The AI and the Future of Skills (AIFS) project at the OECD’s Centre for Educational Research and Innovation (CERI) presents a framework to systematically measure artificial intelligence (AI) and robotic capabilities and compare them to human skills. In this chapter, AIFS discusses possible ways policy makers could draw on the OECD’s AI Capability Indicators to model implications of AI developments in various domains. The first section outlines, an approach to leverage the indicators to estimate how AI will transform occupational demand. In the second section, the OECD presents how education policy makers can use the indicators to inform discussions about the transformation of teacher roles in education and how learning goals may need to shift to account for changing occupational demand.
Introducing the OECD AI Capability Indicators
Abstract
Artificial intelligence (AI) is progressing rapidly, but not all advances will lead to major societal and economic change. The OECD AI Capability Indicators help identify where AI might have transformational impacts. While the indicators are too broad for evaluating specific AI applications, they are well suited to spotting larger shifts in how jobs and learning could evolve.
Because the indicators are new, the OECD has not yet applied them systematically. This chapter outlines how these indicators can be used to map AI progress towards human abilities required at work. It also explores how this mapping can signal possible transformations at work and in education systems, which can guide future policy discussions.
To assess how AI might lead to transformational change in the economy and society, the first step is to understand which occupations – and which component tasks of these occupations – require abilities that AI may soon be able to perform. This analysis begins by linking the OECD AI Capability Indicators to descriptions of occupations and their component tasks to derive capability requirements.
The AI Capability Indicators can be directly linked to the ability and skill requirements of different real‑world jobs by comparing the level descriptions of the nine indicators to the job descriptions. This can be done in two ways: by looking at entire occupations with all their characteristics, or by focusing more narrowly on specific tasks within those occupations. The goal is to determine the capability level needed to carry out a job or task effectively along each of the AI Capability Indicators.
This work can be done using the US-based O*NET system because it is one of the most extensive datasets worldwide of occupational characteristics. O*NET includes data for about 900 occupations and its taxonomy has been adopted in part or compared to the occupational taxonomies used by many other countries. O*NET and the other occupational databases directly modelled on it comprise a set of occupational characteristics – including detailed tasks, work environment and required human abilities – that can be directly compared to the AI Capability Indicators.
This section attempts to illustrate the approach using the example of the teaching profession. As illustrated in Figure 4.1, a substantive part a teacher’s job requires language, social interaction and problem-solving capabilities at some of the highest levels. This becomes particularly evident when the demands of a specific teacher’s task are considered. The O*NET task “Adapting teaching methods and instructional materials to meet students’ varying needs and interests” draws on a range of capabilities described at levels 4 or 5 of the AI Capability Indicators in language, problem solving and social interaction. Thus, teachers must leverage nuanced language abilities (language and social interaction, level 5) to communicate instructions clearly and tailor feedback to learners with different skill levels, cultural backgrounds and learning styles. They also engage in problem solving by interpreting interactions within a complex social environment, identifying individual or group-level barriers to understanding and developing targeted approaches to overcome them (problem solving, level 4). Additionally, teachers build rapport, handle ambiguity and respond to emotional cues, all while nurturing a supportive atmosphere that helps maintain engagement and motivation among diverse learners (social interaction, level 5). These early judgements are still tentative and would need to be refined through further validation and expert review.
This type of analysis can be scaled up using a multi-step process:
A survey of job experts can provide judgements of the level of capabilities required for carrying out occupations and occupational tasks from a representative sample.
These judgements can be extended to cover all occupations and the tens of thousands of tasks included in the O*NET and European Skills, Competences, Qualifications and Occupations (ESCO), using AI-based techniques and statistical inference.
Once occupations and component tasks have been linked to the required level of capabilities on the AI Capability Indicators, the gap between AI’s current capabilities and those required by a specific occupation or task can be calculated. These estimates can then be used to identify occupations and tasks that AI can perform at different levels of AI capabilities.
It will be easiest to identify occupations or tasks where AI possesses all required capabilities, allowing full automation. However, it will also be possible to use these analyses to identify occupations or tasks where AI possesses only some of the required capabilities. This would point towards potential human-AI collaboration with workers who have the complementary abilities that AI lacks. Such analyses can be used to create summary metrics showing how different profiles of AI progress on the indicators could affect different occupations and larger economic implications.
Mapping the AI Capability Indicators to occupational and task requirements can serve as a starting point for an in‑depth analysis of how particular tasks within occupations may evolve with current and improved AI capabilities. Such an analysis will involve a structured discussion with stakeholders from the concerned occupation(s) to understand:
1. whether the society wants AI to ever carry out a particular task
2. whether AI systems already exist or are in the pipeline that can carry out the task
3. what adjustments would be needed for humans to work along with AI systems.
The result will be transformed task scenarios – vignettes that are vetted by the industry itself, education and training providers, and AI experts. The vignettes can illustrate the way the teacher roles and certain other adult roles might change. This methodology could also be applied to capability profiles across different countries and at a subnational level. This will allow policy makers to identify geographies that are more or less likely to be exposed to AI and guide appropriate policy responses. A more detailed discussion of the OECD’s approach can be found in Chapter 14 of the technical volume (OECD, 2025[1]) accompanying this report.
To illustrate the vignettes, Box 4.1 describes a scenario in which AI is at the midpoint on the AI Problem‑solving scale. It has a high level of problem solving ability with respect to natural sciences, medicine and engineering. However, it has more limited abilities with respect to social and ethical reasoning, and problem solving, as well as moderate sensory motor capabilities.
In a blisteringly hot summer, a highly contagious mosquito-borne infectious disease breaks out, rapidly spreading to major cities nationally and internationally. An AI system has identified the problem through analysis of disparate datasets and recommended that public authorities declare the infection as a pandemic. Medical centres, research institutions and government organisations start a collaborative programme to investigate the disease and develop a cure and a vaccine.
AI systems continue to track online news and social media to model how the disease spreads and which symptoms it involves before more centralised data collection can begin. This information helps systems detect promising prevention and containment measures for public authorities to adopt and implement, and to begin considering effective treatments.
AI systems design and execute experiments co‑ordinating dozens of robotic molecular biology labs to analyse samples and data gathered around the world – in real time. Before this event, biomedical literature has informed the AI systems and characterised an enormous diversity of biomedical research data. Using this knowledge, AI can hypothesise pathways where the virus might be interfering.
After prioritising the hypotheses, based on the known literature, human teams implement targeted experiments in mice. Working with the results, an AI system discovers an interesting link to a rare neurological condition and proposes a viable treatment.
AI helps governments globally to manage their stocks of medical material to ensure a fair and efficient distribution in response to the disease. Furthermore, AI-accelerated discovery is the key to identify the novel mechanism of viral action and design an effective vaccine. Within days, thousands of doses are produced and distributed; the disease is under control and those affected are in remission.
Source: Adapted from (Gil and Selman, 2019[2]).
In this vignette, AI systems:
show strong capabilities in data and information retrieval, monitoring, analysis and interpretation, which are linked to their problem identification capacity and related research and development efforts.
are able to identify, support and design solutions, in both the policy and health-care domains.
can work with complex systems relying on scientific knowledge and the analysis of novel data. They work effectively with models in areas spanning human biology and broader ecosystem dynamics – including social and human-nature interactions – demonstrating the capacity of reasoning from evidence.
The human role, by contrast, focuses on:
directing public and private efforts towards containing and eventually solving the crisis once the disease is discovered.
ensuring that AI’s experimental research protocols are administered and that vaccines are distributed effectively once developed and produced by AI and robots. A range of physical skills are required, notably manual and visual skills for various manipulation activities.
employing social skills and ethical reasoning in crisis management and co‑operation activities.
Building on the indicators and the occupational analysis above, a new framework emerges for examining the potential for transformational change in education (Figure 4.2). This example focuses on the education sector given its relevance to an education policy audience but could in principle be extended to any occupation or economic sector. This framework can inform decisions about both the purpose and delivery of education. It can guide discussion of how educational practices might evolve if AI capabilities surpass human performance in certain teaching tasks. It can also handle key questions about which learning objectives and curricular content may stay the same, need removal or revision, or which should be newly introduced – depending on current or future AI capabilities.
Two key types of potential transformation in education stand out.
First, if AI can perform a substantial portion of teachers’ tasks, it could possibly transform education delivery. For example, if AI can reliably deliver instruction or provide feedback, the traditional role of teachers may shift towards mentoring, motivating or handling complex interpersonal dynamics. Such a development could redefine how teaching is delivered, sparking a transformation in classroom dynamics, teacher-student interactions and the overall learning experience.
Second, student learning goals are related to the kinds of tasks that society expects students to perform as adults at work and in their community and in everyday life. If AI can now perform many of these tasks, education systems may need to reconsider what students should be learning. These potential implications are already apparent in the ability of AI to outperform students in many assessments used in education, such as the OECD's Programme for International Student Assessment survey of the competences of 15‑year‑olds. As a result, some current goals may lose relevance, while others – like creativity or ethical judgement – may become more important.
Of course, these changes are not just technical. Societal values will shape whether and how the role of teachers in certain tasks is reduced or revise what students are expected to learn. The AI Capability Indicators and the occupational analysis above do not offer answers to these value questions, but they can highlight areas where major change is technically feasible and perhaps likely.
The OECD will explore these possibilities in greater detail through scenario-based analyses of teaching roles and student expectations. The resulting vignettes will guide the discussion of potential transformation in learning goals and curricular content. Chapter 14 of the technical volume (OECD, 2025[1]) accompanying this report provides a more detailed discussion of how education policy makers can leverage the OECD’s AI Capability Indicators.
The AI Capability Indicators in this report offer a simple but powerful tool to assess how AI progress aligns with human abilities. The indicators are informed by evidence from AI evaluations and expert opinion. By describing AI capabilities linked to human abilities the indicators are understandable to a non-technical audience. They thus provide a clear framework for anticipating the impact of AI progression across a range of domains relevant to work and education. This comparison to human abilities also provides a clear measure of AI’s progress towards artificial general intelligence and will remain stable amid rapid developments in AI capability. By linking AI performance to real-world work demands and educational goals, they help us see where major changes might happen – and where human roles will remain essential.
The indicators also provide a valuable signpost to AI researchers. They indicate the sorts of capabilities that will need to be tested to provide informative evaluations of AI progress as the limitations of current approaches to benchmarks become increasingly salient. They also provide a mechanism through which policy makers can communicate to AI researchers about the sorts of capabilities that need to be evaluated to address societal, political and ethical concerns around AI development.
The work presented here is just the beginning. With further development, the indicators and derivative analyses will enable policymakers to respond to rapid and disorienting AI developments with the clear and concise information they need to capitalise on the huge opportunities offered by advanced AI systems. The OECD will continue to develop and update the indicators so they become the premier international source for trusted information about AI capabilities and their implications for education, work and civil society.
References
[2] Gil, Y. and B. Selman (2019), “A 20-year community roadmap for artificial intelligence research in the US”, arXiv, Vol. 1908.02624, https://arxiv.org/abs/1908.02624.
[1] OECD (2025), AI and the Future of Skills Volume 3: The OECD AI Capability Indicators, OECD Publishing.