Do Adults Have the Skills They Need to Thrive in a Changing World?

To what extent do adults possess and use key skills required to contribute as workers and citizens to the economy and society? And what are these key skills?

In the early 2000s, these questions guided the design of the Survey of Adult Skills, a product of the Programme for the International Assessment of Adult Competencies (PIAAC), the first cycle of which was conducted in 2012.1 At that time, evidence was rapidly accumulating that investments in education were linked to increased earnings for individuals, while country-level differences in education were linked to national growth rates (Deming, 2022[1]). A number of studies attributed the higher growth rates of some countries to a more skilled workforce: the most rapidly growing countries had, on average, greater levels of educational attainment and a better quality of (school) education, according to children’s results in international tests of mathematics and science (Mankiw, Romer and Weil, 1992[2]; Barro, 2001[3]; Hanushek and Woessmann, 2008[4]).

At the same time, adults who struggled to understand written texts and extract relevant information from them to achieve their goals came to be seen as particularly exposed to a number of risks, including unemployment. A report by the OECD had already noted in 1992 that low literacy levels among adults were a serious threat to economic performance and social cohesion (Benton and Noyelle, 1992[5]). As the Survey of Adult Skills was rolled out, it was clear that machines had begun to take over repetitive tasks, gradually replacing workers in many sectors of the economy (OECD, 2013[6]). The employment prospects for adults with strong cognitive and interpersonal skills appeared much brighter: many of the tasks that they could perform were not at risk of automation.

The survey placed a strong emphasis on the direct assessment of literacy and numeracy, which were seen as essential foundations for acquiring further knowledge and skills, in a world where much learning took place through reading and processing written information. It also assessed problem solving in technology-rich environments, to specifically capture the ability to process information in digital environments.

Yet today’s societies and economies differ in many respects from those of the early 2000s, and even more substantive changes lie ahead. Many new technologies and tools are being invented and adopted; population ageing is intensifying; the imperative to reduce net carbon emissions is becoming increasingly urgent; and a rapidly spreading pandemic, COVID-19, brought significant disruption to economies, education systems and the ways people interact with each other, across the world. In light of these recent changes, how relevant are information-processing skills for individuals and societies? How can policy makers use information on the level and distribution of skills in their country to prepare for the changes ahead?

This chapter describes the social and economic context in which the 2023 Survey of Adult Skills was conducted, highlighting similarities and differences with the period when the first cycle of the survey was conceived and implemented. It identifies questions about the future of work and the vulnerability of democratic processes to disinformation campaigns as major contemporary policy concerns and suggests ways in which the results of the survey can inform related discussions and help policy makers identify effective responses.

Against this background, the chapter then describes in some detail the information-processing skills assessed in the survey: literacy, numeracy and adaptive problem solving. It presents example items used in the assessment and provides readers with information about how to correctly interpret the results of the assessment. The chapter concludes with an overview of what the participating countries and economies can learn from the indicators presented in this and in future reports from the Survey of Adult Skills.

Over the past decades, changes in the demand for goods and services, but also, and more importantly, changes in the capabilities of machines and in how they are used to produce goods and services, have profoundly changed the tasks that workers perform in their jobs (Autor, Levy and Murnane, 2003[7]; Spitz‐Oener, 2006[8]; Autor and Price, 2013[9]; Ikenaga and Kambayashi, 2016[10]). The adoption of new technologies has not only affected the distribution of jobs across occupations, but, at a much finer level, also changed the tasks that define individual occupations. Machines have replaced workers in the performance of some tasks, even as workers have developed new skills to work alongside machines in entirely new tasks.

Recent studies suggest that the proportion of the workforce whose job involved routine tasks continued to decline in the early 21st century – probably reflecting the increasing use of machines, rather than workers, to perform such tasks. Meanwhile, non-routine tasks have increased in importance, particularly those related to interactions with clients, customers, service users or co-workers. In the United States, for example, the share of the employed workforce whose job involved social (or interpersonal) tasks continued to increase after 2000 (Deming, 2017[11]); in contrast, the share of those whose job involved analytical tasks (i.e. engaging with textual information, with numbers or with other abstract objects) stopped increasing in the most recent period (Figure 1.1). Among Swedish men, returns to interpersonal skills grew larger between 2002 and 2013, while those associated with cognitive skills remained stable– likely reflecting an increasing demand for interpersonal skill (Edin et al., 2022[12]).2

Until recently, machines had a distinct advantage in the performance of rule-based operations: computers and robots could be programmed to perform tasks which were explicitly codified (most obviously, repetitive ones). The development of more rapid machines, integrating refined sensors to analyse the environment and select the most appropriate action, progressively expanded the contexts in which they could effectively be deployed. These now included more complex operations, in which a larger number of contingencies had to be considered. The tasks that could be automated became less repetitive, but still needed to be explicitly codified.

Then, in the last 10-15 years, a new technological paradigm emerged – commonly referred to as artificial intelligence (AI) (OECD, 2019[14]). Using a rapidly expanding suite of statistical learning models, and building on the increasing availability of (“big”) data and on the falling cost of computing capacity, AI has made rapid advances in its ability to predict the correct answer to problems where formal rules are impossible to codify, and where humans have until recently had a comparative advantage in making decisions based on their training or past experience (OECD, 2023[15]). Machines that embed modern AI technologies can not only do routine tasks faster and more cheaply than ever; they can now imitate and often outperform humans in a much wider range of situations. In 2022 experts judged that AI could already answer around 80% of the literacy questions administered in the Survey of Adult Skills – significantly more than the majority of adults who took the assessment in that year. AI could, in particular, solve most of the easy questions, which typically involve locating information in short texts and identifying basic vocabulary. The remaining tasks are nearly within reach: large language models (LLMs) are expected to be able to “solve” all the tasks in the Survey of Adult Skills assessment within a few years (OECD, 2023[16]).

Given a sufficiently wide corpus of data from which to infer patterns of appropriate responses, machines no longer need human programmers to distil explicit rules; they can learn the rules that guide their behaviour from the data themselves. As a result, the need for workers to rely on “tacit knowledge” and expertise in performing a task is no longer an obstacle to that task’s automation.

A new generation of devices and robots, powered by AI, will probably be able to replicate human performance in a significantly wider array of tasks than past automation technologies (Lassébie and Quintini, 2022[17]; Lane and Saint-Martin, 2021[18]). New types of tasks that can be automated include both cognitive tasks (such as classifying data, summarising texts, detecting deviations, planning and advising) and physical tasks (including some requiring fine psychomotor abilities and hand-eye co-ordination, driven by progress in computer vision technology).

Automation need not necessarily lead to a reduction in overall employment opportunities; the automation of routine tasks did not reduce labour demand overall (OECD, 2019[19]; Georgieff and Milanez, 2021[20]). The impact on workers is uncertain because of the productivity gains that motivate a firm’s decision to automate some tasks: automation allows the same goods and services to be produced at lower costs. As these gains translate into lower prices, the demand for these goods and services can be expected to increase, and, if this increase is sufficiently large, may result in more workers than before being required to perform the remaining, complementary tasks involved in their production.

When projecting the effect of automation on the demand for skills, there are two mechanisms that must be considered: complementarity with non-automated tasks, and the emergence of entirely new jobs (Acemoglu and Restrepo, 2019[21]). Complementarity with a new technology often benefits different occupations and workers from those that were displaced in the first place. But complementarity may also be observed within a given occupation, particularly if jobs involve a complex bundle of tasks. In fact, the complementarity is more likely to operate at the level of tasks rather than at the level of jobs. Workers who perform tasks that are complementary with the new technology will be made more productive by the new technology and may experience substantial changes in the tasks they perform and the skills they need to bring to the task. Considering the type of tasks that AI can do, the transformation of occupations, rather than their disappearance, may be the most frequent pattern observed as its adoption in the economy progresses (Lane and Saint-Martin, 2021[18]).

There is little doubt about the potential of AI to automate many tasks. But there is also considerable uncertainty about the strength of the complementarity effects that result from automation, and about the skills that workers will need to have, or acquire, in order to benefit. To anticipate the consequences of these new technologies for skill demand, it is useful to distinguish two ways in which the statistical learning models behind AI applications can be trained (Acemoglu, 2024[22]): by learning how actions are linked to outcomes, or by learning to imitate the decisions of humans in similar settings. For tasks that have a reliable, observable outcome metric, and where the possible interactions between actions and contextual factors in determining success are relatively simple, statistical learning algorithms can evaluate the success of their own actions and gain a better understanding of how possible actions are linked to success than any human could possibly do. This can lead to large productivity gains and can be expected to result in the targeted automation of particular subtasks.

In many workplace situations, however, either there is no reliable outcome metric to define success, or the link between actions and desired outcomes may depend on a vast number of contextual factors. In the corresponding tasks, AI systems can still learn to imitate humans, but cannot learn from their own actions what the optimal strategy might be. As a result, there will be a tendency for their performance to be similar to the average performance of human decision makers, rather than to the performance of experts (which can reliably be identified only in the former type of tasks). In these cases, AI systems will not outperform humans in some specific subtasks, but rather replicate humans across whole tasks. The gains that firms can expect from adopting AI technologies are more limited in this situation and the workers whose tasks complement those being automated may not see any indirect benefit.

There are not yet any studies available for the period after 2012 describing overall trends in skill demand based on detailed descriptions of occupational tasks, similar to the ones that documented the decline in routine task performance over more than four decades (e.g. Figure 1.1). Even if they existed, such studies would be unlikely to capture the impact that AI technologies will have over the next 10 years. As new technologies are adopted in the economy, they may initially mostly leave workers performing the same occupation in new ways – a change that will not be fully and immediately reflected in standardised descriptions of occupational tasks. Moreover, even in recent years, there are many other factors influencing trends in occupation shares besides the adoption of AI technologies. These include a deepening of more traditional forms of automation (e.g. due to lower costs of computers and faster connectivity), but also changes in the demand for particular goods and services. For example, population ageing or the desire to limit greenhouse gas emissions also change the demand for skills, through their effect on the demand for particular products and services (OECD, 2024[23]).

Another approach to monitoring the evolution of the demand for skills is to use the information available in job postings, which are, nowadays, often published on line. Online vacancies may not be fully representative of all vacancies and certainly do not represent all jobs in an economy, but they reflect the changing content of jobs in real time. Tracking them over time therefore allows emerging trends in skill demand and variations in the skills required for a particular occupation to be captured. Analyses based on a large database of online job postings show, for example, that the share of new vacancies emphasising decision-making responsibilities grew substantially in the United States between 2010 and 2018 (Deming, 2021[24]). Extending this approach to online job vacancies in ten OECD countries, Green (2024[25]) shows that over the decade leading up to 2021-22, firms have increased the emphasis on specific tasks (such as using digital office tools and collaboration software) in job descriptions for occupations that rely heavily on non-routine analytical tasks. This suggests that the corresponding skills have become more important in defining success in these occupations. Next to digital skills (and, in particular, the use of office tools and collaboration software), these include also social skills such as collaboration or teamwork and cognitive skills such as originality and creativity. It is unclear, however, whether these trends will continue as new AI-based technologies are adopted.

Green (2024[25]) also finds that the establishments that are most likely to have already adopted AI technologies over this time period did not reduce hiring overall. However, they did reduce, at the margin, their demand for some types of workers and their skills, relative to their competitors within the same industry (Figure 1.2). In particular, they reduced demand for general resource management skills (including budgeting and accounting); for clerical tasks such as administrative support and record keeping; and for workers with the ability to use basic digital office tools such as spreadsheets and word processors. They also posted relatively fewer vacancies that highlighted originality as an asset for the position. In contrast, for other groups of skills, hiring trends were similar regardless of an establishment’s exposure to AI. These include physical skills, and skills related to science, medicine, law and public safety. Meanwhile, demand for skills involved in “production and technology” tasks (which include design and engineering, but also construction and food production) seemed to increase in establishments with the highest likelihood of adopting AI technologies (Green, 2024[25]).

For specific AI applications, a number of experiments in laboratory or real-world settings have observed the likely strength and direction of skill complementarity effects. These studies examined generative AI technologies that could assist web developers in writing software code (Peng et al., 2023[26]), guide the conversations of customer support agents (Brynjolfsson, Li and Raymond, 2023[27]) or help white-collar professionals with specific writing tasks (Noy and Zhang, 2023[28]).3 In each case, the AI application could automate some subtasks. The largest gains accrued to the less experienced and less productive workers in these occupations, suggesting the possibility of complementarity with general cognitive skills, but not with job-specific skills associated with expertise.

How will AI and human labour complement each other in the future? And how fast will new AI technologies be adopted in the economy? The answers to these questions will determine skill demand over the next 10 years, and the possibilities are still wide open. The future depends not only on the current capabilities of AI but also on the direction in which they are developed: replicating the performance of “average” humans in the production of existing goods and services; enhancing the productivity of most and acting as an alternative to scarce talent in the performance of essential tasks; or creating entirely new products and services, where humans and machines complement each other in new ways. The future also depends on any barriers to the adoption of new technologies that firms face, including regulatory barriers, labour-market frictions and the costs of training the current workforce. Ultimately, the demand for skills in the long term largely depends on the choices of consumers, citizens and policy makers, who can influence the direction of AI research, create new demand for labour in jobs not substituted by AI, lower some barriers to adoption and, at the same time, raise new protections, in order to achieve their desired outcomes.

Technologies are changing rapidly, and with them, the organisation of tasks within almost any occupation. To adapt to this reorganisation and the emergence of new tasks, and to weather potential job losses and the need to find new jobs, workers need to learn new skills. In a changing world, some skills that individuals have acquired through education, training and practice may rapidly become obsolete.

However, this is not the case for high levels of literacy, numeracy and (adaptive) problem solving; indeed these general information-processing skills underpin individuals’ ability to adapt and to weather change and uncertainty. For individuals who master them, these skills make access to new, specialised knowledge and know-how in any domain easier. For younger and older adults alike, they form the foundation for acquiring new skills through formal and informal learning processes. Hanushek et al. (2017[29]), for example, used data from the first cycle of the Survey of Adult Skills and found that labour market returns to literacy and numeracy are higher in countries that have grown rapidly, supporting the hypothesis that such skills facilitate adaptation to change.

Such “foundational” skills are the focus of the Survey of Adult Skills assessments. The tasks used by the survey to measure them, however, are now well within the reach of modern AI (OECD, 2023[16]). If machines can provide better answers to the Survey of Adult Skills items than most humans, in what sense is it relevant for adults to possess the skills that are measured by these tests?

This critique rests, in part, on fallacious logic. When calculators became better than the best humans at solving mathematics operations, the performance of addition, division and other such operations ceased to be part of the everyday job of shopkeepers, bank tellers, ship pilots and other workers. Yet the teaching of such operations continued and the importance of this training for humans, as a means to develop mathematical understanding and numeric reasoning, did not change. Similarly, the ability to complete a few dozen tasks in the Survey of Adult Skills allows us to draw much more wide-ranging conclusions about humans than about machines. The limited number of tasks included in the survey reflect typical progressions in proficiency observed in humans, but this underlying proficiency is an unobservable trait that helps humans in performing a much broader (potentially infinite) set of tasks. Correlations between this underlying proficiency and observable behaviour outside the test situation are expected to have solid roots in human psychology and neurobiology, allowing for reasonably robust predictions of what adults at a certain level of proficiency (as assessed by the Survey of Adult Skills) would be able to achieve in a wide range of situations and tasks.

In contrast, when a machine is able to solve a task in the Survey of Adult Skills, because it has learned to imitate the actions of the most effective humans in similar situations, it is much harder to make inferences about what that machine can do beyond the specific task. While machines may, in the future, continue to expand their capabilities, the capacity of current AI systems to “learn” from their actions remains highly context specific, and depends on the availability of a corpus of relevant and reliable training data.

In many contexts, the use of AI could make some tasks easier for workers, who may no longer be required to perform all of it. Rather than having to produce a translation, or write code, they can instead focus on slightly simpler subtasks: verifying the translation, debugging the code or explaining a new software feature to a client. Current experimental evidence suggests that AI can often replace tacit, job-specific knowledge that is usually only acquired with experience (if at all). This is most likely to happen in contexts where AI can be trained to capture and reproduce the behaviour of the most productive workers. In these contexts, AI technologies could empower a larger share of workers – those with at least minimal skills for the task – to compete with niche experts (Autor, 2024[30]; Brynjolfsson, Li and Raymond, 2023[27]). This could undeniably lead to a form of de-skilling, but workers will still need general information-processing skills to understand, interpret and take advantage of the support given by AI (Noy and Zhang, 2023[28]; Autor, 2024[30]).

While AI is therefore unlikely to make information-processing skills useless, it could potentially change the relevance of some aspects (or subdimensions) of literacy and numeracy, by changing the types of tasks adults are most likely to engage with in the future. When using literacy skills, for example, there may be less emphasis on extracting information, and more on constructing knowledge, including navigating ambiguity and distinguishing fact from opinion. The relevance of such aspects is already quite evident in the ocean of online information that adults have to navigate to access news and information, and which replaced traditional media outlets.

Valid and reliable measures of literacy and numeracy skills are also indicative of other, harder-to-measure traits, of continued value. Adults reach high levels of literacy and numeracy only as a result of long, difficult learning processes. This means that those with a strong command of these skills are likely to also have the attitudes, or emotional skills, that will assist them in learning new skills in the future – including the interpersonal, physical and manual capacities that may be hardest to automate successfully.4 The background questionnaire of the 2023 Survey of Adult Skills can be used to validate some of these claims, with measures of adults’ “readiness to learn” and participation in formal and informal learning activities, and an assessment of social and emotional skills.

Finally, even if AI threatens to displace workers in cognitive-intensive occupations (such as financial analysts, translators and interpreters), these workers will not necessarily be the ones who suffer the most significant consequences from a new wave of automation. An initial substitution of workers with machines may have a ripple effect, with these workers then displacing other workers. In the worst-case scenario, in which machines perform a larger share of the tasks where individuals with high literacy and numeracy skills currently have a comparative advantage without creating new opportunities for them to use their advanced skills in the workplace, then these individuals will compete with less skilled workers for the remaining jobs.

Ultimately, for individuals and countries, investment in literacy and numeracy skills may pay off in one of two ways. In the most likely scenario, these individuals and economies will be the ones to benefit from new employment opportunities that result from innovative uses of AI. In the worst-case scenario in which few new opportunities emerge, it is the individuals (and countries) with low literacy and numeracy skills who are likely to suffer the most. Unless they can count on an entirely different set of skills (e.g. fine motor skills) to shield them from competition, they risk losing their comparative advantage over individuals with greater cognitive skills.

Technological disruptions, the climate crisis, population ageing and migration have the potential to reshuffle the distribution of income, wealth and power in societies, and to influence economic growth and collective well-being. The degree and the direction of such changes, however, are not predetermined: public policies and collective action can harness the forces behind these trends to reach desirable outcomes. The way policies are designed to respond to the complex questions and challenges raised will significantly influence individual and collective well-being in the long run.

Public policies can influence how the gains resulting from the use of AI are shared, and how the resulting income contributes to funding public expenditure – for example, by regulating the access and use of personal data and intellectual works to train AI applications. Public policies, combined with consumer choices and social dialogue, can rein in undesirable uses of AI technologies, such as scam industries, arms races in IT security or the creation of media that exploit psychological addiction mechanisms. They can also drive the direction of future AI research and facilitate the adoption of new technologies which best complement human capacities or overcome scarcity issues.

Public policies can also accelerate the green transition, or slow it down, through regulations, subsidies, taxes and direct investment. Perhaps more importantly, policy levers can maximise the efficiency of investment in the green transition to obtain the greatest possible reduction in emissions at the lowest possible economic and social cost.

Public policies can also create training and insurance policies to help workers at risk of losing their jobs (e.g. due to the introduction of regulations aimed at reducing emissions) in the transition to new occupations and encourage firms to invest in the training of all workers and improved working conditions to keep older workers in the workforce for longer. Public policies can also respond to ageing societies by reforming pension systems and promoting measures that boost birth rates.

In democratic societies, the responses to these challenges will depend on citizens’ preferences, in particular those expressed in elections and through collective action. The context in which policy preferences are formed, and collective decisions taken, has significantly changed over the past decade, due in particular to the rise of digital media and the increasing socio-economic and cultural diversity of modern societies.

In the decade that separated the two cycles of the Survey of Adult Skills, OECD countries have achieved near-universal access to the Internet. On average, the share of adults using the Internet increased from 76% in 2012 to 93% in 2023, with 87% of adults reporting using it daily (up from 61% in 2012; Figure 1.3). In 2022, 73% of adults reported accessing the Internet from a mobile device and 71% used it to read or download newspapers or news magazines (OECD, 2024[31]).

As a result, the role of traditional media (the printed press, radio and television) has declined, and adults increasingly use social media to get information and discuss and debate political opinions. In the European Union, 64% of adults reported using the Internet for reading online news sites or newspapers in 2023, and 18% used it for civic or political participation (Eurostat, 2024[32]). A recent survey by the Pew Research Center found that about half of adults in the United States got news from social media “sometimes” or “often” in 2022 (Pew Research Center, 2023[33]).

The rapid diffusion and circulation of information is not without risks. There is a concern that the increase in the sheer amount of information available has been accompanied by a decrease in the quality (OECD, 2023[34]). Internet platforms and social media also make it easier to spread false and misleading information. Some actors may do so deliberately to distort public debates and fuel polarisation. At times this may form part of a hybrid warfare tactic, to erode the social fabric of open societies and weaken their defences. Such disinformation campaigns have already been observed. They cast doubt on factual evidence and aggravate existing societal divisions, making it difficult to build the societal consensus essential to address complex policy challenges. A survey conducted by the Lloyd’s Register Foundation in 2019 found that 56% of adults in OECD countries were concerned about receiving false information online (World Risk Poll, 2019[35]).

Disinformation is not a new phenomenon, but modern digital communication technologies have fundamentally changed its reach and impact. Today, anyone with an Internet connection can produce and distribute content, without any responsibility to adhere to the journalistic or academic and scientific ethics and standards, which have been developed over the years to support information integrity. Generative AI tools may further amplify and strengthen the production of false and misleading content (OECD, 2024[36]).

The ease with which false and misleading information spreads can have very practical negative consequences. For example, misinformation may already be hindering action to improve the environment and fight climate change (Benegal and Scruggs, 2018[37]) and fake news can increase political polarisation, making political action more difficult (Tucker et al., 2018[38]). People appear to be over-confident in their ability to distinguish false news from true information (Corbu et al., 2020[39]).

On the other hand, this same accessibility provides unprecedented access to knowledge and can foster citizen engagement and innovative news reporting. In a survey conducted by the Pew Research Center in 19 countries (Australia, Canada, Israel, Japan, Korea, Malaysia, Singapore, the United States and 11 large European countries, including France, Germany, Italy and the United Kingdom) the majority of respondents tend to see social media as a positive thing for democracy (Pew Research Center, 2022[40]).

For democracies to function effectively, individuals must be able to actively seek out relevant information and to assess its quality. Without these skills, adults will not be able to form well-grounded opinions on the complex issues and challenges that policies are attempting to address. Together with other important foundations, knowledge and skills thus contribute to the successful operation of a democratic society.

Literacy and numeracy skills, in particular, determine how individuals access and process information about available options and their associated prospects, thereby influencing the decisions people ultimately make. These skills influence which media or content adults choose to access, and the extent to which they are able to understand and act upon the information they receive. Navigating this new world of online information requires the ability to understand and evaluate information in a written text. These are core cognitive processes that support proficiency in literacy, defined as the ability to “access, understand, evaluate and reflect on written texts” (Rouet et al., 2021[41]).

The role of literacy and numeracy skills in helping individuals orient themselves and reach their goals in the information landscape is a clear example of how the social benefits of skills go well beyond the private benefits. Greater skills do not just benefit the individuals who hold them. When it comes to how the public opinion is formed, and how this can influence electoral outcomes and policy decisions, having large shares of the population lacking the basic skills to effectively participate in complex social and democratic processes can have very important negative implications for the entire society, including for individuals with high levels of skills.

Items for the literacy assessment in the 2023 Survey of Adult Skills have purposefully been updated to better reflect the kinds of reading tasks that adults are likely to encounter in modern digital information environments. Similarly, the numeracy framework emphasises the importance of “using and reasoning critically with mathematical content” (Tout et al., 2021[42]), a crucial skill in a world where “data” and “scientific evidence” are often presented in partial ways in order to make deceptive arguments in support of a particular opinion or position. The specific mental processes that underpin literacy and numeracy need to be complemented by (more general) metacognitive skills. These are the ability to calibrate one’s comprehension of the problem, evaluate potential solutions and monitor progress towards the goals (OECD, 2021[43]) They are a core element of the assessment of adaptive problem solving (Greiff et al., 2021[44]). Indeed, effectively operating in complex information landscapes requires having knowledge about how information is generated and the limitations inherent to different information-generation processes, and also being aware of one’s own and other people’s cognitive limitations (OECD, 2023[34]).

Critical thinking helps citizens to identify and counter falsehoods, steer clear of disinformation campaigns, and defuse misleading narrations. Critical thinking involves assessing the strength and appropriateness of a statement, theory or idea through a questioning and perspective-taking process (Vincent-Lancrin, 2021[45]) – or, to put it differently, engaging in “reasonable reflective thinking focused on deciding what to believe or do” (Ennis, 2016[46]). Although critical thinking clearly has a strong cognitive base, it also relies on non-cognitive attitudes or dispositions (OECD, 2023[34]). The literature has stressed, for instance, the role of self-regulation; of an open, fair and reasonable mindset; of commitment to self-improvement, truth-seeking and curiosity; and the avoidance of cultural- or trait-induced bias and dichotomous thinking (Vardi, 2015[47]; Thomas and Lok, 2015[48]).

More generally, there are several indications that attitudes and dispositions, or social and emotional skills, have increased their relevance in recent years. Economic returns to social and emotional skills have grown faster than returns to cognitive skills in recent years (Deming, 2017[11]; Edin et al., 2022[12]) and this trend may well continue in the future, as generative AI increases its ability to perform non-routine cognitive tasks. But the importance of attitudes and social and emotional skills is likely to be even greater for society at large, as they are key for restoring (or building) the interpersonal and institutional trust and co-operation that are essential to the functioning of liberal democracies.

Modern societies are becoming increasingly diverse, bringing together people with very different experiences and backgrounds. The entry of new migrants intending to settle permanently in OECD countries reached a record high of 6.1 million in 2022, and the share of foreign-born residents across OECD countries increased from 8.9% in 2012 to 10.6% in 2022, with greater increases observed in countries already hosting a relatively large share of immigrants (OECD, 2023[49]). This suggests that diversity will increase further in the years to come.

The speed at which information travels across borders, potentially reaching every individual around the globe thanks to the Internet, also makes it more likely that citizens will be exposed to a wider variety of opinions and points of view. At the same time, and perhaps in reaction, many are finding refuge in homogeneous virtual communities, which in turn increase the polarisation of opinions and the refusal to consider other points of view (Suhay, Bello-Pardo and Maurer, 2017[50]; Bakshy, Messing and Adamic, 2015[51]; Hobolt, Lawall and Tilley, 2023[52]) Many of these virtual communities span borders, and may therefore introduce a disconnect between the community individuals identify themselves with and the community in which they live and whose collective decisions they can influence through voting and other forms of civic and political engagement.

Building effective trust and co-operation in this context requires developing attitudes and values to understand global and intercultural issues. These considerations are at the base of the numerous efforts made by the OECD to better understand the role of social and emotional skills, and how to measure and foster them. Specific examples in this sense include the Programme for International Student Assessment (PISA) assessment of global competence (OECD, 2019[53]) and the OECD Survey on Social and Emotional Skills (Chernyshenko, Kankaraš and Drasgow, 2018[54]). The Survey of Adult Skills gathers information on attitudes and dispositions toward learning and teamwork, albeit only indirectly (i.e. by relying on information about tasks performed on the job). The second cycle of the survey has also introduced an assessment of social and emotional skills, grounded in the Big Five taxonomy (Soto and John, 2017[55]; Kankaraš, 2017[56]) and administered within the background questionnaire of the survey.

The Survey of Adult Skills was born as a programme to assess the adult competencies that are considered relevant to modern societies. So far, the focus has mostly been on foundational cognitive skills. To remain relevant as societies evolve, the assessment must continue to stay attuned to changes in the types of tasks adults are confronted with, and for which they need to rely on literacy and numeracy skills. Looking ahead, the programme must stay attuned to shifts in skills demand in societies and remain open to adapting or expanding the set of skills and competencies that it tracks and assesses.

The 2023 Survey of Adult Skills assessments of literacy, numeracy and adaptive problem solving are based on conceptual frameworks that define what these skills are and describe how to design assessment items to measure them. The assessment tasks reflect how these skills are applied across a wide range of situations in adults’ lives. A prominent role is played by tasks embedded in data-intensive, complex digital environments, which are increasingly common in the workplace and everyday life in modern societies.

To this end, the assessment was exclusively administered on digital devices (tablets). This constitutes an important difference from the previous cycle of the survey, where respondents had the option to sit the assessment using paper-based instruments. While a statistical equivalence was established at the time between paper-based and computer-based instruments, the paper-based assessments were necessarily limited in the range of tasks they could accommodate.

Assessment items differ along various dimensions: cognitive processes (i.e. the mental strategies that form part of the skill in question), content (i.e. the artefacts, knowledge, representations and situations to which these cognitive processes are applied) and contexts (i.e. the settings in which the skill is used). A large number of items are needed to cover all aspects of a skill domain, as well as the variety of contexts in which adults are required to use that particular skill to solve tasks.

Literacy and numeracy were also assessed in the first cycle of the Survey of Adult Skills. The frameworks used in the second cycle build on past frameworks, but have been revised and extended to ensure relevance to contemporary reality and understanding of the phenomena measured (Rouet et al., 2021[41]; Tout et al., 2021[42]; Tout et al., 2017[57]). The links with the first cycle of the survey remain strong, both at the conceptual level and at the practical level, as the second cycle relies on many assessment items already used in the first cycle. Such common items allow strong psychometric links to be established across the two assessments.

In the 2023 Survey of Adult Skills, literacy is defined as “accessing, understanding, evaluating, and reflecting on written texts in order to achieve one’s goals, to develop one’s knowledge and potential, and to participate in society” (Rouet et al., 2021[41]). Proficiency in literacy is crucial for adults across their personal, social and professional spheres, given the prevalence of written communication in various aspects of life. Throughout the day, adults engage in a diverse range of reading activities, spanning from delving into extensive pieces of continuous text to swiftly scanning pages for pertinent information. Examples include reading emails, leaflets, timetables and instruction manuals.

An example of a literacy item is presented in Figure 1.4. In this item, readers must make inferences based on the information presented in the text in order to determine if a set of statements is true for bread, crackers or both. Respondents are asked to tap on a response for each of the presented statements. Only one response can be selected for each row.

In the 2023 Survey of Adult Skills, numeracy encompasses “accessing, using, and reasoning critically with mathematical content, information and ideas represented in multiple ways in order to engage in and manage the mathematical demands of a range of situations in adult life” (Tout et al., 2021[42]). The skills and knowledge needed for work and civic participation, and in more personal spheres of life, have changed. Individuals are presented with ever-increasing amounts of information of a quantitative or mathematical nature through online or technology-based resources, which have to be located, selected or filtered, interpreted, and at times questioned and doubted, and analysed for their relevance to the responses needed.

Figure 1.5 presents an example numeracy item, which relies on an interactive tool. For this item, the wallpaper calculator has already been used to determine the number of rolls needed. However, an error was made when one or more values were entered into the tool. The task is to identify the error(s) and enter the correct value(s).

Adaptive problem solving (APS) involves “the capacity to achieve one’s goals in a dynamic situation in which a method for solution is not immediately available. It requires engaging in cognitive and metacognitive processes to define the problem, search for information, and apply a solution in a variety of information environments and contexts” (Greiff et al., 2021[44]). The ability of adults to adapt to new circumstances and learn throughout life has likely become more important in complex modern societies, which are evolving at an accelerating pace (Greiff et al., 2017[58]). This motivated the inclusion of this new domain in the 2023 Survey of Adult Skills, replacing the assessment of problem solving in technology-rich environments administered in the first cycle of the survey.5

APS has three important features. First, it emphasises individuals’ capacity to flexibly and dynamically adapt their problem-solving strategies to a dynamically changing environment. Second, it tests their ability to identify and select among a range of available physical, social and digital resources. Third, individuals need to monitor and reflect on their progress in solving problems, through metacognitive processes.

An example item is presented in Figure 1.6. In it, an initially static situation becomes dynamic due to obstacles that change the problem and the solutions available. In the first item, the problem solver needs to use an interactive map to find the fastest route to accomplish three goals, keeping a set of time constraints in mind. The problem solver needs to take a child to school by a designated time, purchase groceries and return home by a designated time. The total driving time (shown at the bottom right of the screen) updates as the route is selected by the respondent. This could be considered a standard problem-solving task, in which a solution needs to be found given some constraints that need to be satisfied. In the second item, the situation becomes dynamic as the problem solver has to deal with new circumstances that interfere with the initial solution. The solver has to overcome impasses and take additional constraints into consideration when adapting the initial problem solution.

The 2023 Survey of Adult Skills assessments require adults to complete a set of tasks (also called assessment items),6 that can only be solved if they have a sufficient level of literacy, numeracy and adaptive problem solving skills. Their performance in these assessments is used to estimate their proficiency in each of these skill domains. These estimates are reported on three distinct 500-point scales. The same scales are used to classify assessment items according to their difficulty. At each point on the scale, an individual with a given level of proficiency has a 67% chance of successfully completing tasks located at that same level. Adults with a given level of proficiency will have a lower probability of being able to successfully complete more difficult tasks (those with higher values on the scale). They will similarly have a higher probability of successfully completing easier tasks. The relationship between adults’ proficiency and task difficulty is illustrated in Figure 1.7. For example, Adult C, with low proficiency, is unlikely to complete items II to IV and is also less likely to complete item I. Adult A, with high proficiency, is likely to successfully complete items I to V and probably also item VI.

It is important to note that literacy, numeracy and adaptive problem solving are separate skills, measured on distinct scales. This means that direct comparisons of literacy and numeracy scores, for example, are not meaningful. In other words, the fact that individuals (or a group) have a higher score in literacy than in numeracy does not allow to conclude that their literacy skills are higher than their numeracy skills. Any comparison of performance across different domains must necessarily be of a relative nature. For example, it can be said that individuals (or a group) are relatively better at literacy than numeracy if their rank in the ordered literacy distribution is higher their rank in the numeracy distribution.

Finally, it is worth noting and stressing that proficiency is necessarily linked to a specific language, the one used in the assessment items. The language of the assessment is, in most cases, the official language of the country or economy. However, participating countries were allowed to choose the assessment language, and some countries decided to administer the assessment in multiple languages.

The importance of this link between skills and language proficiency is particularly salient when looking at adults who lack sufficient mastery of the assessment language to take the assessment. These adults are present (although generally as a very small percentage of the population) in all countries and are defined as “literacy-related non-respondents”. In the first cycle of the Survey of Adult Skills, the proficiency of these adults was not estimated, which meant a country’s average results did not cover the entire adult population7. In order to collect more information on such adults, so as to allow their likely proficiency in the assessment language to be estimated, the 2023 Survey of Adult Skills introduced a new survey instrument, the doorstep interview (Box 1.1).

The Survey of Adult Skills directly assesses adults’ literacy, numeracy and adaptive problem solving skills. A basic command of these skills is thought to be necessary for fully integrating and participating in the labour market, education and training, and social and civic life. Results from the first cycle of the survey show that adults with more advanced mastery of these skills also benefit from greater earnings, well-being, agency and prestige (OECD, 2013[6]). Literacy, numeracy and adaptive problem solving skills are domain-general and highly transferable, and relevant to many social contexts and work situations; they can also be learned and are therefore subject to the influence of policy. Understanding the level and distribution of these skills among adult populations in participating countries is thus important for policy makers in a range of social and economic policy areas. To this end, Chapter 2 provides a descriptive, comparative analysis of the distribution of skills within the adult population. Chapter 2 also addresses the question of who has low, medium or high proficiency in literacy, numeracy and adaptive problem solving in 2023.

The vast majority of countries and economies that participated in the 2023 Survey of Adult Skills also took part in its first cycle, and some of them also participated in previous international assessments of adult skills. The evolution of the information-processing skills of their adult populations can therefore be tracked over time, although some caution is always advised in these analyses because of changes in the design, implementation and methods of the surveys. As each survey covers all adults born during a period of almost 50 years, more than three-quarters of participants in the most recent Survey of Adult Skills would have also been eligible to participate in the first cycle of the survey. In other words, the populations behind the statistical indicators presented in this report are largely the same populations whose skills were assessed in previous editions. This makes it possible not just to compare the results of different age groups with those of the same age in the previous survey but, for older cohorts, to also compare them with the same cohort at a younger age.

Skills are not fixed for life, at any age. Comparisons such as these, presented in Chapter 3, can offer insight into how well governments, firms and other actors, including individuals themselves, have been able to retain and expand the information-processing skills of the adult population, through investment in formal and informal education and training, and how recent migration flows have contributed to skill dynamics.

Additional comparisons, based on background questionnaires that measure adults’ participation in training and readiness to learn, can also provide insights into the effectiveness of skill policies. Future reports will analyse adults’ participation in non-formal and informal learning and how it has changed over time, looking at the nature of such training and barriers to accessing it.

At the same time, inward and outward migration means that the composition of the eligible population will have changed over time. Not all these compositional changes can be fully accounted for when analysing the evolution of results over time. For example, it is possible to restrict comparisons to individuals born in the country, in an attempt to neutralise the impact of inward migration. However, little can be done to control for outward migration.

Future reports will also analyse the skills of adults with an immigrant background in greater depth – e.g. to explore the extent to which gender gaps differ depending on where adults were born and went to school, or how the skills of foreign-born adults evolved over time, and in comparison to those of native-born adults of the same age.

For adults who completed their initial education more recently, the results can also indicate how well this has equipped them with a solid foundation to engage in further learning throughout their life. Comparing the results of younger cohorts with those of young adults surveyed in the first cycle can complement the information about trends among 15-year-olds who were assessed in related domains (reading and mathematics) in PISA. The analysis of changes in the scores of young adults across cycles, presented in Chapter 3, forms the basis for such comparisons.

For countries that participated in the first cycle of the survey, it is also possible to compare the growth in skills between those measured by PISA among 15-year-olds and those of 19-25 year-olds who participated in either cycle of the Survey of Adult Skills. Only participants in the most recent cycle will have been affected by the disruptions to education and early career experiences due to the COVID-19 pandemic. With careful comparison, it may therefore also be possible to understand whether countries were able to successfully mitigate lost learning opportunities during the pandemic, or whether the negative shock experienced by young adults in 2020-21 is likely to have long-lasting consequences for their careers.

Not all adults participate in the workforce, and some of those who do may not be using their skills to their full extent. Based on the belief that skill requirements are rapidly evolving and cannot be adequately conveyed through job titles, the Survey of Adult Skills has continued to expand its indicators on the use of skills in the workplace. Chapter 4 examines how skills relate to labour-market participation and wages, and the extent of skills mismatches. Future reports will also analyse how the use of skills have evolved in response to the adoption of new technologies and other trends that have changed the demand for skills over the past decade. In particular, the Survey of Adult Skills allows changes in workers’ tasks within and across occupations to be documented, providing a picture of job requirements that could reflect the emerging impact of automation or changes in workplace practices triggered by the COVID-19 pandemic.

A number of studies have documented the growing importance of interpersonal skills, such as teamwork and leadership, for labour-market success in the 21st century. Data from the 2023 Survey of Adult Skills will enable analysis of how differences in social and emotional skills relate to information-processing skills, economic outcomes, lifelong learning dispositions and behaviour, and past educational qualifications. This will be the focus of future reports.

In an era of rapid changes in skill requirements, and of longer working lives, the need to invest in lifelong learning takes on an added urgency. Workers, employers and unions need to be aware of this imperative. Data from the Survey of Adult Skills can document inequality in access to and participation in adult learning. In particular, they show how the uptake of formal, non-formal and informal learning opportunities varies across demographic groups and levels of educational attainment, but also how literacy, numeracy or social and emotional skills relate to adult behaviour and dispositions towards learning. This will be the focus of future reports.

Through its background questionnaire, the Survey of Adult Skills assesses more than economic well-being and labour-market outcomes. It also collects data on social and civic dimensions of well-being and self-assessed health status. These data can be used to explore the nexus between skills, skill development and well-being – both in cross-sectional comparisons (at a single point in time) and in cohort-level, longitudinal comparisons. These questions are partly covered in Chapter 4 of this report but will be analysed in greater depth in future reports.

[22] Acemoglu, D. (2024), “The simple macroeconomics of AI”, NBER Working Paper, No. 32487, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w32487.

[21] Acemoglu, D. and P. Restrepo (2019), “The wrong kind of AI? Artificial intelligence and the future of labour demand”, Cambridge Journal of Regions, Economy and Society, Vol. 13/1, pp. 25-35, https://doi.org/10.1093/cjres/rsz022.

[30] Autor, D. (2024), “Applying AI to rebuild middle class jobs”, NBER Working Paper, No. 32140, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w32140.

[7] Autor, D., F. Levy and R. Murnane (2003), “The skill content of recent technological change: An empirical exploration”, The Quarterly Journal of Economics, Vol. 118/4, pp. 1279-1333, https://doi.org/10.1162/003355303322552801.

[9] Autor, D. and B. Price (2013), The changing task composition of the US labor market: An update of Autor, Levy and Murnane (2003), https://economics.mit.edu/sites/default/files/publications/the%20changing%20task%20comp%202013.pdf (accessed on 21 November 2024).

[51] Bakshy, E., S. Messing and L. Adamic (2015), “Exposure to ideologically diverse news and opinion on Facebook”, Science, Vol. 348/6239, pp. 1130-1132, https://doi.org/10.1126/science.aaa1160.

[3] Barro, R. (2001), “Human capital and growth”, American Economic Review, Vol. 91/2, pp. 12-17, https://doi.org/10.1257/aer.91.2.12.

[37] Benegal, S. and L. Scruggs (2018), “Correcting misinformation about climate change: The impact of partisanship in an experimental setting”, Climatic Change, Vol. 148/1-2, pp. 61-80, https://doi.org/10.1007/s10584-018-2192-4.

[5] Benton, L. and T. Noyelle (1992), Adult Illiteracy and Economic Performance, Centre for Educational Research and Innovation, OECD Publishing, Paris.

[27] Brynjolfsson, E., D. Li and L. Raymond (2023), “Generative AI at work”, NBER Working Paper, No. 31161, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w31161.

[54] Chernyshenko, O., M. Kankaraš and F. Drasgow (2018), “Social and emotional skills for student success and well-being: Conceptual framework for the OECD study on social and emotional skills”, OECD Education Working Papers, No. 173, OECD Publishing, Paris, https://doi.org/10.1787/db1d8e59-en.

[39] Corbu, N. et al. (2020), “‘They can’t fool me, but they can fool the others!’ Third person effect and fake news detection”, European Journal of Communication, Vol. 35/2, pp. 165-180, https://doi.org/10.1177/0267323120903686.

[1] Deming, D. (2022), “Four facts about human capital”, Journal of Economic Perspectives, Vol. 36/3, pp. 75-102, https://doi.org/10.1257/jep.36.3.75.

[24] Deming, D. (2021), “The growing importance of decision-making on the job”, NBER Working Paper, No. 28733, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w28733.

[13] Deming, D. (2017), Replication Data for: “The Growing Importance of Social Skills in the Labor Market”, Harvard Dataverse, https://doi.org/10.7910/DVN/CYPKZH.

[11] Deming, D. (2017), “The growing importance of social skills in the labor market”, The Quarterly Journal of Economics, Vol. 132/4, pp. 1593-1640, https://doi.org/10.1093/qje/qjx022.

[12] Edin, P. et al. (2022), “The rising return to noncognitive skill”, American Economic Journal: Applied Economics, Vol. 14/2, pp. 78-100, https://doi.org/10.1257/app.20190199.

[46] Ennis, R. (2016), “Critical thinking across the curriculum: A vision”, Topoi, Vol. 37/1, pp. 165-184, https://doi.org/10.1007/s11245-016-9401-4.

[32] Eurostat (2024), ICT usage in households and by individuals, Eurostat Data Browser, https://ec.europa.eu/eurostat/databrowser/view/isoc_ci_ac_i__custom_11597620/default/table?lang=en (accessed on 30 May 2024).

[20] Georgieff, A. and A. Milanez (2021), “What happened to jobs at high risk of automation?”, OECD Social, Employment and Migration Working Papers, Vol. 255, https://doi.org/10.1787/10bc97f4-en.

[25] Green, A. (2024), “Artificial intelligence and the changing demand for skills in the labour market”, OECD Artificial Intelligence Papers, No. 14, OECD Publishing, Paris, https://doi.org/10.1787/88684e36-en.

[44] Greiff, S. et al. (2021), “PIAAC Cycle 2 assessment framework: Adaptive problem solving”, in The Assessment Frameworks for Cycle 2 of the Programme for the International Assessment of Adult Competencies, OECD Publishing, Paris, https://doi.org/10.1787/3a14db8b-en.

[58] Greiff, S. et al. (2017), “Adaptive problem solving: Moving towards a new assessment domain in the second cycle of PIAAC”, OECD Education Working Papers, No. 156, OECD Publishing, Paris, https://doi.org/10.1787/90fde2f4-en.

[29] Hanushek, E. et al. (2017), “Coping with change: International differences in the returns to skills”, Economics Letters, Vol. 153, pp. 15-19, https://doi.org/10.1016/j.econlet.2017.01.007.

[4] Hanushek, E. and L. Woessmann (2008), “The role of cognitive skills in economic development”, Journal of Economic Literature, Vol. 46/3, pp. 607-668, https://doi.org/10.1257/jel.46.3.607.

[61] Heckman, J. and S. Mosso (2014), “The economics of human development and social mobility”, Annual Review of Economics, Vol. 6/1, pp. 689-733, https://doi.org/10.1146/annurev-economics-080213-040753.

[52] Hobolt, S., K. Lawall and J. Tilley (2023), “The polarizing effect of partisan echo chambers”, American Political Science Review, Vol. 118/3, pp. 1464-1479, https://doi.org/10.1017/s0003055423001211.

[10] Ikenaga, T. and R. Kambayashi (2016), “Task polarization in the Japanese labor market: Evidence of a long‐term trend”, Industrial Relations: A Journal of Economy and Society, Vol. 55/2, pp. 267-293, https://doi.org/10.1111/irel.12138.

[56] Kankaraš, M. (2017), “Personality matters: Relevance and assessment of personality characteristics”, OECD Education Working Papers, No. 157, OECD Publishing, Paris, https://doi.org/10.1787/8a294376-en.

[18] Lane, M. and A. Saint-Martin (2021), “The impact of Artificial Intelligence on the labour market: What do we know so far?”, OECD Social, Employment and Migration Working Papers, No. 256, OECD Publishing, Paris, https://doi.org/10.1787/7c895724-en.

[17] Lassébie, J. and G. Quintini (2022), “What skills and abilities can automation technologies replicate and what does it mean for workers? New evidence”, OECD Social, Employment and Migration Working Papers, No. 282, OECD Publishing, Paris, https://doi.org/10.1787/646aad77-en.

[2] Mankiw, N., D. Romer and D. Weil (1992), “A contribution to the empirics of economic growth”, The Quarterly Journal of Economics, Vol. 107/2, pp. 407-437, https://doi.org/10.2307/2118477.

[28] Noy, S. and W. Zhang (2023), “Experimental evidence on the productivity effects of generative artificial intelligence”, Science, Vol. 381/6654, pp. 187-192, https://doi.org/10.1126/science.adh2586.

[36] OECD (2024), Facts not Fakes: Tackling Disinformation, Strengthening Information Integrity, OECD Publishing, Paris, https://doi.org/10.1787/d909ff7a-en.

[31] OECD (2024), ICT access and usage by households and individuals, OECD Telecommunications and Internet Statistics (database), https://doi.org/10.1787/b9823565-en (accessed on 29 May 2024).

[23] OECD (2024), OECD Employment Outlook 2024: The Net-Zero Transition and the Labour Market, OECD Publishing, Paris, https://doi.org/10.1787/ac8b3538-en.

[59] OECD (2024), Survey of Adult Skills 2023: Reader’s Companion, OECD Skills Studies, OECD Publishing, Paris, https://doi.org/10.1787/3639d1e2-en.

[16] OECD (2023), Is Education Losing the Race with Technology?: AI’s Progress in Maths and Reading, Educational Research and Innovation, OECD Publishing, Paris, https://doi.org/10.1787/73105f99-en.

[15] OECD (2023), OECD Employment Outlook 2023 : Artificial Intelligence and the Labour Market, OECD Publishing, https://doi.org/10.1787/08785bba-en.

[49] OECD (2023), “Recent developments in international migration movements and labour market inclusion of immigrants”, in International Migration Outlook 2023, OECD Publishing, Paris, https://doi.org/10.1787/3da593b1-en.

[34] OECD (2023), “Skills and attitudes for new information landscapes”, in OECD Skills Outlook 2023: Skills for a Resilient Green and Digital Transition, OECD Publishing, Paris, https://doi.org/10.1787/fe52b3c2-en.

[43] OECD (2021), The Assessment Frameworks for Cycle 2 of the Programme for the International Assessment of Adult Competencies, OECD Skills Studies, OECD Publishing, Paris, https://doi.org/10.1787/4bc2342d-en.

[14] OECD (2019), Artificial Intelligence in Society, OECD Publishing, Paris, https://doi.org/10.1787/eedfee77-en.

[19] OECD (2019), “Overview: The future of work is in our hands”, in OECD Employment Outlook 2019: The Future of Work, OECD Publishing, Paris, https://doi.org/10.1787/e4718721-en.

[53] OECD (2019), “PISA 2018 Global Competence Framework”, in PISA 2018 Assessment and Analytical Framework, OECD Publishing, Paris, https://doi.org/10.1787/043fc3b0-en.

[6] OECD (2013), OECD Skills Outlook 2013: First Results from the Survey of Adult Skills, OECD Publishing, Paris, https://doi.org/10.1787/9789264204256-en.

[60] OECD (forthcoming), Survey of Adult Skills 2023 Technical Report, OECD Publishing, Paris.

[26] Peng, S. et al. (2023), The Impact of AI on Developer Productivity: Evidence from GitHub Copilot, arXiv, https://doi.org/10.48550/arXiv.2302.06590.

[33] Pew Research Center (2023), “Social media and news fact sheet”, Fact Sheet, Pew Research Center, https://www.pewresearch.org/journalism/fact-sheet/social-media-and-news-fact-sheet/ (accessed on 30 May 2024).

[40] Pew Research Center (2022), Social Media Seen as Mostly Good for Democracy Across Many Nations, But U.S. is a Major Outlier, Pew Research Center, https://www.pewresearch.org/wp-content/uploads/sites/20/2022/12/PG_2022.12.06_Online-Civic-Engagement_REPORT.pdf (accessed on 30 May 2024).

[41] Rouet, J. et al. (2021), “PIAAC Cycle 2 assessment framework: Literacy”, in The Assessment Frameworks for Cycle 2 of the Programme for the International Assessment of Adult Competencies, OECD Publishing, Paris, https://doi.org/10.1787/7b3bf33b-en.

[55] Soto, C. and O. John (2017), “The next Big Five Inventory (BFI-2): Developing and assessing a hierarchical model with 15 facets to enhance bandwidth, fidelity, and predictive power.”, Journal of Personality and Social Psychology, Vol. 113/1, pp. 117-143, https://doi.org/10.1037/pspp0000096.

[8] Spitz‐Oener, A. (2006), “Technical change, job tasks, and rising educational demands: Looking outside the wage structure”, Journal of Labor Economics, Vol. 24/2, pp. 235-270, https://doi.org/10.1086/499972.

[50] Suhay, E., E. Bello-Pardo and B. Maurer (2017), “The Polarizing Effects of Online Partisan Criticism: Evidence from Two Experiments”, The International Journal of Press/Politics, Vol. 23/1, pp. 95-115, https://doi.org/10.1177/1940161217740697.

[48] Thomas, K. and B. Lok (2015), “Teaching critical thinking: An operational framework”, in The Palgrave Handbook of Critical Thinking in Higher Education, Palgrave Macmillan US, New York, https://doi.org/10.1057/9781137378057_6.

[57] Tout, D. et al. (2017), Review of the PIAAC Numeracy Assessment Framework: Final Report., Australian Council for Education Research (ACER), https://research.acer.edu.au/transitions_misc/29 (accessed on 21 November 2024).

[42] Tout, D. et al. (2021), “PIAAC Cycle 2 assessment framework: Numeracy”, in The Assessment Frameworks for Cycle 2 of the Programme for the International Assessment of Adult Competencies, OECD Publishing, Paris, https://doi.org/10.1787/c4221062-en.

[38] Tucker, J. et al. (2018), “Social media, political polarization, and political disinformation: A review of the scientific literature”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3144139.

[47] Vardi, I. (2015), “The relationship between self-regulation, personal epistemology, and becoming a ’critical thinker’: Implications for pedagogy”, in The Palgrave Handbook of Critical Thinking in Higher Education, Palgrave Macmillan US, New York, https://doi.org/10.1057/9781137378057_13.

[45] Vincent-Lancrin, S. (2021), Skills for Life: Fostering Critical Thinking, Inter-American Development Bank, https://doi.org/10.18235/0003743.

[35] World Risk Poll (2019), The Lloyd’s Register Foundation World Risk Poll, https://wrp.lrfoundation.org.uk/ (accessed on 30 May 2024).