How to ensure education is relevant for all in a fast-changing world - Andreas Schleicher 5 April 2023.pptx

Editor's Notes

• #3 Across the OECD and beyond, educators and policymakers face the double challenge of helping populations recover from recent shocks (e.g. the geopolitical implications of the war in Ukraine, the COVID-19 pandemic) while ensuring individuals and societies adapt to longer-term trends that are changing the way we live and work (e.g. digitalisation, automation, demographic change, mass information and climate) Education and training can play an important role in addressing this dual challenge, but only if it is able to empower people of all ages and backgrounds to learn, unlearn and relearn dynamically throughout their lives as the needs of economies and societies change. The Education Policy Outlook 2023 provides policymakers with practical guidance on how to do this, starting with how to enhance the relevance of the education and training offer.
• #4 So what are the shorter-term disruptions that make transforming education pathways so important in 2023?
• #5 In 2022, global economic uncertainty put pressure on national and personal finances. In the wake of the invasion of Ukraine and the ensuing energy and food crises, as well as continued COVID-related lockdowns in China, economic recovery has lost momentum. Nearly all OECD countries had their projected growth in GDP downgraded in June 2022. By September 2022, high inflation had become deeply entrenched across most OECD economies.
• #6 Alongside these shorter-term disruptions, recent shocks have accelerated longer-term evolutions that were already changing the way we live and work.
• #7 Recent events have also further entrenched the digital transformation of work, society and, increasingly, education itself. Even before the pandemic, 15-year-olds spent an average of 35 hours per week on the internet, and in Denmark and Sweden it was close to 50 hours, a large chunk of which was in school. For them, the digital world has become the real world. Education must empower people of all ages to confidently navigate the digital world and ensure everyone benefits from digital transformations.
• #8 The dynamics behind that are not difficult to understand The kind of things that are easy to teach and test have also become easy to digitise and automate And technology intensive tasks are a on the rise
• #9 The dynamics behind that are not difficult to understand The kind of things that are easy to teach and test have also become easy to digitise and automate And technology intensive tasks are a on the rise
• #10 There are over 30 research areas in the domain of Natural Language Processing with at least one existing benchmark task. Human language competencies are often divided in four broad categories: Reading, Writing, Listening and Speaking. The work we commissioned aims to compare human levels of competence with current NLP system performance levels. In the chart above, you can see Dialogue open domain: Carry out a conversation with another person about any topic (chit-chat models) Conversational Intelligence Challenge. Evaluation challenging, bc it requires human evaluators to talk to chatbots about an open or prescribed topic before rating their quality. A recent evaluation of SOTA models: the highest performing models rated by human judges at approximately 52%. SOTA models in this area are based on transformers that learn from extremely large language models and produce highly fluent output that often is appropriate for the input provided by its human conversation partner. However, despite fluent output, models still lack consistency in conversations and the ability to reliably incorporate knowledge or learn new information from conversations. Dialogue task-oriented: Assist someone in completing a task where the help provided is through automated conversation (spoken or text-based). Dataset: KVRET, containing over 3,000 dialogues across the in-car assistant domain, for calendar scheduling, weather information retrieval, and point of interest navigation in English. Models are evaluated using entity-F1, a metric that evaluates the model’s ability to generate relevant entities from an underlying knowledge base and to capture the semantics. The current best task-oriented dialogue system achieves a score of approximately 71%; human evaluation results of 75%. But several evaluation challenges and limited data sets (given the large number of use cases).  Most closely to mid-level performance. Essay scoring: automatically assessing student essays (Direct applications in education: get feedback about the quality of their writing without requiring human time and resources The Automated Student Assessment Prize (ASAP) data set, released by Kaggle, contains 8 essay sets (150 - 550 words per response, written by students in US grades 7 to 10). Essays within the data set are hand-graded and double-scored to test rater reliability. Results for state-of-the-art systems show performance at approximately 80% weighted kappa. Discussions of human parity in this area are unlikely to develop. [Kaggle is an online community of data scientists and machine learning practitioners https://www.kaggle.com https://www.kaggle.com/competitions/asap-aes/data]   Question answering: Automatically finding answers to questions (or identifying when a question should not be answered due to some ambiguity or lack of info). General knowledge or knowledge about a specific topic; Ability to find relevant information with the help of technology, e.g. search engine. QA is an extensively studied area of NLP and there exists a vast number of data sets (> 100) and benchmarks for evaluating systems. The Stanford Question Answering data set (aka Squad) is a reading comprehension data set containing a sample of questions and answers about Wikipedia articles. In Squad, systems are required to either find the answer to each question in the corresponding article or identify that the question is in fact unanswerable. Results for state-of-the-art QA systems on leader boards for Squad and other QA data sets such as CommonsenseQA reveal excellent system performance and have led to discussions of systems reaching human parity within the QA research community. Information retrieval: Finding (and ranking) the content most relevant to a specific user information need. Finding material (usually documents) of an unstructured nature (usually text) that satisfies an information need from within large collections (usually stored on computers). Even the most basic IR algorithms from the early days already surpassed what is possible for a human in terms of information retrieval due to the scale of data needed to be searched.  benchmarking against humans are less relevant. Specialist on human side bc the task corresponds to the job of a trained librarian.
• #11 SLIDE FOR MOROCCO
• #13 As well as transforming our classrooms and workplaces, digital developments have revolutionised the way people access information and the amount of it readily available to them. The explosion of social media has played a particularly significant role. By 2021, Meta, the technology company that owns Facebook, Instagram and WhatsApp, hosted more than 6.2 billion active user accounts. If each account were owned by a different person, this would be the equivalent of 80% of the total global population. While social media and other digital innovations expand opportunities for citizens to express their voices and interact, they also increase their exposure to misinformation, blurring the line between reality and performance, fact and opinion, truth and lies. This places the onus on educators and policymakers to equip all students with the foundational literacy and critical thinking skills they need to navigate the information, misinformation and disinformation they are bombarded with daily.
• #14 There are signs that we are moving more quickly towards a greener future. After a mild increase in previous years, in 2022, clean energy investment is projected to account for almost three-quarters of the growth in overall energy investment. But, with the world in the midst of a global energy crisis, governments will need to strengthen their efforts towards a cleaner energy agenda. Policymakers must therefore ensure that education and training pathways give young people the knowledge, skills and agency to act for a more sustainable society.
• #15  Recent shifts in the demand for labour and skills and ongoing global uncertainty mean that transforming pathways to make them more relevant is no longer just an important challenge, it’s an urgent one. Education is becoming less about what you learn, and more about what you do with what you know. This means that education systems must provide learners with the opportunities to acquire the knowledge and skills that they require as they progress through these pathways. To achieve this, we will need to abandon the old system architecture of pre-determined routes, fixed destinations and divergent trajectories. Our systems need to become more learner-centred, evolving around their contexts, needs, interests and passions. This means responsive and flexible learner centred routes, multi-directional and ongoing learning journeys and interconnected and permeable pathways. Much remains to be done for education systems to get to this point.
• #16 This is the focus of the Education Policy Outlook 2022. The report outlines one key challenge for education: transforming pathways for lifelong learners and highlights three key approaches for doing this. Among them is, enhancing the relevance of learning pathways. In order to enhance it, policy makers can…
• #17 …strengthen their ability to anticipate future skills needs, starting with short term forecasts. The most effective anticipation exercises draw on data from different levels of the economy (local, regional, or national) as well as from different employment sectors. This is because the effects of trends and shocks can vary significantly across local economies. Policymakers also need to focus their attention on the population groups most affected by these shocks and trends. This includes women, young people, workers over 50 and the lowest-skilled. Several of the policy examples in the report make use of a range of anticipation models and data and involve collaboration between different stakeholders. For example, the UK has recently conducted a forecasting exercise drawing on 130 sources to generate long-term insights. This was complemented by macro-economic projections informed by qualitative and quantitative data.
• #18  Second, countries need to become better at adapting the learning offer to evolving skills needs. Adaptions to the learning offer should ultimately seek to empower learners to navigate change. To achieve this, several countries have developed overarching education strategies where high-level strategic objectives break down into concrete actions. For example, Latvia’s Education Development Guidelines set out measurable goals for the education system to be achieved by 2027. Representatives from government, industry and wider society worked together to identify Latvia’s skills, challenges and opportunities using evidence from different sources. They also identified the key priorities and indicators that feature in the plan. Another key trend involves strengthening co-operation with employers and industry. In Japan, for example, 40 higher education institutions worked with local employers to develop over 60 flexible courses in high-demand fields such as healthcare and ICT. These target the unemployed and those on temporary contracts.
• #19    And third, countries need to become more able to make a more frequent practice assessing the impact of policy efforts through monitoring and evaluation. To put this into action, policymakers should focus on enhancing the monitoring capacity of local actors, such as teachers or municipal officials to help the system know if they are indeed offering relevant education opportunities to the different learners. The quality of evidence within an education system depends on the ability of these local teams to implement robust monitoring and evaluation strategies. One policy trend in this area involves testing and evaluating innovative approaches to adult learning. For example, Australia funded over 50 innovative projects to generate evidence on what works for getting at-risk adults back into education or training. They used the evidence to identify 15 promising projects and highlight their common features.
• #20 You can help governments navigate the risks and leveraging the opportunities. What is the right balance between modernising and disruption. How do we reconcile new goals with old structures How do we support globally minded and locally rooted students and teachers How do we foster innovation while recognising the socially highly conservative nature of education How do we leverage potential with existing capacity How do we reconfigure the spaces, the people, the time and the technologies to create powerful learning environments. Again, this is not a government project, this is a whole of society project and I believe the philantrophic sector can make such a difference.
• #21 These are main findings from the second chapter of Education Policy Outlook 2022, available on the OECD iLibrary.   Let’s empower learners through relevant education that helps them to act in a world of change. Thank you and I look forward to our discussion.