Arbitrator-Robot : Is A(I)DR the future?

Abhilasha Vij 

The practice of law has seen a boom in the use of technology, particularly in the arena

of alternative dispute resolution (ADR). Owing to the characteristics of disputes catered to,

ADR generally involves the use of technology as ‘fourth participant’ in the proceedings.

Regardless of the efficiency and cost effectiveness introduced by technology, until quite

recently, parties and law practitioners showed preference for in-person court or ADR

proceedings. Before the Covid-19 pandemic disrupted the incumbent administrative and

commercial activities around the world, virtual courts and virtual ADR proceedings were

hardly in use. Now, these encompass the truth of the practice of law. With the unforeseeable

change in the demands of consumers of legal services, as well as, the manner in which justice

has to be administered, there is an increasing need to find effective tools for the purpose.

In this background, this article aims to discuss the feasibility of using artificial

intelligence(AI) for arbitral decision making. This article explores the current use-cases of AI

to lay down the foundation for its use in arbitral decision making. Further, the article discusses

the suitability of an Arbitrator-Robot (ArBot) for the process of arbitration. The article also

discusses the limitations of AI based arbitral decision making in light of its current models and

use-cases, and further, plausible solutions to overcome these shortcomings. Finally, the article

concludes that the use of AI in arbitral decision making will cater to the changing expectations

of the consumers of legal services. If access to justice can be provided in a demonstrably cost

and time effective manner, the market can be expected to opt for such an alternatives.


Attorney-at-law, LL.M., 2020 (Stanford University). Former Senior Associate at L&L Partners Law Offices,
Delhi, India.

Electronic copy available at: https://ssrn.com/abstract=3815946

 Contents

1. Introduction……………………………………………………………..……………2

2. Evolution of Online Dispute Resolution and the need for AI tools………………….4

3. Artificial Intelligence Based Decision Making….……... …………………………...8

4. Artificial Intelligence Based Arbitrator-Robot: ArBot...............................................14

Notice of Arbitration……………………………………………………………..15

Appointment of ArBot…………………………………………………………...16

Data processing and communication……………………………………….........18

Rendering of Award…………………………………………………………...…19

5. Challenges to ArBot in the Current Legal and Technical Landscape..………..……21

Lacks emotions and empathy……………………………………………...……..21

Lack of reliable data sets: Confidentiality in Arbitration…………......................23

Need for reasoned awards………………………………………………..………24

Replicating existing biases……………………………………………………….26

Costs involved in formulating AI decision making model…………………........28

6. Conclusion……………………………………………………………….……….…29

Electronic copy available at: https://ssrn.com/abstract=3815946

 I. Introduction

John von Neumann1, popularly known as the founding father of computer science and

mathematics, once said: "You insist that there is something a machine cannot do. If you will

tell me precisely what it is that a machine cannot do, then I can always make a machine which

will do just that!" 2. He made this statement in 1948, and ever since, technology has advanced

multi-folds, depending on the growing and varying demands of humans. Today, decisions vital

to our welfare and freedom are made using artificial intelligence (AI) that improves with data,

commonly known as Machine Learning systems. 3 Our day to day decisions are becoming

increasingly reliant on the recommendations made by machine learning algorithms when

compared to those made using human intelligence. Perhaps, the algorithms have indeed become

smarter than humans in understanding human needs and preferences in music, food, books, or

for that matter, even spouses. Almost every sphere of human life is penetrated by algorithmic

decision making.

The world of law is not far behind. AI already touches upon many areas of law,

including contract analysis, legal research, e-discovery, etc.4 With constant advancements in

technology fuelling the growth of AI, it is increasingly being used by courts in management of

1 John Von Neumann was a Hungarian-American mathematician, physicist, and a computer scientist. Von
Neumann's work in areas such as game theory, mathematics, physics, and meteorology formed the building
blocks for the most important discoveries of the century: the modern computer, game theory, the atom bomb,
radar, and artificial intelligence, to name just a few. See NORMAN MACRAE, JOHN VON NEUMANN: THE
SCIENTIFIC GENIUS WHO PIONEERED THE MODERN COMPUTER, GAME THEORY, NUCLEAR DETERRENCE, AND
MUCH MORE, (1992).
2
E.T JAYNES, PROBABILITY THEORY: THE LOGIC OF SCIENCE, 7 (2003). The author claims to have been present
when John Von Neumann gave a talk in Princeton University in 1948 and said that.
3
Lillian Edwards & Michael Veale, Slave to the Algorithms? Why a right to explanation is probably not the
remedy you are looking for, 6 Duke L. Tech. Rev. 18, 19 (2017).
4
See RICHARD SUSSKIND, TOMORROW’S LAWYERS: AN INTRODUCTION TO YOUR FUTURE, (2nd ed., 2017).

Electronic copy available at: https://ssrn.com/abstract=3815946

cases5, and by legal practitioners in strategic case planning based on algorithmic predictions. 6

Contrary to the general perception that legal decision making is beyond the domain of

computers, the impact of AI tools, such as, predictive coding and natural language processing,

is pervading legal decision making. In the U.S., these techniques are already used in criminal

justice system to set bail, determine sentences, and even contribute to determinations about guilt

or innocence. 7 In Mexico, the AI system is currently advising judges and clerks ‘upon

determination of whether the plaintiff is or is not eligible for granting him or her a pension’. 8

The world of arbitration is not unperturbed by algorithmic disruption either. AI is

already in use by different stakeholders in arbitration for performance of tasks, such as

appointment of arbitrator, legal research, drafting and proof reading of written submissions,

translating of documents, case management and document organisation, cost estimations,

hearing arrangements and drafting of standard sections of awards. 9 This paper aims to discuss

AI powered legal decision making in arbitration using technologies such as machine learning,

natural language processing, and quantum computing with a view to initially supplement and

eventually replace human arbitrators in small-claim disputes. In the second section of the paper,

I discuss the evolution of Online Dispute Resolution (ODR) and the need for deploying AI

5
AI can leverage more data than the human brain could assess at any given point. From advances in data
processing speeds, lower costs, big data volume, to the integration of data science into technology, AI has
become the perfect assistant in Case Management. See James Bailey, Top Uses of Artificial Intelligence in Case
Management, ARMEDIA BLOG (Oct. 29, 2018), https://www.armedia.com/blog/top-uses-ai-in-case-
management/.
6
The advances in Natural Language Processing and Machine Learning provide us with the tools to
automatically analyze legal materials, so as to build successful predictive models of judicial outcomes. See
Nikolaos Aletras et al., Predicting judicial decisions of the European Court of Human Rights: a Natural
Language Processing perspective, PEERJ COMPUTER SCIENCE, (Oct. 24, 2016), https://doi.org/10.7717/peerj-
cs.93.
7
Algorithms in the Criminal Justice System: Risk Assessment Tools, ELECTRONIC PRIVACY INFORMATION
CENTER, https://epic.org/algorithmic-transparency/crim-justice/ (last visited Nov. 25, 2020).
8
Tania Sourdin, Judge V Robot? Artificial Intelligence and Judicial Decision Making, 41(4) UNSW Law
Journal 1114, 1119 (2018).
9
Maxi Scherer, Artificial Intelligence and Legal Decision-Making: The Wide Open? Study on the Example of
International Arbitration, Queen Mary School of Law Legal Studies Research Paper No. 318/2019
https://poseidon01.ssrn.com/delivery.php?ID=08911910506400412010402010406708712402204603902104205
506408612509609807511710910302806303300000810901602609211702210002312408600604106204603600
808402811311211210909102903002110008412201700611506411111402208903109803010110312000601001
6090111072071123024&EXT=pdf.

Electronic copy available at: https://ssrn.com/abstract=3815946

powered ODR for small claims cases. In the third section, I provide an overview of a few

successful AI models being used in decision making, and how they lay down the foundation for

AI based legal decision making in arbitration. In the third section of this paper, I will discuss

the core characteristics of an AI based arbitrator from the perspective of traditional mechanisms

of arbitral decision making, such as appointment of arbitrator, review of documents, and

rendering of award. In the next section, I focus on the technical and legal complexities involved

in launching of AI based arbitrator. I also discuss how technologies such as explainable machine

learning and quantum computing may provide plausible solutions for the conduct of an

autonomous arbitration. The last section is the conclusion.

II. Evolution of Online Dispute Resolution and the need for AI tools.

Decades back, when the personal computer was invented, it was beyond imagination that

years down the line, the personal computer would become the basis for conducting business

and transacting across borders. While the internet had been invented in 1969, during the first

twenty years of its existence there were relatively few disputes.10 Its prime users during the

period were in academia and the military, and whenever there were disputes in the relatively

small user group, they were settled informally. 11 Gradually, with the invention of the World

Wide Web, the number of internet users began to grow with a growth in the online interactions

and transactions. It began to be clear that cyberspace, in the future, would not be a harmonious

place and there would be a need for tools, resources and expertise in responding to the disputes

that would occur. 12 In 1996, the National Centre for Automated Information Research

10
Ethan Katsh, ODR: A Look at History: A Few Thoughts About the Present and Some Speculation About the
Future, in ONLINE DISPUTE RESOLUTION: THEORY AND PRACTICE, A TREATISE ON TECHNOLOGY AND DISPUTE
RESOLUTION, 21, 21 (Mohamed S. Abdel Wahab et. al. eds., 1st ed., 2012). During this period of time, few
ordinary citizens were aware of the Internet and if they had been aware, they would have had no way to connect
to it since the first Internet Service Provider did not appear until 1992.
11
Id.
12
Id. at 23.

Electronic copy available at: https://ssrn.com/abstract=3815946

(NCAIR) sponsored the first conference devoted to ODR, and funding from NCAIR launched

the first significant projects aimed at ODR: the Virtual Magistrate, the Online Ombuds Office

at the University of Massachusetts and a family dispute ODR project at the University of

Maryland.13

Today, there are a number of disputes stemming from online activities. Commerce is

no longer dependent on in-person interaction between the buyer and the seller. Advancements

in information and communication technologies have significantly altered the way business is

conducted, and have led to ever increasing use of electronic instead of paper based means of

communication and data storage. 14 This has led to the emergence of a digitized global

economy, and businesses are witnessing immense growth based on e-commerce beyond

national borders. E-commerce portals like eBay and Amazon are the steady examples of how

parties from different parts of the world enter into e-sale contracts without having the need for

their representatives to meet in person. In fact, technology makes it possible for buyers to

download the digitized version of the goods without expecting a physical delivery of the same.

Even better, in some instances, a 3D printer may be used for the process of making three

dimensional solid objects from the digital file so received.15 This has improved the economic

efficiency in conducting business, enhancing their profitability. More business means more

disputes. The process of litigating disputes is notorious for undue delays, comprehensive and

inflexible procedures, and high costs. Thus, there is an overwhelming demand for more cost

effective and less time consuming mechanisms for dispute resolution of small and medium

sized claims.

13
Id.
14
Mohamed S. Abdel Wahab, ODR and e-Abritation: Trends & Challenges, in ONLINE DISPUTE RESOLUTION:
THEORY AND PRACTICE, A TREATISE ON TECHNOLOGY AND DISPUTE RESOLUTION, 399, 399 (Mohamed S. Abdel
Wahab et. al. eds., 1st ed., 2012).
15
The creation of a 3D printed object is achieved using additive processes. In an additive process an object is
created by laying down successive layers of material until the object is created. Each of these layers can be seen
as a thinly sliced horizontal cross-section of the eventual object. See, What is 3D printing?,
https://3dprinting.com/what-is-3d-printing/ (last visited Nov.28, 2020).

Electronic copy available at: https://ssrn.com/abstract=3815946

 Resolution of cross-border disputes is complicated because connecting factors, such as,

the place of domicile, the place of business, and the place of performance are difficult to

determine, and more so, when the claims involve small sums of money as compared to the size

of the business of the disputing entities. Dispute resolution is even more burdensome for B2C

transactions involving individuals. This has raised serious concerns of access to justice since

many people are excluded from an effective redressal of their grievances. Individuals,

especially those involved in B2C transactions, often refrain from pursuing small claims for the

following reasons: a) if the seller does not have a redressal mechanism in place, implying that

the resolution of dispute would then be entirely up to the mercy of courts, the individual buyer,

in such circumstances, would not want to spend more money on filing a suit; (b) even if a

dispute resolution mechanism is provided for in the sale contract, individuals are sceptical in

pursuing it owing to the costs and time involved in ADR mechanisms. Conflicts arising from

a typical purchase on the internet are almost never resolved in courts because the cost of

bringing such claims outweighs the value of the dispute, especially when parties are located in
16
different jurisdictions. Further, the trend of restraint in invoking dispute resolution

mechanisms has encouraged many business entities and corporations to include arbitration as

the default dispute resolution clause. This deters an individual claimant from pursuing cost and

time intensive arbitration for redressal of small claims. Although cross-border disputes are

suitable for resolution by arbitration, however, the costs and time involved in the process leaves

the aggrieved with no feasible solution.

ODR mechanisms offer the opportunity to widen access to justice by introducing

flexibility and cost effectiveness. This allows the parties to resolve their disputes in a place and

at a time convenient to them. In theory, ODR can be used in most civil and commercial disputes

16
See PABLO C ORTES , THE LAW OF CONSUMER REDRESS IN AN EVOLVING DIGITAL MARKET (2017).

Electronic copy available at: https://ssrn.com/abstract=3815946

ranging from contracts to torts, from family to business, and from domestic cases to

international cases. However, disputes that involve e-transactions or small claims are more

suitable for ODR services as these cases involve online form of order and payment, and

therefore, are largely based on documents that are unimpeachable in nature. These documents

can be readily submitted as evidence to ODR mechanisms, such as, arbitration or mediation.

This also finds support in the development that individual expectations about redress are also

moving online. If the conflicts arise online, it seems logical that they should be resolved

online.17 This is already happening in a number of areas, such as traditional B2C and B2B

disputes in eBay and PayPal. 18 Easy access to justice would enable individuals to develop trust

in the system and would help in building a stronger market place. Further, although use of ODR

is currently aimed at small claim online disputes, but this would also help in laying the

foundation for trust of users in ODR for bigger international disputes.

Rather than finding disputes that can utilize ODR, the new challenge is finding tools

that can deliver trust, convenience and expertise for different kinds of conflicts.19 Therefore,

practitioners are increasingly looking at employing AI based technologies for facilitating and

conducting ODR. For years, parties and practitioners have preferred an in-person conduct of

arbitral proceedings, including performance of miscellaneous administrative tasks. The

emergence of Covid-19 pandemic has impacted dispute resolution in an unforeseen manner.

As a consequence of the containment restrictions imposed by different countries, technology

is now a necessity in the conduct of dispute resolution. With courts functioning at a limited

capacity, the demand for cost effective and time saving tools for dispute resolution has

intensified. Besides, taking dispute resolution online also ensures safety of the participants of

17
Pablo Cortés, Online Dispute Resolution for Consumers: Online Dispute Resolution Methods for Settling
Business to Consumer Conflicts, in ONLINE DISPUTE RESOLUTION: THEORY AND PRACTICE, A TREATISE ON
TECHNOLOGY AND DISPUTE RESOLUTION, 151, 171 (Mohamed S. Abdel Wahab et. al., eds., 1st ed., 2012).
18
Id.
19
Ethan Katsh, The Online Ombuds Office: Adapting Dispute Resolution to Cyberspace,
https://www.umass.edu/dispute/ncair/katsh.htm#fn1 (last visited, Nov. 20, 2020).

Electronic copy available at: https://ssrn.com/abstract=3815946

the process. Arbitration, being one of the most prominent forms of alternate dispute resolution,

has the potential of being transitioned into an ODR mechanism using minimal human

intervention by utilising state-of-the-art technologies that are integrated and embedded into

arbitration proceedings conducted wholly and substantially online.20

Despite the sudden induction of technological solutions for a seamless conduct of

dispute resolution processes, the degree of utilization varies. The degree of utilisation of

technology in ODR may be categorised into three groups: (a) Technology assisted ODR

mechanisms, where technology is used as a secure means of communication and information

exchange; (b) Technology based ODR mechanism where fully fledged application of cutting

edge technology is used to resolve disputes; and (c) Technology facilitated online dispute

prevention guarantees, which help reduce the risk of potential disputes. 21 Although so far,

utilisation of technology in arbitration is restricted to and would fall under category (a), the

potential use of AI in arbitration to substitute the human arbitrator would fall within the

category (b). In its current form, the use of ODR is the “Fourth Party” which primarily assists

the neutral third party, but is likely to encroach upon the role of the third party just as ADR

encroached upon more formal dispute resolution process.22 Therefore, it is relevant to explore

whether and how AI can be used to eliminate the need for human arbitrator in online arbitration

and the issues associated with the proposition.

III. Artificial Intelligence Based Decision Making

20
Wahab, supra note 14, at 399, 400. Arbitration, as the most prominent form of traditional out-of-court
process, is widely used and globally recognized and applied owing to its default binding nature, finality, global
regulation, and enforceability of arbitral awards. Accordingly, taking arbitration online by utilizing state-of-the-
art technologies, that are integrated and embedded into arbitral proceedings conducted wholly or substantially
online, is a necessary evolutionary phase that matches the transition to a paperless world.
21
Id. at 402.
22
Id.

Electronic copy available at: https://ssrn.com/abstract=3815946

 AI, in its current form, is capable of exhibiting superior powers of comprehension and

decision making. Consider the following news snippet that appeared on the Los Angeles Times

website on 18 March 2014:

A shallow magnitude 4.7 earthquake was reported Monday morning

five miles from Westwood, California, according to the U.S. Geological

Survey. The temblor occurred at 6:25 a.m. Pacific time at a depth of 5.0

miles.

According to the USGS, the epicenter was six miles from Beverly Hills,

California, seven miles from Universal City, California, seven miles

from Santa Monica, California and 348 miles from Sacramento,

California. In the past ten days, there have been no earthquakes

magnitude 3.0 and greater centered nearby.

This information came from the USGS Earthquake Notification Service and this post

was created by an algorithm written by the journalist and programmer Ken Schwencke.23 His

algorithm automatically generates a short article when an earthquake occurs. 24 This post

appeared within three minutes of the earthquake. Journalism is not the only field permeated by

algorithmic cognitive abilities. IBM’s ground breaking Watson computer, which beat a series

of champions in a popular American game show in 2011, is now being used to diagnose

medical conditions.25 The Language Technology group at the University of Hamburg with the

23
Ken Schwencke, a journalist and computer programmer, left the Los Angeles Times in 2014. He previously
developed news applications for latimes.com and conducted analysis for reporting projects.
24
Robot writes LA Times earthquake breaking news article, BBC NEWS (Mar. 18, 2014)
https://www.bbc.com/news/technology-26614051.
25
Lauren F. Friedman, IBM's Watson Supercomputer May Soon Be The Best Doctor In The World, BUSINESS
INSIDER (Apr. 22, 2014, 7:44 PM) https://www.businessinsider.com/ibms-watson-may-soon-be-the-best-doctor-
in-the-world-2014-4.

Electronic copy available at: https://ssrn.com/abstract=3815946

Bucerius Law School are collaborating with IBM’s Watson Knowledge Studio to analyze court

decisions, and offers a faceted search interface “LawStats” to aid law practitioners. 26

We fail to acknowledge that computers have become capable of more sophisticated

tasks, and even if we do grudgingly acknowledge the fact, we find it hard to believe that a

computer could perform typically human tasks as well as humans. 27 But the emergence of big

data has allowed a significant shift in the development of AI, and AI is progressing towards

inculcating human intelligence. Rather than developing complex algorithms for cognitive

processes, AI is being used to learn from existing data.28 This subfield of AI research is called

machine learning, which is concerned with computer programs that learn from experience and

improve performance over time.29 Machine Learning, in combination with Natural Language

Processing, can sift through bundles of unstructured legal data to identify complex statistical

relationships within a dataset and predict the outcome. This model is being deployed globally

to find solutions to the otherwise mechanical and time consuming tasks in the domain of human

intelligence.

An example of how such technologies could cause disruption in an industry dominated

by human intelligence is robot advising in financial investment. These platforms provide

tailored financial advice by applying algorithms that analyze and make decisions based on

customer provided information relating to financial circumstances and goals. 30 An average

American lacks the basic financial and investing knowledge 31, and robo-advice industry seizes

upon opportunities presented by digital investors who may be financially inexperienced and

26
Eugen Ruppert et al., LawStats – Large-scale German Court Decision Evaluation using Web Service
Classifiers, Dept. of Computer Science, Universit ̈at Hamburg, Germany (2018) https://www.inf.uni-
hamburg.de/en/inst/ab/lt/publications/2018-ruppertetall-cdmake-lawstats.pdf.
27
Paul Cohen & Sophie Nappert, The march of the robots, GLOBAL ARBITRATION REVIEW (Feb. 15, 2017)
https://globalarbitrationreview.com/article/1080951/the-march-of-the-robots.
28
Scherer, supra note 9, at 6.
29
STUART J. RUSSELL & PETER NORVIG, ARTIFICIAL INTELLIGENCE: A MODERN APPROACH, 693 (3rd ed., 2018).
30
Nicole G. Iannarone, Computer as Confidant: Digital Investment Advice and the Fiduciary Standard, 93
CHIC. KENT L. REV. 141, 151 (2017).
31 Id. at 143.

10

Electronic copy available at: https://ssrn.com/abstract=3815946

wary of high fees. Robo-advisory firms differ from traditional human investment advisors

because they generally do not have a high account minimums for fees and charge fees that are

substantially less than human investment advisors.32 Robo-advisors can also offer conflict-free

advice based on the information they receive from the end users, i.e., the investor customers.

While most cognitive aspects of the legal practice remain outside the domain of AI’s

capabilities, it is increasingly being employed in ancillary tasks performed by law

professionals. The use cases of AI have now percolated the areas of discovery, research, case

management, due diligence, and outcome predictions – to name a few. For instance, IBM’s

Ross Intelligence understands and interprets the legal issue, relevant facts, and procedural

posture fed in natural language.33 It then uses AI to interpret the query, read its entire database

of all U.S. federal and state cases, and finds all the most relevant case law for the research

question.34 Support tools used in legal research, such as LexisNexis’ LegalAdvance, include a

range of AI tools in different parts of the research process. 35 Kira 36 , a contract analysis

platform, uses machine learning and natural language processing tools to automatically

highlight and extract provisions that are important and helps in organizing the data for

analysis.37 The company asserts that that its software is capable of performing a more accurate

due diligence contract review by searching, highlighting, and extracting relevant content for

32
Id. at 141.
33
ROSS quickly finds all the most relevant case law for your research questions — often finding diamonds in
the rough that traditional research tools miss. This is because you can describe your legal issue, relevant facts,
and procedural posture in natural language, rather than relying on clunky Boolean terms or keywords. ROSS
Intelligence Offers A New Take On Legal Research, Above the Law (29 May 2019, 5:15 PM),
https://abovethelaw.com/2019/05/ross-intelligence-offers-a-new-take-on-legal-research/.
34
Id.
35
John Armour, et. al., Augmented Lawyering (Aug. 21, 2020), https://dx.doi.org/10.2139/ssrn.3688896.
36
Kira Systems help enterprises make meaning from their unstructured contracts and related documents. It
enables enterprises to easily find and use trusted information and answers from their documents. See,
https://kirasystems.com (last visited Nov. 28, 2020).
37
Id.

11

Electronic copy available at: https://ssrn.com/abstract=3815946

analysis, as compared to junior lawyers.38 Everlaw39, an e-discovery portal, uses AI to look

into the contents and metadata to classify documents.40

Many start-ups aiding in the conduct of arbitration have found success with

practitioners around the globe. ArbiLex41, a Harvard Law School legal tech start-up, is one

such example. ArbiLex leverages Bayesian Machine Learning 42 to offer an empirical, coherent,

and explainable framework for counsel to quantify uncertainties, reduce errors, and maximize

desirable outcomes in high-stakes international arbitration cases in a cost-effective manner.43

The product complements the function of human intelligence in making strategic calls related

to global disputes involving arbitration. Global law firms have deployed ArbiLex’s proprietary,

explainable AI solutions, to optimize strategy in a portfolio of cases worth more than $3

billion.44

A corollary to this would be the use of algorithmic legal decision making. And while

we may have predicted this to be worlds away from happening in the near future, it is happening

already. In State v. Loomis45, a judge used AI to assess an appropriate sentence. 46 AI is also

38
Daniel Faggella, AI in Law and Legal Practice – A Comprehensive View of 35 Current Applications, Emerj
(March 14, 2020), https://emerj.com/ai-sector-overviews/ai-in-law-legal-practice-current-applications/.
39
Everlaw is a cloud-based platform that unlocks the collaborative power of litigation and investigative teams.
See, https://www.everlaw.com/product/# (last visited Nov. 28, 2020). Everlaw uses its predictive coding feature
to create prediction models based on at least 300 documents that were classified before as relevant or irrelevant
by the user. See, Faggella, supra note 38.
40 Faggella, supra note 38.
41
See, https://www.arbilex.co/welcome (last visited Nov. 28, 2020).
42
Bayesian ML differs from traditional ML algorithms as probabilities determined in the beginning,
intermediate, and final output of a model rely on initial, reasonable guesses of an event happening, called the
prior, instead of only building on the observed frequency of an event occurring. A Bayesian approach to
machine learning is preferred for this particular problem because one can leverage an expert’s opinion to
quantify the prior probability of a given factor in a case. An informed prior is critical to the success of getting a
sensible posterior probability, or a probability that coherently adjusts the prior beliefs from limited past data.
Frederick Daso, ArbiLex, A Harvard Law School Legal Tech Startup, Uses AI To Settle Arbitrations, Forbes
(Feb. 4, 2020, 09:07 AM), https://www.forbes.com/sites/frederickdaso/2020/02/04/arbilex-a-harvard-law-
school-legal-tech-startup-uses-ai-to-settle-arbitrations/#435a5cd452c5.
43
https://www.arbilex.co/welcome (last visited Nov. 28, 2020).
44
Jodi Goldstein, Harvard Innovation Labs 2019 Year in Review, Harvard Innovation Labs (Dec. 17, 2019),
https://innovationlabs.harvard.edu/about/news/2019-year-in-review/.
45
State of Wisconsin v. Eric L. Loomis, 881 N.W.2d 749 (Wis. 2016).
46
Though subsequently, it was held by the Supreme Court that the judge used the robot as a check rather than
actually delegating the decision to it. See, Loomis v. Wisconsin, 881 N.W.2d 749 (Wis. 2016), cert. denied, 137
S.Ct. 2290 (2017).

12

Electronic copy available at: https://ssrn.com/abstract=3815946

routinely used in bail hearings in the US to predict the risk of absconding. 47 In 2016,

researchers at University College London, the University of Pennsylvania, and University of

Sheffield developed AI software which analyzed the language used in the submissions and

previous judgments to predict the outcome of European Convention on Human Rights litigation

in 79% of the cases.48 In fact, a recent report by Carl Frey, co-director of the Oxford Martin

Programme on Technology and Employment at Oxford University, highlights how exponential

improvements in machine learning mean highly trained workers are at risk of having their jobs

automated, with judges and magistrates listed as at medium risk (40.1%) along with detectives,

economists, computer programmers, commercial pilots and personal financial advisers. 49

The incursion of robots into mediation hit a new milestone in February 2019, as

Smartsettle One, an AI tool, replaced a human mediator and, in less than an hour, settled a three

month dispute of over $2000 unpaid bill for a personal counselling course. 50 The system allows

parties to make offers and counter offers by moving flags along sliders – a green one seen by

the other side and a yellow flag which is not. The yellow flag is effectively a blind bid. 51 The

algorithms learn the bidding tactics and priorities employed by the parties and help nudge them

into a settlement, without revealing their secret bids.52

There is no doubt that algorithms and big data are fast entering the confidentiality bound

world of alternative dispute resolution. What began with the use of internet for the purpose of

research and sharing of documents has now come to the use of sophisticated technologies such

as smart contracts, blockchain, AI, and quantum computing for facilitating resolution of

47
Karen Maxwell, Summoning the demon: robot arbitrators: arbitration and artificial intelligence, THOMPSON
REUTERS ARBITRATION BLOG (Jan. 17, 2019), http://arbitrationblog.practicallaw.com/summoning-the-demon-
robot-arbitrators-arbitration-and-artificial-intelligence/.
48
Id.
49
Cohen & Nappert, supra note 27.
50
Kate Beioley, Robots and AI threaten to mediate disputes better than lawyers, myFT (13 Aug 2019),
https://www.ft.com/content/187525d2-9e6e-11e9-9c06-a4640c9feebb.
51
Nick Hilborne, Robot mediator settles first ever court case, (19 Feb 2019),
https://www.legalfutures.co.uk/latest-news/robot-mediator-settles-first-ever-court-case.
52
Id.

13

Electronic copy available at: https://ssrn.com/abstract=3815946

disputes. A properly equipped computer can digest new data, look for patterns, and make

predictions and recommendations. 53 With each passing day, more and more tasks in the domain

of human intelligence are being taken over by computers. With the progress of AI tools such

as noted above, it is conceivable that on the current trajectory, a computer would be able to

serve as a fact finder and arbitrator on its own. 54 Such a computer would sift through the

applicable law more thoroughly, assess the credibility of witness evidence more accurately,

and deliberate more quickly than human arbitrators.55

The author is not asserting that a machine of this calibre exists as of now. But with the

current progression, especially with the advent of quantum computing, it is foreseeable that a

machine will develop the ability to perform multiple tasks simultaneously, thereby reducing

the need for human intervention. 56

IV. Artificial Intelligence Based Arbitrator-Robot: “ArBot”

Arbitration is a process of dispute resolution by which the parties refer their disputes to

a neutral third party. The conduct of an arbitration involves the neutral undertaking a number

of tasks. The neutral reviews the facts of the case and the evidence produced by the parties,

applying procedural rules agreed upon by the parties and/or mandated by the lex arbitri,

applying substantive rules of law as selected by the parties or directed by the lex arbitri, and

53
Paul Bennett Marrow, et. al., Artificial Intelligence And Arbitration: The Computer as an Arbitrator-Are We
There Yet?, 17(4) Disp. Resol. Jour. 35, 38 (2020).
54
With the progress of artificial intelligence programmers such as Watson, it is conceivable that on the current
trajectory, a computer would be able to serve as a fact finder and arbitrator on its own within the next two
decades, if not much sooner. Again, that is a conclusion that would have seemed inconceivable until very
recently. Sophie Nappert & Paul Cohen, Chapter 6: Case Study, The Practitioner’s Perspective in Arbitration,
in THE DIGITAL AGE: THE BRAVE NEW WORLD OF ARBITRATION, 126, 142 (Maud Piers & Christian Aschauer
eds. (2018).
55
Id. at 143.
56
AI technology could become “smarter” and more accurate by gaining access to previously unheard-of dataset
sizes, which the quantum machine could help it process. This could be a case where two technologies could
become stronger when deployed together. John Breeden, Will Quantum Computing Supercharge Artificial
Intelligence? (Nov. 4, 2020), https://www.nextgov.com/ideas/2020/11/will-quantum-computing-supercharge-
artificial-intelligence/169757/.

14

Electronic copy available at: https://ssrn.com/abstract=3815946

performing miscellaneous tasks. 57 In deliberating whether AI can take over human arbitral

decision making, it becomes pivotal to illustrate whether an ArBot would be capable of filling

in the shoes of a human arbitrator in the context of the traditional meaning of arbitration.

Following is a detailed analysis of whether and how an ArBot could undertake various tasks

that an arbitrator performs or oversees the performance of, and the legal sanctity of such

performance:

(i) Notice of Arbitration

The first and the foremost step in the arbitral process is invocation, by a party to the

contract, of the clause containing an arbitration agreement between the parties. Generally, the

invocation is governed by the steps enumerated in the arbitration agreement itself. Article 3(4)

of the UNCITRAL Arbitration Rules58 provides that the notice of arbitration may include the

proposal for the appointment of a sole arbitrator or an appointing authority. For invoking an

arbitration for the appointment of an ArBot, a party can, thus, send an electronic notice

containing the demand for a reference to the other party, as well as, to the administrator of AI

based arbitration. Technologies such as blockchain may be employed for the purposes of

encryption and cybersecurity.

Before referring the dispute to an ArBot, it would be relevant that there is an agreement

between the parties to refer the disputes to a machine rather than a human neutral. While the

existence of a prior agreement between the parties to refer the dispute to a machine would be

ideal, it may not be necessary. A submission agreement 59 could state the intent of the parties.

57
Horst Eidenmüller & Faidon Varesis, What is an Arbitration? Artificial Intelligence and the Vanishing
Human Arbitrator, 23 (Jun. 17, 2020) https://dx.doi.org/10.2139/ssrn.3629145.
58
United Nations Commission on International Trade Law (UNCITRAL)
Arbitration Rules, General Assembly Resolution 31/98 adopted by the General Assembly on 15 December
1976, https://www.uncitral.org/pdf/english/texts/arbitration/arb-rules/arb-rules.pdf.
59
A submission or post-dispute arbitration agreement is entered into by parties to an existing dispute. In the
submission agreement, the parties agree to submit only a specific dispute to arbitration. They can do so at the
time the dispute arises, while the parties are engaged in negotiations for a resolution, or even if the dispute is
already being actively litigated in court. Laura MacDonald & Elizabeth A. Edmondson, Submission Agreements
- How to Join the Pandemic Arbitration Party Even If You Weren’t Invited (Jun. 29, 2020),
https://www.lexology.com/library/detail.aspx?g=6c9325f5-9b83-47cd-8ec4-7744df75707e.

15

Electronic copy available at: https://ssrn.com/abstract=3815946

A major drawback here would be that parties, especially the ones with huge stakes in the

dispute, could be reluctant in having a machine helm their dispute resolution. The author

believes that a shift in the mindset of the parties to demonstrate trust in an ArBot will be

dependent on the accuracy and effectiveness of AI based decision making.

(ii) Appointment of Arbitrator

The “Appointment of Arbitrator” as provided under Article 6 of the UNCITRAL

Arbitration Rules does not envisage the possibility of referring a dispute to a machine, as it

requires that “..the appointing authority shall have regard to such considerations as are likely

to secure the appointment of an independent and impartial arbitrator and shall take into account

as well the advisability of appointing an arbitrator of a nationality other than the nationalities

of the parties.” The rule contemplates a natural person who could be independent and impartial

as an arbitrator, and would belong to certain nationality. Administrator rules also use the term

“persons” or “individuals” when referring to an arbitrator. 60 However, arbitration is conducted

within the four corners of party autonomy, and the parties might as well agree to designate a

computer as an arbitrator. 61

While the procedural rules may be modified and agreed upon by the parties, the will of

the parties cannot amend the substantive law applicable to the arbitration. For instance, the

French Law on Arbitration issued by virtue of Decree No. 2011-48 of 13 January 2011

explicitly endorses the requirement for an independent and impartial arbitrator.62 The English

Arbitration Act 1996 deals with the death of an arbitrator, thereby implying that an arbitrator

60
Marrow, supra note 53, at 74.
61
Id.
62
In principle, it is worth noting that almost all existing rules and laws envisage arbitrators as humans and
requires them to possess the necessary capacity, impartiality, independence, and ability to decide a case. The
most recent French Law on Arbitration issued by virtue of Decree No. 2011-48 of 13 January 2011 explicitly
endorses such requirement and universal presumption that an arbitrator cannot be but a human. Wahab supra
note 14, at 420.

16

Electronic copy available at: https://ssrn.com/abstract=3815946

shall be human. Such requirements are shared by many other countries. States worldwide

would need to step in to reconcile such laws to include an ArBot within the definition of

“arbitrator”.

In the past, many states have shown a tendency towards adapting to developments in

arbitral practices by making amends to existing statutes and legislations. The amendment of

the Singapore Arbitration Act63 for inclusion of “emergency arbitrator” within the definition

of “arbitral tribunal” is an illustration of the State’s intent of providing the requisite legal

framework.64 Many other states followed suit. 65 Similarly, adapting to technological advances

in arbitral processes would also depend on the appetite of the State towards having an

arbitration friendly approach.

Once the statutes become amenable to appointment of an ArBot, the next requisite

would be to see that the ArBot meets the desired criteria of independence and impartiality. It

should be easier to achieve, as it is relatively limited set of factual and legal prescriptions which

need to be considered and observed. 66 For instance, Rule 2.3.8 of the IBA Guidelines on

Conflicts of Interest in International Arbitration (2014), that have been adopted by many states

in their arbitration laws, provides that the arbitrator must not have a close family relationship

with any of the parties. Therefore, issues such as conflict of interests or other inherent biases

that entail a human element would not put an embargo on the appointment of an ArBot.

63
Singapore Arbitration Act 2001, Original Enactment: Act 37 of 2001 https://sso.agc.gov.sg/Act/AA2001#xv
(last visited Nov. 29, 2020).
64 Section 2(1), Singapore Arbitration Act, 2001- In this Act, unless the context otherwise requires — “arbitral

tribunal” means a sole arbitrator or a panel of arbitrators or an arbitral institution, and includes an emergency
arbitrator appointed pursuant to the rules of arbitration agreed to or adopted by the parties including the rules of
arbitration of an institution or organization.
65
Hong Kong has made amendment to its Arbitration law, which provides that courts may grant leave to enforce
relief granted by emergency arbitrators, both inside and outside Hong Kong, so long as the relief falls within one
of the types of interim order listed therein. Arbitration Ordinance 2011, c 609, Arts 22A and 22B, as amended
by the Arbitration (Amendment) Ordinance 2013 with effect from 19 July 2013.; Emergency arbitrators' orders
or directions shall be enforceable in the same manner as court judgments, New Zealand Arbitration Act 1996,
Art 2(1), as amended on 17 October 2016, with effect from 1 March 2017.
66
Eidenmüller & Varesis, supra note 57, at 21.

17

Electronic copy available at: https://ssrn.com/abstract=3815946

(iii) Review of Documents: Data Processing and Communication

E-proceedings normally commence by filing an e-request for arbitration and would

involve e-hearings (audio and/or video conferencing), e-submissions and e-production of

documents/evidence, e-deliberations, and/or e-communications (whether inter-party, inter-

tribunal, and/or party-tribunal), and would normally end by an e-award. 67 Review of

documents and communications form integral parts of the arbitral proceedings starting from e-

filing until the rendering of the e-award. The legal world is not oblivious to the use of Machine

Learning to review electronic documents for the purposes of e-discovery. The abundance of

information that is typically available in relation to a dispute promotes the use of technology

for cost effective discovery processes. While e-discovery, in its current form, may not be apt

for large scale disputes where the information may be too specific and mandating the training

of a new Machine Learning model for every dispute to which the technique is applied. 68 In

small and medium sized disputes, such as, e-commerce disputes, or simple contractual disputes,

the evidence comprises of documents generally found in electronic form and are readily

available to be submitted before the tribunal for evaluation. More often than not, these are

standard documents evincing sale, purchase, receipt, delivery of articles, etc. Such basic level

of evaluation and comprehension based on Machine Learning and Natural Language

Processing may not be too much to ask for from sophisticated AI in simple small claims

disputes.

Encryption technologies such as blockchain are already being used by many entities

across the globe for e-transfer or e-submission of important documents. Such technologies may

also be introduced in AI based arbitration models so as to ensure cybersecurity in the arbitral

process.

67 Wahab, supra note 14, at 412.

68
Armour, supra note, at 21.

18

Electronic copy available at: https://ssrn.com/abstract=3815946

(iv) Rendering the Arbitral Award

The most important function of an arbitrator is rendering of an arbitral award, which is

final and binding on the parties to the arbitration. Rendering an award involves applying the

applicable substantive laws to the facts of the case. 69 Drawing from several examples of AI

based legal decision-making in the foregoing paragraphs, it may possible to train a computer

to render decisions after applying the relevant rules and laws on a particular fact set. However,

arbitral awards so rendered must generally be in writing and duly signed by the arbitral

tribunal.70 For instance, article IV(1)(a) of the New York Convention on the Recognition and

Enforcement of Foreign Arbitral Awards (1958) requires that the party seeking recognition

and/or enforcement of the foreign award produce the ‘duly authenticated original award or duly

certified copy’ thereof. 71 An award rendered by an ArBot would need to comply with the

requirements stated above in order to be binding and enforceable.

In this regard, it is relevant to take note of Article 9(2) of the UN Convention on the

Use of Electronic Communications in International Contracts (2005), which states: ‘Where the

law requires that a communication or a contract should be in writing, or provides for the

consequences for the absence of a writing, that requirement is met by an electronic

communication if the information contained therein is accessible so as to be usable for

subsequent reference.’ Mohamed S. Abdel Wahab, in his article entitled ‘ODR and E-

Arbitration’72, has interpreted the word ‘accessible’ to imply that information in the form of

computer data should be readable and interpretable, and that the software that might be

necessary to render such information be retained. He also interprets the word ‘usable’ to be

intended to cover both human and computer use. By applying this to the form of arbitral award

69
Eidenmüller & Varesis, supra note 57, at 25.
70
Wahab, supra note 14, at 424
71
Id.
72
Id.

19

Electronic copy available at: https://ssrn.com/abstract=3815946

that may be rendered by an ArBot, it may be deduced that the award would qualify as ‘writing’

insofar as it is accessible to and interpretable by humans. (This may be possible with the advent

of explainable AI and quantum computing, which are discussed in detail the next section of the

paper).

As far as authentication and formalization of the award is concerned, it may be noted

that although many countries may require a paper award that is manually signed by the

arbitrator, certain national laws now recognize e-signature and the process of validating e-

awards. For example, in the U.S., the Revised Uniform Arbitration Act of 28 August 2000

states under Article 19: ‘An arbitrator shall make a record of an award. The record must be

signed or otherwise authenticated by any arbitrator who concurs with the award.’ 73 The

modified requirement of authentication may be easily complied with by an ArBot.

Additionally, award scrutiny may be an important feature to be applied to awards

rendered by an ArBot. The function of scrutiny is a unique selling point of some arbitral

institutions. For instance, The International Chamber of Commerce Rules provide that the ICC

Court must approve all awards as to their form. 74 In an ICC arbitration, scrutiny is a key

element ensuring that arbitral awards are of the highest possible standards and thus less

susceptible to annulment in the national courts than they might otherwise be.75 The scrutiny

process provides the parties with an additional layer of protection that would not otherwise be

available, since arbitral awards are generally not subject to appeal. 76 If scrutiny is made

applicable to award rendered by an ArBot, it would ensure that such award is in compliance

with the requirements of an arbitral award in form, as well as, in substance. In fact, since there

73
Id. at 425
74
International Chamber of Commerce Arbitration Rules, as of 1 March 2017 https://iccwbo.org/dispute-
resolution-services/arbitration/procedure/ (last visited Jan. 20, 2020).
75
Afzalian Iman, The scrutiny of the award by the ICC Court of Arbitration,
https://www.memoireonline.com/03/10/3232/Scrutiny-of-the-award-by-the-ICC-court-of-arbitration.html (last
visited Nov. 29, 2020).
76
Id.

20

Electronic copy available at: https://ssrn.com/abstract=3815946

cannot be any allegations of fraud by the parties or mishandling by the tribunal, the chances of

an award being set aside on procedural lapses is minimal. 77

V. Challenges To An ArBot In The Current Legal And Technical Landscape

(i) Lack of Emotions and Empathy

Emotions act as a source of information, cause of motivation and influence information

processing by coloring our perception, memory encoding and judgments. 78 Critics of

autonomous arbitration believe that computers, unlike humans, cannot spontaneously feel

emotions, because they can’t recognize nor understand cues such as facial expression, gestures,

and voice intonation. 79 No susceptibility to emotions and an inherent lack of empathy is

considered to be a huge limitation for AI. As a consequence, an ArBot may inherently be

designed with the same limitation. A question thus arises, how relevant is the emotional

quotient to qualify as an arbitrator?

Many academicians have expressed their scepticism regarding the substitution of

human decision maker by AI. It is believed that the human element ensures empathy and

emotional justice in legal decision making, and an emotionless arbitrator without empathy and

the ability to explain itself would not be able to fully understand the drama of the parties, their

intent and the provided meaning besides the written text of the contract and documents.80 The

counter argument to this could be that emotions and empathy are a part of the form and style

of advocacy that may be designed to appeal to a human arbitrator’s sympathies and sub-

77
Madhvendra Singh, Blockchain & AI: Is the Cocktail Right for Conduct of Arbitration?, BAR & BENCH (Apr.
23, 2020, 3:40 PM) https://www.barandbench.com/columns/policy-columns/blockchain-ai-is-the-cocktail-right-
for-conduct-of-arbitration
78
José María de la Jara et al., Machine Arbitrators: Are we ready?, KLUVER ARBITRATION BLOG (May 4, 2017)
http://arbitrationblog.kluwerarbitration.com/2017/05/04/machine-arbitrator-are-we-ready/.
79
Id.
80
Id.

21

Electronic copy available at: https://ssrn.com/abstract=3815946

conscious biases; the role of the counsel is to draw the arbitrator’s attention to the facts and

law of the cases that are in his or her client’s favour. 81 The ArBot could presumably know and

process the facts without pointers from counsels. Especially, small claims disputes, such as a

consumer dispute, would involve review of minimal documentary evidence, thereby leaving

no scope for emotional consideration by the arbitrator. Thus, in the forms of arbitration where

the arbitrators’ discretionary powers are restricted by choosing either party’s final requests, the

future would yield opportunities for ArBot to engage in an assessment process for both parties’

proposals before rendering a decision in favour of either. 82 Such assessment would be an

objective assessment of proposals without necessitating the involvement of emotions and

empathy of the ArBot. Moreover, an ArBot would be designed using the same training data

that an arbitrator would consider when making a judgment leading to an award. 83

Technology will no doubt catch up and provide solutions. 84 In fact, scientists are

already working to inculcate in a machine the ability to quantify and assess human emotions.

Eonum, a Californian start-up, harnesses the power of quantum computers to make sense of

the data involving a high level of uncertainty, emotional volatility, power imbalances, and

mysterious information. 85 It provides the clients with a more realistic assessment of their

chances and supporting them in the processes of risk assessment, litigation, or arbitration. 86

The use of quantum computing is at a nascent stage in the field of law. At its full potential,

81
Cohen & Nappert, supra note 27.
82
It is not impossible that, at least for some forms of arbitration where the arbitrators’ discretionary powers are
restricted by choosing either party’s final requests, would yield opportunities for AI applications to engage in an
assessment process for both parties’ proposals before rending a decision in favour of either. Wahab, supra 14, at
421.
83
Marrow, supra note 53, at 65.
84
Id.
85
See https://eonum.ai (last visited Nov. 30, 2020).
86
Any type of conflict, either between companies, political opponents, or world-powers, involves a high level of
uncertainty, emotional volatility, power imbalances, and mysterious information. Eonum’s quantum model aims
to make sense of this data, providing their clients with a more realistic assessment of their chances and
supporting them in the processes of risk assessment, litigation, or arbitration. Catalina Sparleanu, A Paradigm
Shift. The Role of Quantum Mechanics in Arbitration, Politics, and Defense, (Oct. 23, 2020)
https://www.supertrends.com/a-paradigm-shift-the-role-of-quantum-mechanics-in-arbitration-politics-and-
defense/.

22

Electronic copy available at: https://ssrn.com/abstract=3815946

processes may be specifically designed to train an algorithm for it to render an arbitral award

that embodies not only logic, but also the emotional derivative of the data set.

(ii) Lack of Reliable Data: Confidentiality in Arbitration

Machine learning was developed as a result of the advent of big data. With an increase

in the number of internet users, the availability of data that could be fed to the Machine

Learning System also grew considerably. This data could be fed to the system for it to pick up

patterns, learn and render predictions. Hence, access and availability of data is the basic

requirement for the success of any data driven AI program. The larger the sample data, the

more accurate the model’s predictive value. 87

A free-flowing availability of data, such as awards, interim awards, procedural orders,

etc., will be needed to fuel the decision making algorithm. Gaining access to such data may not

be an easy task. Confidentiality is touted as one of the most sought after features of arbitration.

Arbitral proceedings are conducted privately with only the parties, their counsels, the arbitrator,

and the administrator being privy to the information being disclosed during the proceedings.

The awards are generally not published and are, therefore, not easily available in public

domain. Even if such data is made available, the volume of data would be considerably low,

making impossible the constitution of an appropriate database to train an algorithm.88

However, there is an increasing demand for transparency in the practice of arbitration.

Arbitral institutions are moving towards publication of arbitral awards. A recent example of

this ideological shift is visible in the revised note of International Chamber of Commerce (ICC)

released on 20 December 2018. 89 The revised text, which came into effect from 1 January 2019,

87
Scherer, supra note 9, at 15.
88
See, Scherer, supra note 9, at 15, 16.
89
Article III(D)(40-46) Publication of Awards, Note to parties and arbitral tribunals on the conduct of the
arbitration under the ICC Rules of Arbitration, Jan. 1, 2019
https://iccwbo.org/content/uploads/sites/3/2017/03/icc-note-to-parties-and-arbitral-tribunals-on-the-conduct-of-
arbitration.pdf.

23

Electronic copy available at: https://ssrn.com/abstract=3815946

adopts a new and very different default position that, barring party opt out, ICC arbitral awards

will be published two years after they have been notified to the parties. 90 Many other notable

institutions, such as the Stockholm Chamber of Commerce (SCC), the Singapore International

Arbitration Centre (SIAC), and the Hong Kong International Arbitration Centre (HKIAC)

allow publication of redacted or anonymised awards.91 Redacted or anonymised awards may

be devoid of inessential information such as the names of the parties and other identifying

information, but such awards would provide insight into the functioning of the tribunal, as well

as, in identifying standard practices.92 The awards available in public domain could then be

utilised as input for training the ArBot.

Another feasible solution to this issue would be the development of AI models for

decision making by arbitral institutions such as the SIAC or the ICC, which already have access

to arbitral awards. Such data could be used as input to train the AI decision making models,

and then apply the model to the incoming disputes.93

(iii) Need for Reasoned Decision Making

Providing a reasoned decision is one of the fundamental features of legal decision

making. 94 Explanation to a judicial pronouncement makes the judge accountable for the

90
Vanessa Naish, The transparency conundrum: will the ICC’s new pro-publication approach to arbitral
awards win over the majority? (Part 1), THOMSON REUTERS ARBITRATION BLOG (Apr. 15, 2019)
http://arbitrationblog.practicallaw.com/the-transparency-conundrum-will-the-iccs-new-pro-publication-
approach-to-arbitral-awards-win-over-the-majority-part-1/.
91 Id.
92
There are certainly some benefits that could result from greater transparency and access to more data in
arbitration. It is arguable that publishing arbitral awards gives practitioners access to procedural decisions that
may be useful for identifying “standard practice” and new approaches. Data about challenges to arbitrators helps
to identify whether a challenge may have grounds and limits the potential for unmeritorious challenges. Vanessa
Naish, The transparency conundrum: will the ICC’s new pro-publication approach to arbitral awards win over
the majority? (Part 2), THOMSON REUTERS ARBITRATION BLOG (Apr. 18, 2019)
http://arbitrationblog.practicallaw.com/the-transparency-conundrum-will-the-iccs-new-pro-publication-
approach-to-arbitral-awards-win-over-the-majority-part-2/.
93
Daniel M. Katz et al., A General Approach for Predicting the Behavior of the Supreme Court of the United
States, 12(4) PloS One (2017). This study was conducted by the researchers on the prediction of US Supreme
Court decisions. Overall, the model predicted the votes of individual justices with 71.9% accuracy, and the
outcome of the decisions with 70.2% accuracy.
94
Scherer, supra note 9, at 22.

24

Electronic copy available at: https://ssrn.com/abstract=3815946

decision. The explanation not only provides legitimacy to the decision but also forms the basis

for challenge to the decision before appellate authority. In the context of AI, it may be fair to

say that such explanation would form the very basis for constituting the Machine Learning

algorithm to be deployed for legal decision making. However, programmers face issues with

providing reasons for the decisions made by an algorithm. Similarly, it may raise significant

issues in making an ArBot capable of providing reasoned awards.

Article 31(2) of UNCITRAL Model Law95 provides that the arbitral tribunal shall state

the reasons upon which the award is based, unless the parties have agreed that no reasons are

to be given. Many countries around the world have adopted the UNCITRAL Model Law, or

have adopted the law with some variations. Therefore, it may not be out of context to say that

unless parties agree, an ArBot acting under the legal regime in most countries would have to

provide explanation for its decision. The award rendered must be interpretable by not only the

parties, but also the judicial forum exercising its authority in enforcing or hearing a challenge

to such an award.

Explanations expose information about specific individual decisions without

necessarily exposing the precise mechanics of the decision-making process. 96 Neither

identifying important factors nor reasoning about their counterfactuals requires knowing the

flow of bits through an AI system, no more than explanation from humans requires knowing

the flow of signals through neurons.97 In the context of artificially intelligent decision making,

an ArBot would be required to assign reasons and justifications for a particular outcome and

not lay down the decision making process. With the ongoing research on the arena of

Explainable AI (XAI), it may be possible to provide explanations to the decisions rendered by

95 Article 31(3), UNCITRAL Model Law On International Commercial Arbitration, as adopted by the United
Nations Commission on International Trade Law on 21 June 1985.
96
Finale Doshi-Velez et al., Accountability of AI Under the Law: The Role of Explanation, 2
https://arxiv.org/pdf/1711.01134.pdf.
97
Id. at 12.

25

Electronic copy available at: https://ssrn.com/abstract=3815946

the AI. XAI refers to methods and techniques in the application of artificial intelligence such

that the results of the solution can be understood by human experts. Major tech companies are

now working towards finding effective solution to the problem of machine learning black

box.98

XAI encompasses a range of efforts to explain, or help humans interpret, how a

particular machine learning model reached its conclusion. 99 We can discern two types of

explanations for the decisions made by algorithms. The first, known as Model-Centric

Explanation, revolves around the model of algorithm itself. However, an ArBot would not be

required to provide in the award its decision making process. The second category, Subject-

Centric Explanation, concerns the input record. 100 Subject-Centric Explanation is a local

explanation and can be provided with reference to a given query. 101 It may be given when one

needs a justification for a specific decision, for instance, why a particular loan application was

rejected. An approach to achieving Subject-Centric Explanation may be by quantifying how

changes in inputs impact outputs to identify the important factors and reasoning about their

counterfactuals. Such a system could be easily probed to enlist the important factors in a

decision. In view of such technologies, the possibility of a reasoned award rendered by an

ArBot may be ruled out indefinitely.

(iv) Replicating Existing Biases

98
In a recent interview, Rohit Prasad, Amazon’s chief Alexa scientist said that the company is working on
Alexa’s ability to provide explanation for its actions. See Dave Gershgorn, Why it’s so hard for Amazon Alexa
to really explain itself? (Sept. 26, 2019), https://onezero.medium.com/why-its-so-hard-for-amazon-alexa-to-
really-explain-itself-9d5ede67a68.
99
Ashley Deeks, The Judicial Demand for Explainable artificial intelligence, Virginia Public Law and Legal
Theory Research Paper No. 2019-51 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3440723 (last visited
Nov. 30, 2020).
100
Subject-centric explanations are built on and around the basis of an input record. They can only be provided
in reference to a given query—which could be real or fictitious or exploratory. Edwards & Veale, supra note 3,
at 56.
101
Id.

26

Electronic copy available at: https://ssrn.com/abstract=3815946

 AI constantly faces backlash in almost all industries for exhibiting biases. A software

used to predict future criminal behaviour was found to be biased against minority racial

groups.102 In 2015, for example, Google’s photo app embarrassingly tagged some black people

as gorillas.103 Although a machine learning algorithm would improve with use, but in the early

stages the algorithm could suffer from biases based on the input data as an algorithm is only as

good as the creator’s data set. 104 Mistake will thus usually result from the input data and are

more difficult to detect and fix. 105 Would AI learn biases in favour of large companies that win

disputes more frequently, disputing parties from certain countries, or certain types of claims?

Should AIs be challenged for issue bias? 106 The algorithm could amplify and perpetuate

embedded biases and generate even more bias-tainted data to feed a vicious cycle. 107 In the

context of an AI powered arbitral decision making, a machine, while comprehending the

training data comprising of awards rendered by human arbitrators, may also replicate the biases

of the human arbitrators.

Despite the limitation, many academicians and researchers believe that AI models have

the advantage of algorithmic objectivity and infallibility over humans who inevitably make

mistakes and are influenced by subjective, non-rational factors.108 Cognitive biases appear in

day to day human decisions. For instance, a notorious study of Israeli parole judges over an

eight-month period revealed that after lunch, an applicant’s chance of parole rose to 65% before

steadily falling again. The severity of the applicants’ crime bore no relation to the order in

which they came before the judges; they were simply more likely to be merciful when they

102
Christine Sim, Will Artificial Intelligence Take over Arbitration? Academia.edu,
https://www.academia.edu/36646259/Will_Artificial_Intelligence_Take_Over_Arbitrators_2018_forthcoming_
Asian_Journal_of_International_Arbitration (last visited Nov. 20, 2020).
103
Peter Rejcek, The Struggle of Making AI Less Biased than its Creators, SINGULARITYHUB (Jan. 31, 2017)
https://singularityhub.com/2017/01/31/the-struggle-to-make-ai-less-biased-than-its-creators/.
104
Sim, supra note 102.
105
Scherer, supra note 9, at 21.
106
Sim, supra note 102.
107
Karen Hao, AI is sending people to jail—and getting it wrong, MIT TECHNOLOGY REVIEW (Jan. 21, 2019)
https://www.technologyreview.com/s/612775/algorithms-criminal-justice-ai/.
108
Scherer, supra note 9, at 18.

27

Electronic copy available at: https://ssrn.com/abstract=3815946

were not hungry.109 This example makes it clear that human decision making is affected by

extraneous factors, which could be irrelevant to the merit of the case. 110 An AI based ArBot

would be free from such extraneous biases. Further, although the risk of AI learning from and

replicating the inherent biases of humans is real, but that does not necessarily mean that the

award rendered would be unfair. What gives the computer the edge is the ability to uncover

and remove bias using simulation techniques. 111 A machine is capable of vindication, whereas

in the case of a human arbitrator, it would be impossible to assess the inherent bias.

Further, quantum computing combined with AI is capable of eradicating the

shortcomings of AI. Many of the problems affecting AI today, including things like

accidentally programming inherent bias into their systems, could be caused by the limitations

of the datasets they are using. 112 Employing quantum computers to give an AI quick access to

trillions more data points could supercharge its accuracy and usefulness. 113 There are also

initiatives such as diversity.ai that are working towards preventing racial, age, gender,

disability and other discrimination by humans and AI. Diversity.ai addresses these issues

through meetings, seminars, hackathons, open database sharing and audits of artificially

intelligent systems to promote and encourage inclusion, equality, balance and diversity.114

(v) Costs Involved in Formulating AI Decision Making Model

109
Cohen & Nappert, supra note 27.
110
Id.
111
Marrow, supra note 53, at 66.
112
Breeden, supra note 56. See Gary Fowler, How Can AI and Quantum Computers Work Together? (Oct 27,
2020, 07:40AM), https://www.forbes.com/sites/forbesbusinessdevelopmentcouncil/2020/10/27/how-can-ai-and-
quantum-computers-work-together/?sh=519440e16ad1. As AI entrepreneur Gary Fowler explains it, today’s
computer intelligences are limited by the amount of data they have access to, and how quickly they can analyze
it. The size of the dataset determines the quality of the results, so if the information isn't ample, the output won't
be promising either. However, thanks to quantum computing's ability to go beyond the traditional binary coding
system, it makes it possible to enlarge and enrich the data set, both in terms of volume and diversity.
113
Breeden, supra note 56.
114
See http://diversity.ai (last visited Nov. 30, 2020).

28

Electronic copy available at: https://ssrn.com/abstract=3815946

 Another challenge to decision making by an ArBot would be the expenditure involved

in its constitution. This would be onerous task requiring funding for getting access to reliable

data sets, assimilation of the available data sets, setting up of review process etc. Considering

we are talking about application of the AI model to small claims’ disputes, there will be dearth

of sufficient incentive for any start up to invest in a project like this. There have been several

instances where e-commerce websites have either constituted their own ODR systems, or

collaborated with companies involved in providing dispute resolution mechanisms. The way

eBay came up with its dispute resolution mechanism is illustrative of this. Today, eBay claims

that it resolved 60 million cases in the past year. A possible solution to this may be incentivizing

the existing arbitral institutions and companies to develop the AI models using the data sets

available in public domain. The cases that such model will cater to are currently outside of the

purview of most of these arbitral institutions, since small claims disputes rarely opt for

institutional arbitration owing to the costs involved. This AI model would enable the arbitral

institutions to cater to a new set of clients and build a reputation in the market for innovation

and trust.

VI. Conclusion

Industrial norms globally entail execution of standard form employee contracts,

consumer contracts, insurance contracts etc. Such contracts are not the best examples of equal

bargaining power of the contracting parties, and more often than not, contain vexatious clauses

which put the parties on different pedestals. A dispute resolution clause forcing the employee

or the consumer to arbitrate the dispute is an example of such a clause. Owing to the costs,

effort and time consumption involved in arbitration, the employees, consumers, or such other

29

Electronic copy available at: https://ssrn.com/abstract=3815946

parties refrain from resorting to dispute resolution, thereby benefiting the party which

possessed greater bargaining power at the time of execution of the contract.

The legal profession has a valuable role in facilitating convenient and inexpensive

access to justice for disputing parties to instil their trust in the legal machinery. Litigants don’t

want courts. 115 They want their disputes resolved fairly and with finality. 116 Developing

effective mechanisms such as ArBot will provide access to justice to such parties. With a

reduced dependency on the physical movement of persons involved in the arbitral process,

such advancement in technology will simultaneously address the demands for inexpensive and

greener arbitration. Besides the ethical advantage, developing such mechanisms will lay down

the foundation for innovation in the legal field. It would bring the stakeholders on the same

page as other flourishing professions in the world that have embraced technology.

As has been argued by Richard Susskind 117 , it is time that working practices must

ultimately serve the needs of customers, not workers.118 Similarly, the legal profession needs

to progress based on the demands of justice, fairness and equity of the persons availing the

legal professional services. An ‘Output-based’ thinking would lay down the foundation for

developments in the field of Law-tech. In the near future, even if AI based decision making

does not seem feasible, at least the procedural aspects of arbitration could be more effectively

delegated to AI systems. At some point, a human tribunal might consult with AI as a

supplement to, or check on, their decisions. This kind of approach might provide a happy

medium between those who insist on human interaction in an arbitral proceeding and those

willing to place their faith in the less fallible, but less palatable, prospect of “robot justice”.119

115Richard Susskind, AI, work and outcome thinking, British Academy Review, No. 34, Autumn 2018, 30.
116
Id.
117
Richard Susskind is Technology Adviser to the Lord Chief Justice of England and Wales, and Chair of the
Advisory Board and Visiting Professor at the Oxford Internet Institute.
118
Richard Susskind, supra note 115, at 30.
119 Cohen & Nappert, supra note 27.

30

Electronic copy available at: https://ssrn.com/abstract=3815946