a chaotic stream of thoughts into a “draft zero”, comprising
a structured, coherent narrative. It can help you to overcome the initial

134
 Writing & editing with AI

hurdle of writer’s block but also provide you with a foundation for further
development of your ideas. Remember that the goal is not necessarily the
production of a finished product, but to kickstart your writing process and
generate a base from which to build.

Generating outlines
Another strategy for overcoming writing roadblocks with AI chatbots is
to employ their capabilities to create a comprehensive outline for your
upcoming publication. By generating headings and subheadings, AI can
help create a “paint by numbers” scenario, eliminating the guesswork
about what to write next. This method draws on the approach of Professor
Rowena Murray,4 who visited my university during my PhD candidature
and changed the way I approached future writing tasks. During her ses‑
sion, and as is well documented in her many books, she recommended
creating highly prescriptive outlines, including specifying word counts for
each section.
These detailed outlines allow you to approach your writing flexibly,
enabling you to work on any section at any time, simply by filling in the
required number of words. Essentially, they provide you with a framework
for structuring your thoughts and can take away some of the stress of
the intimidating expanse of white space, replaced by a clear action plan
which you can dip in and out of as time permits.
To create an outline with the help of AI, I recommend beginning by
assigning your AI assistant a role.
An example role prompt could look something like this:

Please assume the role of a senior research assistant with

expertise in [Insert topic, e.g., educational technology and AI
implementation in higher education]. You have a background
in qualitative research and twenty years of experience writing
academic papers for high‑quality journals. Communicate in a
professional yet approachable tone, and don’t hesitate to ask for
clarification if needed.

You might also provide an exemplar outline, based on an existing paper.

This approach is particularly beneficial when you aim to emulate a

135
AI-Powered Scholar

specific structure or style common in your field of study. By doing so,

you’re essentially giving the AI a template to work from, ensuring that your
outline aligns with established conventions in your discipline. But, before
uploading anything, ensure you have the necessary permissions. Most
leading AI models are capable of extracting the outline from an uploaded
document, so there’s no need for you to go through the ­document and
type it out.
For instance, you might prompt:

I’ve uploaded a paper titled [Paper Title] that follows the structure
I’d like to emulate. Please analyse this paper’s structure, focusing
on its headings and subheadings. Then, using this structure as a
guide, create an outline for a 5,000‑word paper on [Your Topic].
Adapt the headings and subheadings to fit my topic while main‑
taining a similar overall structure. For each section, please sug‑
gest an approximate word count.

If you don’t have an exemplar outline/paper, you might use a prompt like
this:

Please create a detailed outline for a 5,000‑word academic

paper on [Your topic]. The paper should follow a standard
research article structure including an introduction, litera‑
ture review, methodology, results, discussion, and conclusion.
Please provide main headings and subheadings and suggest an
approximate word count for each section. The outline should
be suitable for a qualitative research paper to be submitted to a
peer‑reviewed journal in [Your field]. Thanks!

For example, if your topic was “The Impact of AI on Higher Education

Teaching Practices”, your prompt might look like this:

Please create a detailed outline for a 5,000‑word academic

paper on ‘The Impact of AI on Higher Education Teaching
Practices’. The paper should follow a standard research article
structure including an introduction, literature review, methodol‑
ogy, results, discussion, and conclusion. Please provide main
headings and subheadings and suggest an approximate word

136
 Writing & editing with AI

count for each section. The outline should be suitable for a qual‑
itative research paper to be submitted to a peer‑reviewed journal
in educational technology. Thanks.

Remember, while AI can provide a valuable starting point, the final struc‑
ture should always be refined based on your unique insights, research
findings, and the specific requirements of your academic project.

Expanding text
There are often times when we find ourselves needing to elaborate on our
ideas or deepen the level of description to justify our arguments. In other
words: we find ourselves in need of more words so that we can satisfy
required wordcounts. AI tools can be used for this task by assisting us in
rephrasing our already‑written text, suggesting additional points, and by
offering ways we could expand our writing to achieve greater depth and
clarity.
If you’re thinking of using AI for this task, it’s important to exhibit
some caution.
As was discussed in Chapter 7, AI‑generated content contains many
limitations (e.g. hallucinations) and should always be critically evaluated
and refined through your own expertise and judgement. Therefore, while,
in theory, people are presumed innocent until proven guilty in court‑
rooms, the opposite mindset is wise to apply to AI‑generated content,
especially when that content is contributing to the creation of academic
publications! I always take a “guilty until proven innocent” approach
when encountering AI‑generated text – every AI suggestion is treated
with a healthy dose of scepticism, and only “acquitted” after thorough
verification.
With this firmly front of mind, let’s explore how you might use AI
chatbots to expand a sample of your academic writing:
To use an AI chatbot for expanding text:

1. Identify the section to expand: Highlight the specific paragraph or

passage that you want to develop further.
2. Provide a clear prompt: Give the AI chatbot a specific instruction for
expansion, such as “Please provide more detailed explanations of the

137
AI-Powered Scholar

key concepts in this paragraph”, “Suggest ways to elaborate on the

evidence presented here to strengthen the argument”, or simply use a
prompt such as “Please expand this text”.
3. Review and refine: Read through the AI’s suggestions and integrate
the most relevant and useful expansions. Be sure to maintain your
own voice and style and ensure that any added material enhances
rather than detracts from your overall argument.

If you try this approach, I urge you to think of yourself as both a collabora‑
tor and a gatekeeper – working with the AI to generate ideas, but carefully
filtering and refining its output. Remember, the goal is not just to increase
your word count, but also to enhance the depth, clarity, and overall qual‑
ity of your work.

Contracting text
At the start of a writing project our goal tends to be focused on generating
more words. However, there will likely come a time when you also need
to cut them back. For example, when you receive a dreaded “desk reject”
and find out that the next journal on your list of potential homes for your
paper has a significantly lower maximum word count. AI chatbots can
help speed up the process by suggesting ways to eliminate redundancies,
trim extraneous details, and articulate your core message.
Here’s one way you might get the job done:

1. Select the text you want to shorten: First, highlight the paragraph or
passage you’re focusing on. Then, save it temporarily by copying it.
This allows you to use it again right away without having to retype it.
2. Provide a clear prompt: Give the AI chatbot a specific instruction for
contraction, such as “Please suggest ways to eliminate unnecessary
details and make this passage more concise”, or “Identify any redun‑
dant or repetitive phrases that could be removed from this paragraph”.
3. Review and refine: Read through the AI’s suggestions and integrate the
most relevant and useful contractions into your text. Be mindful not
to remove any essential information or nuance and ensure that your
condensed version still effectively communicates your key points.

138
 Writing & editing with AI

To demonstrate how this process works in practice, let’s look at an

example. The following output was generated using Claude, an AI chat‑
bot running its 3.5 Sonnet model. Claude generated a result (reproduced
below) after being provided with the following sample text and prompt:

Sample text: The researchers conducted in‑depth, semi‑struc‑

tured interviews with 15 university lecturers who had imple‑
mented AI‑powered adaptive learning systems in their courses.
The interviews, lasting between 60 and 90 minutes each, were
conducted via video conferencing software due to COVID‑19
restrictions. The interview questions focused on the lecturers’
experiences with the AI systems, their perceptions of the impact
on student learning, and the challenges they faced during imple‑
mentation. All interviews were audio‑recorded with participant
consent and subsequently transcribed verbatim. The researchers
then employed a thematic analysis approach, following Braun
and Clarke’s six‑step process, to identify key themes and patterns
in the data. This involved multiple rounds of coding and theme
refinement, with regular team meetings to discuss and validate
the emerging findings.
Prompt: “Please suggest ways to make this passage more
concise by removing unnecessary details and streamlining the
description of the qualitative research process, while maintain‑
ing the essential methodological information”.
Contracted text: “Semi‑structured interviews were con‑
ducted with 15 university lecturers who had implemented
AI‑powered adaptive learning systems. Interviews, conducted
virtually, explored lecturers’ experiences, perceived impacts on
student learning, and implementation challenges. Using Braun
and Clarke’s thematic analysis approach, researchers identified
key themes through iterative coding and team validation”.

As with all aspects of AI‑assisted writing, it’s important to approach

these suggestions with a critical eye and with a commitment to main‑
taining your own authorial voice and intent. It’s best to consider the
chatbot’s recommendations as a starting point for refinement, ensuring
that you always make the final decisions about what to include and
what to cut.

139
AI-Powered Scholar

Editing with AI
Between teaching responsibilities, conducting research, analysing data,
and juggling university admin, having the headspace to conduct a thor‑
ough edit of your work can be a challenge. For those fortunate enough to
have access to a professional editor, willing colleague, caring supervisor,
or generous friend, the benefit of having a skilled human look over your
work can’t be overstated. Unfortunately, there are many barriers to acquir‑
ing editorial assistance. And while AI tools are currently far from perfect
compared to a professional editor, they do offer some benefits worth con‑
sidering, including:

1. Round‑the‑clock availability: AI chatbots don’t keep office hours.

They’re ready to assist you whenever you need.
2. Instant feedback: With an AI chatbot, you don’t have to wait days or
weeks for feedback on your writing. Suggestions are provided in real
time, allowing you to iterate and improve your work according to
your schedule.
3. Cost‑effective solution: Human editors can cost thousands of dollars,
especially when tasked with editing lengthy academic documents. AI
chatbots offer a relatively affordable alternative, at least in terms of
your out‑of‑pocket expenses.5
4. A “safe space”: Sometimes we’re hesitant to share our writing with
others for fear of judgement. Chatbots can offer a sounding board
without that fear attached. Additionally, they are typically trained
to offer supportive advice, unless, of course, you decide to prompt
your chatbot to “provide critiques in the style of a grumpy aca‑
demic at a conference, poised to unleash scathing comments dur‑
ing the Q&A”.

To use an AI chatbot in the editing process, you will (obviously) need

some text that requires editing. It could be as small as a sentence, or
as large as thesis. When you’ve got some words to work with, here’s an
example workflow for completing the process:

1. Choose your AI platform: For longer edits, select a platform that has
document upload capabilities. This feature allows you to input larger

140
 Writing & editing with AI

portions of text or entire documents for analysis. Importantly – ensure

that the platform you choose complies with your institution’s data
protection and privacy policies.
2. Prepare your document: Before uploading, review your document to
ensure it doesn’t contain any sensitive or confidential information.
3. Upload the document: Follow the platform’s instructions for
uploading your document. This usually involves selecting the file
from your computer (typically by clicking on a paperclip icon),
or cloud storage provider, or copying and pasting your text into a
designated area.
4. Prompt the chatbot: This is where your skills in crafting effective
prompts (if you need a refresh, see Chapter 6) come into play. Be
specific about what kind of editing assistance you need. For example:
“Please review this text for clarity and coherence. Suggest improve‑
ments in sentence structure, identify any awkward phrasing, and
highlight areas where the argument could be strengthened. Pay par‑
ticular attention to the flow between paragraphs and the use of aca‑
demic language”.
5. Review the analysis: Carefully read through the AI’s suggestions.
While AI can provide valuable insights, it’s not infallible.

Lastly, always use your critical thinking skills to evaluate each suggestion,
considering:

• Does the suggested change improve clarity or flow?

• Does it maintain your intended meaning?
• Is it consistent with your voice and style?

As you become more comfortable with AI‑assisted editing, you might find
it particularly useful for tasks such as:

• Checking for consistency in terminology and style

• Identifying overused words or phrases
• Suggesting more precise vocabulary
• Highlighting areas where citations might be needed
• Improving the overall structure and flow of your argument

141
AI-Powered Scholar

It’s important to recognise that what’s been covered in this chapter, and
the book more broadly, is just the tip of the iceberg when it comes to
leveraging AI in academic tasks.
If you’re enthusiastic to explore additional AI applications, you’ll find
the final chapter of this book to be an invaluable resource. It provides
access to online materials, including practical use cases, allowing you to
continue your journey into AI‑powered scholarship.

That’s a wrap!
It was tempting to conclude this book with a playful snippet of AI‑gen‑
erated text. Perhaps by using a prompt such as: “Please craft a three‑line
poem, that expresses to my family and friends that, finally, the book has
been written, and I’ll soon be returning to pre‑ChatGPT patterns of social
interaction after having a day off to celebrate”.
But just as non‑disclosed AI‑generated content continues to be
judged as a misrepresentation of human effort, so too would that poem’s
sentiments misrepresent any intention of dialling down my enthusiasm for
building AI skills and sharing them widely. I long jumped down the rabbit
hole of AI, and, like many other AI enthusiasts who attached themselves
to the AI rocket ship, I continue to thrive on its energy.
The pace of advancement, including the almost daily updates about
new features, apps, business cases, etc. provides more dopamine hits than
a bottle of Vyvanse.6 AI has significantly enhanced my life, and through
sharing some of what I’ve learnt so far, I hope to have nudged you towards
unlocking similar benefits.
Yet, many across the sector are understandably panicked by AI,
viewing it as the end of academic scholarship and the death of academic
writing.
Which is why I’d like to ask you a small favour.
Remember when we metaphorically unwrapped the gift of AI in
Chapter 1? Now that you’ve journeyed through this book, you’ve hope‑
fully not only unwrapped that gift but have begun to feel comfortable
using it. This gift is not a zero‑sum game. I believe lifting others up on their
AI journey will help lift academia more broadly, and thus help us to retain
integrity and relevance (and the chances of ongoing employment!) in a
sector that’s facing massive disruption. I encourage you to become an AI

142
 Writing & editing with AI

literacy ambassador in your academic community. Share your newfound

skills and insights with colleagues and students. Organise workshops, ini‑
tiate discussions, or simply offer a helping hand to those still grappling
with the blinking cursor on a blank page.
Thank you for coming along on the journey through this book. If you
now feel a little more comfortable using AI, are a more informed user,
or just got some enjoyment from my often‑times self‑indulgent narrative
tangents, I’ve achieved what I set out to do. If you found value in these
pages, I would be very grateful if you could do the proverbial “like and
subscribe” on social media to help promote the book online.
Lastly, I urge you to be mindful that the AI revolution in academia
isn’t just about technology – it’s about reimagining the very essence of
higher education. But that’s a topic for a whole other book!

Notes
1. Jensen, L. X., Buhl, A., Sharma, A., & Bearman, M. (2024). Gen‑
erative AI and higher education: a review of claims from the first
months of ChatGPT. Higher Education, 1–17. https://doi.org/10.1007/
s10734‑024‑01265‑3
2. Cameron, J. (2020). The Artist’s Way: A Spiritual Path to Higher Cre‑
ativity. Souvenir Press.
3. The provided response was generated by Claude.ai, using Claude 3.5
Sonnet.
4. To learn more about Professor Rowena Murray, visit https://www.
anchorage‑education.co.uk/about‑me/
5. For insights into AI’s costs, including environmental impacts, refer to:
Crawford, K. (2021). The Atlas of AI: Power, Politics, and the Planetary
Costs of Artificial Intelligence. Yale University Press.
6. Prescription medication used to treat Attention Deficit Hyperactivity
Disorder (ADHD).

143
 11 Bonus chapter

If this book had included mentions of the “best AI tools for the job”, they
would likely be outdated before the ink dried on the page.
Providing this information via an online companion resource allows
it to be as fresh as your morning coffee, updated as new tools emerge and
others become obsolete.
Unlock bonus material by visiting: www.broneager.com/aischolar

DOI: 10.4324/9781032665276-11144
 Glossary

As you work through the chapters in this book, you’ll likely encounter
some unfamiliar terms. This glossary is designed to be your handy com‑
panion for navigating common AI terminology. It’s not an exhaustive
encyclopaedia of AI jargon but rather a friendly guidebook. Here, you’ll
find clear and concise explanations of key concepts, essential for building
your AI literacy skills.

A
Accuracy (in machine learning) A metric used to evaluate the perfor‑
mance of a machine learning (ML) model, measuring the proportion
of correct predictions among the total number of cases examined.
AGI (Artificial General Intelligence) A hypothetical form of AI that would
have the ability to understand, learn, and apply intelligence in a way that
matches or surpasses human cognitive abilities across a wide range of
tasks. Unlike narrow AI, which is designed for specific tasks, AGI would
be capable of generalising its knowledge and skills to new situations.
AI‑assisted (AI‑augmented) This term describes the use of artificial intel‑
ligence to enhance human capabilities in various tasks. The primary
goal is to increase efficiency, accuracy, and productivity, allowing
academics to focus on more innovative and strategic elements of their
work. AI‑assisted tools act as digital collaborators, augmenting rather
than replacing human expertise.

145
Glossary

AI‑generated (outputs) Content or outputs predominantly created by

artificial intelligence systems. Examples include text, images, music,
videos, and other data forms. These are often crafted using sophis‑
ticated AI techniques such as deep learning and natural language
processing. AI systems analyse vast datasets to produce innovative
content that showcases human‑like creativity or analytical capabili‑
ties. The quality and complexity of AI‑generated content hinge on
various factors, including the algorithms and training data used.
AI literacy The understanding and skill set required to effectively inter‑
act with, manage, or leverage artificial intelligence technologies. This
encompasses a basic knowledge of what AI is, how it works, and
its potential applications. AI literacy goes beyond technical skills,
embracing an awareness of ethical considerations, biases in AI, and
the impact of AI on various sectors such as employment, privacy, and
security.
AI tools Software applications or platforms that incorporate artificial
intelligence to perform or improve tasks. AI tools are increasingly
being integrated into academic workflows to enhance research and
writing processes.
AI winter/AI summer In the history of artificial intelligence research,
an “AI winter” refers to a period where progress stagnated, leading to
reduced interest and funding. Conversely, an “AI summer” describes
a time of accelerated progress, sparking renewed interest and invest‑
ment in the field.
Algorithm A set of step‑by‑step instructions or rules designed to perform
a specific task or solve a particular problem. In computing, algorithms
are used to manipulate, process, and analyse data, forming the basis
of computer programs. The efficiency and effectiveness of an algo‑
rithm are crucial, as they directly impact the performance and capa‑
bilities of a software application or system.
Alignment The consistency of an AI system’s actions and decisions with
human values and intentions. This involves designing AI models that
understand and adhere to ethical guidelines, societal norms, and the
specific objectives set by their human developers or users. Alignment
is a critical concern in AI safety and ethics, particularly as AI systems
become more capable.
Anthropic An artificial intelligence research and safety company
founded by former employees of OpenAI (the company behind

146
 Glossary

ChatGPT). Anthropic focuses on developing large‑scale AI models

while emphasising AI safety and ethical considerations. Their popu‑
larly used chatbot, Claude, is available at www.claude.ai.
API (Application Programming Interface) A set of protocols, routines,
and tools for building software applications. In the context of AI,
APIs allow developers to integrate AI capabilities into their applica‑
tions or access AI services provided by companies like OpenAI or
Google.
Artificial intelligence A branch of computer science focused on ­creating
systems that can perform tasks which typically require human intel‑
ligence. These tasks include learning, reasoning, problem‑solving,
perception, language understanding, and creativity. AI systems range
from simple programs that solve specific problems to complex learn‑
ing systems that adapt and operate autonomously.
Attention mechanism A key component in many modern AI models,
particularly in natural language processing. It allows the model to
focus on specific parts of the input data when producing an output,
mimicking human attention and greatly improving the performance
of tasks like translation and text summarisation.
Augmentation The enhancement of human capabilities or experiences
through the use of technology, particularly artificial intelligence and
other digital tools. In academic contexts, augmentation can refer to
the use of AI to support and enhance research, writing, and analytical
processes.
Automation/Automated The technology‑enabled process of making
systems or processes function automatically, without human interven‑
tion. In academia, automation can streamline repetitive tasks, allow‑
ing researchers to focus on other aspects of their work.
Autonomous systems AI systems capable of operating and making deci‑
sions without direct human control. These systems use ML algorithms
to adapt to new situations and improve their performance over time.

B
Bias Systematic errors or unfairness in data, algorithms, or interpretation
of results. It can manifest in various forms, including data bias, algo‑
rithmic bias, and interpretation bias. Addressing bias in AI is critical to

147
Glossary

ensure fairness, accuracy, and ethical use of technology in academic

research. This involves careful data selection, algorithm design, and
ongoing monitoring to identify and mitigate potential biases that
could skew research outcomes or perpetuate societal inequalities.
Big Data Extremely large datasets that may be analysed to reveal pat‑
terns, trends, and associations, especially relating to human behav‑
iour and interactions. In academia, Big Data offers unprecedented
opportunities for research but also presents challenges in terms of
storage, analysis, and ethical considerations.
Black Box AI AI systems whose decision‑making processes are not trans‑
parent or easily interpretable by humans. This lack of explainability
can be problematic in academic research, where understanding the
reasoning behind conclusions is crucial.
Bodies of knowledge In the context of AI and academic research, this
term refers to the vast collections of information, theories, and data.
It’s essentially the AI’s “textbook” – a comprehensive repository of
information it can draw upon to generate responses or perform tasks.
Bot Short for “robot”, a bot is a software application that runs auto‑
mated tasks over the internet. Bots can be used for tasks like literature
searches, data collection, or even as research assistants.

C
Chatbot A software application designed to simulate conversation
with human users, especially over the internet. It uses natural lan‑
guage processing (NLP) and artificial intelligence to understand user
queries and generate responses. Chatbots can range from simple,
rule‑based systems that respond to specific keywords or phrases, to
more advanced AI‑driven bots capable of learning and adapting to
user interactions over time.
ChatGPT An AI chatbot developed by OpenAI, based on the Generative
Pre‑trained Transformer (GPT) architecture. ChatGPT has gained sig‑
nificant attention in academia for its ability to engage in human‑like
conversations, answer questions, and assist with writing tasks.
Clustering A ML technique that involves grouping similar data points
together. It’s used in academic research for tasks, such as data explo‑
ration and pattern recognition.

148
 Glossary

Context window The amount of text (information) an AI can “remem‑

ber” and consider when generating a response. Think of it as the AI’s
short‑term memory or attention span.
Conversation‑style chatbot A type of chatbot designed to mimic
human‑like conversational interactions.
Compute The amount of computational power and resources required
to run AI models or perform technological tasks. The availability of
compute can significantly impact the scale and complexity of AI
models that can be developed and deployed.
Computational power The capacity of a computer or a system to process
and manipulate data. It is typically measured by the speed and effi‑
ciency with which a system can execute instructions, handle complex
calculations, or process large amounts of data. Advancements in com‑
putational power have been a key enabler for the development and
deployment of sophisticated AI applications in academic research.
Computer program A sequence of instructions written to perform a
specified task with a computer. These instructions provide the guide‑
lines for how the program operates and interacts with other software
and hardware.
Co‑pilot A system or tool that assists users in completing tasks. In the
context of AI, it refers to AI‑powered assistants that work alongside
humans, augmenting their capabilities.

D
Dataset A collection of data used in research and analysis. Key char‑
acteristics include: size (the number of records and features); type
(e.g. numerical, categorical, textual, or a combination); structure (e.g.
organised in a defined schema or unstructured); quality (accuracy,
completeness, and consistency); and source (origin and collection
method, which can affect reliability and relevance).
Deep learning A subset of ML based on learning data representations,
as opposed to task‑specific algorithms. It involves the use of large
neural networks with many layers of nodes (hence “deep”), trained on
vast amounts of data. These neural networks attempt to simulate the
behaviour of the human brain – albeit in a limited sense – to process
data and create patterns for decision making.

149
Glossary

Deep learning architecture The structure and design of neural networks

used in deep learning. These architectures are layered constructs of
algorithms intended to model high‑level abstractions in data.
Document analysis The use of AI tools to extract, analyse, and sum‑
marise information from academic papers, reports, or other docu‑
ments. This can be particularly useful for literature reviews or data
extraction from multiple sources.
Draft zero AI tools can create initial drafts of academic writing. Ideally,
these drafts should be used for brainstorming and/or getting your ini‑
tial thoughts on the page, and not be submitted for publication as a
replacement for human contributions.

E
Ethical AI The development and use of artificial intelligence in ways
that adhere to moral principles and values. For academics, this might
involve considering the ethical implications of using AI tools in
research, such as potential biases, privacy concerns, or the impact on
human subjects.
Explainability The degree to which the internal workings or deci‑
sion‑making processes of an AI system can be understood and inter‑
preted by humans. Explainability allows researchers to understand
how conclusions were reached, validate results, and identify poten‑
tial biases or errors. Explainable AI (XAI) refers to methods and tech‑
niques that make AI systems more transparent and interpretable,
which is particularly important in fields where the reasoning behind
decisions or predictions needs to be clearly articulated.
Everyday AI AI technologies and tools that have become integrated
into daily academic workflows. This term encompasses the various
AI‑powered applications that researchers might use regularly, such as
writing assistants, reference managers, or data analysis tools, which
can enhance productivity and efficiency in academic work.
Expert systems A branch of artificial intelligence that emulates the deci‑
sion‑making ability of a human expert, designed to solve complex
problems by reasoning through bodies of knowledge. In academic
contexts, expert systems can assist researchers in specialised fields

150
 Glossary

by providing guidance or analysis based on established knowledge

in the discipline.

F
Feedback loop The iterative process of providing input to an AI system,
receiving output, and then refining the input based on the output. This
cycle can help in improving the relevance and accuracy of AI‑gener‑
ated content.
Few‑shot learning An AI capability where a model can make accurate
predictions based on only a few examples. This can be useful when
working with limited datasets or when trying to apply AI tools to
niche research areas.
Fine‑tuning The process of adapting a pre‑trained AI model to a spe‑
cific task or domain. For academics, this could involve customising a
general‑purpose AI writing assistant to better understand and generate
content in their specific field of study.

G
Generative AI (GenAI) AI systems capable of creating new content,
such as text, images, or code. Generative AI can assist with tasks like
drafting literature reviews, generating research questions, or creating
visual aids for presentations.
GPT (Generative Pre‑trained Transformer) A type of large language
model that uses deep learning to produce human‑like text. GPT
models, such as those powering ChatGPT, are increasingly used
by academics for tasks like writing assistance, brainstorming, and
summarisation.
GPT‑3 (GPT‑4, etc.) Specific iterations of the GPT language model
developed by OpenAI. GPT‑3 was groundbreaking at the time.
GPT‑4 was its successor. These models power various AI tools used
in academic workflows, such as writing assistants, language transla‑
tion, and text summarisation. While powerful, academics should be
aware of their limitations, including potential biases and the need to

151
Glossary

fact‑check outputs. As new versions are released, they are likely to

offer enhanced capabilities but may also present new challenges for
academics.

H
Hallucinations Instances where AI models generate plausible‑sounding
but factually incorrect or nonsensical information. For academics
using AI tools, it’s crucial to be aware of this limitation and to verify
AI‑generated content.
Hybrid intelligence The combination of human and artificial intel‑
ligence to solve complex problems. In academic workflows, this
often involves using AI tools to augment human capabilities, such as
using AI for initial data analysis before applying human expertise for
interpretation.
Human‑in‑the‑loop An approach where human input (intervention and
oversight) is incorporated into AI systems to improve decision‑making
and outcomes.

I
Intelligent machines Systems or devices that mimic human intelligence,
capable of learning, problem‑solving, and adapting to new situations.
Intelligent machines can range from AI writing assistants to complex
research tools that analyse data.
Information retrieval The process of obtaining relevant information
from a large collection of data. AI‑powered information retrieval sys‑
tems can significantly enhance academic research by quickly iden‑
tifying and extracting information from vast databases of scholarly
literature.
Iterative prompting The process of refining and adjusting prompts given
to AI systems to improve the quality and relevance of outputs.
Interpretability The degree to which an AI system’s decision‑making
process can be understood by humans. For academics using AI tools,

152
 Glossary

interpretability is crucial for understanding the reliability and poten‑

tial biases of AI‑generated outputs.
Implicit bias in AI Unintended prejudices that can be present in AI
­systems, often reflecting biases in training data or algorithm design.
Academics should be aware of potential implicit biases when
using AI tools for research or analysis to ensure fair and accurate
results.

K
Knowledge base A structured collection of information that an AI system
uses to generate responses or make decisions.
Keyword extraction An AI‑powered process that automatically identifies
and extracts the most important or frequently used words or phrases
from a text. This functionality can be particularly useful for academics
when summarising research papers or identifying key themes in large
bodies of literature.
Knowledge graph A network representation of relationships between
entities (e.g. concepts, people, events) in a knowledge domain.
AI‑powered knowledge graphs can help researchers visualise con‑
nections within their field of study, potentially revealing new research
directions or unexplored relationships between concepts.

L
Large language models (LLMs) Advanced AI systems trained on vast
amounts of text data, capable of understanding and generating
human‑like text. LLMs, such as GPT‑3 and GPT‑4, represent the AI
technology underpinning many AI writing assistants and research
tools used in academia.
Large neural network models Complex artificial neural networks with
many layers and parameters, designed to process and analyse large
amounts of data. These models underpin various AI tools used for
tasks such as natural language processing, image recognition, and
data analysis.

153
Glossary

M
Machine learning A subset of AI that focuses on the development of
algorithms and statistical models that enable computer systems to
improve their performance on a specific task through experience. ML
is used in various tools for data analysis, pattern recognition, and
predictive modelling across disciplines.
Model(s) In AI, a model refers to the computational representation of a
system or process, trained on data to make predictions or decisions.
Multi‑modal AI systems that can process and integrate information from
multiple types of input (e.g. text, images, audio). Multi‑modal AI tools
can be particularly useful for researchers working with diverse data
types or in interdisciplinary fields.

N
Narrow AI Also known as “weak AI” or ‘specialised AI’, narrow AI refers
to artificial intelligence systems designed to perform specific tasks
within a limited domain. Unlike the sci‑fi dream of general AI that
can match human‑level cognition across all areas, narrow AI excels
at particular functions but lacks broader capabilities. Think of it as a
highly skilled specialist rather than a jack‑of‑all‑trades.
Natural language The language that humans naturally speak or write,
as opposed to artificial or computer languages. In the context of
AI, natural language refers to the ability of AI systems to under‑
stand and generate human language in a way that feels natural and
intuitive.
Natural language processing (NLP) A branch of AI that focuses on the
interaction between computers and human language. NLP enables
AI systems to understand, interpret, and generate human language.
NLP powers many writing assistance tools, chatbots, and research
aids.
Neural network A computing system inspired by the biological neural
networks in human brains. Neural networks are the foundation of
many modern AI systems, including those used in academic tools for
data analysis, language processing, and pattern recognition.

154
 Glossary

O
OpenAI A research organisation focused on developing AI, perhaps best
known for creating ChatGPT – putting the higher education system
into a spin, largely due to concerns that students would use it in ways
that compromised academic integrity.
Optical Character Recognition (OCR) AI‑powered technology that con‑
verts different types of documents, such as scanned paper documents,
PDFs, or images, into editable and searchable data.

P
Persona A set of characteristics and behaviours assigned to an AI system
to give it a distinct personality or role. Academics might use different
AI personas for various tasks, such as a “research assistant” persona for
literature reviews or a “writing coach” persona for manuscript editing.
Program A set of instructions that tell a computer how to perform a
specific task.
Prompt An input or instruction is given to an AI system to elicit a spe‑
cific response or action.
Prompt engineering The practice of designing and refining inputs to AI
systems to achieve desired outputs. This skill is increasingly important
for academics using AI tools, as it can significantly impact the quality
and relevance of AI‑generated content.
Plagiarism detection tools Tools that compare text against a large data‑
base of sources to identify potential instances of plagiarism. Please
question the reliability and validity of their outputs before putting
your trust in them!

R
Reinforcement learning A type of ML where an AI agent learns to make
decisions by taking actions in an environment to maximise a reward.
Robustness The ability of an AI system to maintain its performance
under various conditions or when faced with unexpected inputs.

155
Glossary

For academics, considering the robustness of AI tools is important

when relying on them for critical research tasks.

S
Semantic (search) A search method that attempts to understand the
intent and contextual meaning of search queries, rather than just
matching keywords.
Sentiment analysis AI techniques used to determine the emotional tone
behind words. This can be useful in analysing large text datasets.
Singularity A hypothetical future point at which artificial intelligence
surpasses human intelligence.
Strong AI Also known as “artificial general intelligence” (AGI), strong
AI refers to hypothetical AI systems capable of human‑like general
intelligence, having the ability to understand, learn, and apply intel‑
ligence across a wide range of domains. It’s a concept that remains
largely theoretical and is a subject of ongoing research and debate.
Supervised Learning A type of ML where the algorithm is trained on a
labelled dataset.
Systems (AI Systems) Integrated sets of AI technologies designed to per‑
form complex tasks or solve specific problems.

T
Training data The dataset used to teach an AI model how to perform its
task. Understanding the nature of training data can help academics
critically evaluate the potential biases or limitations of AI tools.
Turing Test A test of a machine’s ability to exhibit intelligent behaviour
equivalent to, or indistinguishable from, that of a human.

U
Unsupervised Learning A type of ML where the algorithm tries to find
patterns in unlabelled data. This concept is relevant for academics

156
 Glossary

working with exploratory data analysis or trying to discover hidden

patterns in large datasets.

V
Vectorisation The process of converting text into numerical vectors that
can be processed by ML algorithms. This underlies many AI‑based
text analysis tools used in academic research.

W
Word embedding A technique in NLP where words or phrases are
mapped to vectors of real numbers. This underlies many advanced
text analysis tools used in academic research.

X
XAI (Explainable AI) AI systems designed to be interpretable and under‑
standable by humans.

Z
Zero‑shot learning The ability of an AI model to solve tasks it wasn’t
explicitly trained on.

157