See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/362861163

Artificial Intelligence, 21 st Century Competences, and Socio- Emotional

Learning in Education: More than High-Risk?

Preprint · August 2022

DOI: 10.13140/RG.2.2.13681.15209

CITATIONS READS

0 1,506

1 author:

Ilkka Tuomi
Meaning Processing
68 PUBLICATIONS 2,943 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ilkka Tuomi on 23 August 2022.

The user has requested enhancement of the downloaded file.

 Artificial Intelligence, 21st Century Competences, and Socio-
Emotional Learning in Education: More than High-Risk?

Ilkka Tuomi
Meaning Processing Ltd., Arkadiankatu 20 A 20, 00100 Helsinki, Finland

Forthcoming in the European Journal of Education, special issue on the Futures of Artificial Intel-
ligence in Education, vol 57(4)

Abstract
Over the last two decades, 21st century competences and socio-emotional skills have become a
major focus in educational policy. In this article, skills for the 21st century, soft skills, as well as
social and emotional skills, are contextualised in the context of technological change, machine
learning, and the ethics of artificial intelligence. The use of data-driven AI technologies to model
and measure these skills—in this article defined as non-epistemic competence components—can
lead to major social challenges that have important implications for educational policies and
practices. A moratorium on the use of data on these competence components in machine learning
systems is proposed until the society-wide impact is better understood.

KEYWORDS: AIED, non-epistemic competence, regulation of AI, social impact of machine

learning, socio-emotional learning, future of education

1 INTRODUCTION
There exists now a strong policy-level agreement on the importance of 21st century competences
and social and emotional skills for educational, social, economic, and other life outcomes
(Chernyshenko et al., 2018; Kankaraš, 2017). The field, however, is also described using “a
dizzying array of terms, definitions, and measures” (Soto et al., 2022, p. 193). A recent extensive
review found 136 socio-emotional competence frameworks (Berg et al., 2017), and many
alternative terms have been used to characterise these skills and competences, including
‘noncognitive skills,’ ‘character skills,’ ‘transferable skills,’ and ‘soft skills’ (Abrahams et al., 2019).
The terms competence and skill are often used interchangeably in the literature, and the commonly
used term noncognitive skills—intended to comprise personality traits, attitudes, and motivations—
is acknowledged to be unfortunate because social, emotional, and other ‘soft’ skills are influenced
and depend on human cognition (Borghans et al., 2008).

In contrast to skills and knowledge that have traditionally been the focus of education and training,
21st century competences and socio-emotional skills are closely related to individual personality
traits and characteristics. For example, a recent major initiative in this area, the OECD Study on
Social and Emotional Skills (Kankaraš & Suarez-Alvarez, 2019) organises these skills based on the
Big Five personality domains (John & Srivastava, 1999). The original Big Five model defined five
top-level factors: (1) Openness to Experience, (2) Conscientiousness, (3) Extraversion, (4)
Agreeableness, and (5) Neuroticism, and the resulting Five Factor Theory of personality was built
on the postulate that personality traits are highly stable and biologically determined (McCrae &

1
Costa, 1999). Such a view, clearly, challenges policies and educational interventions that aim at
developing these skills.

Recent research suggests that there are age and life-event-related changes in the average levels of
these skills, and that educational interventions may have some limited impact on their change
(Borghans et al., 2008; Heckman & Kautz, 2012). In fact, social, emotional, noncognitive, and soft
skills are often defined as skills that are susceptible to interventions and policy measures, especially
during the early years of life (Chernyshenko et al., 2018). Despite defining these skills as malleable,
it is also known that they show predictable patterns of change across age groups (Roberts et al.,
2006), and the relative arrangement of individuals along personality traits (known as rank-
consistency) is surprisingly stable over time (Roberts & DelVecchio, 2000). Empirical studies,
however, are also known to provide ambiguous evidence because these skills and personal qualities
are measured in many different ways using varying conceptual frameworks (Duckworth & Yeager,
2015), and, despite hundreds of studies, high-quality scientific evidence is still rare (Smithers et al.,
2018).

When the emerging possibilities of artificial intelligence (AI) are appropriated to model 21st century
competences and social and emotional skills, important and qualitatively new ethical challenges
emerge for education policy and practice. To understand these, in this article a new conceptual
framing of 21st century competences and social and emotional skills is proposed that distinguishes
epistemic and non-epistemic competence components. Non-epistemic competence components are
shown to be related to personality characteristics that are, on average, highly stable and difficult or
impossible to change after preschool age. Although it is now commonly claimed that, for example,
social and emotional skills are malleable and can be influenced by interventions, in this article I
adopt a more literal reading of existing research that emphasises the possibility to predict life
outcomes using data on non-epistemic competence components. Deconstructing competence into its
epistemic and non-epistemic components is useful from the point of education policy and learning
theory, but it also makes explicit the social and technological dimensions of competence.
Competent action requires integration of all these four components of competence.

Recent developments in artificial intelligence have shown that, given enough data, machine learning
can recover often surprising associations that can effectively be used for prediction. The
developments in AI over the last two decades have to a large extent been driven by commercial
interests in modeling users and customers based on their online behaviour. A core technology for
major AI-based platforms, such as Google, Amazon, Netflix, and Facebook, is social clustering
based on behavioural similarity. Recent breakthroughs in data-driven AI, including in image
processing and natural language processing, now provide novel ways to analyse and model personal
characteristics. For example, a recent study successfully used colours in social media images to
predict the Big Five personality traits (Khorrami et al., 2022). In learning analytics, students’
interactions with the learning platform and the generated trace data have been used to map students
to the Big Five traits, as well as to their characteristic learning strategies (Matcha et al., 2020). In
general, there has been increasing interest in using AI in education (AIED) for the development of
non-epistemic competence components (Joksimovic et al., 2020), including motivation (du Boulay,
2018), emotion (Harley et al., 2017), social interaction skills (Porayska-Pomsta et al., 2018), as well
as metacognition and self-regulated learning (Azevedo et al., 2019).

2
The analysis presented in this article draws conclusions from research on personality psychology,
economics of personal characteristics, and AI to show that the use of machine learning technologies
to model and measure non-epistemic competence components may have important social
consequences in the near future. Although the argument is based on existing research, the expected
futures of AI and AIED are also considered, and the argument, therefore, is to some extent
speculative. Although many caveats are necessary, the outline of the argument in this article is
simple. Innovation, globalisation and access to knowledge—and information and communications
technologies that underpin these—reduce the importance of epistemic competence components and
increase the social and economic importance of non-epistemic ones (Tuomi, 2015). Various terms,
such as 21st century competences, soft and noncognitive skills, and social and emotional skills have
been used in the literature to capture the essence of these competence components that have now
become important. Non-epistemic competence components, however, are closely related to
personality characteristics that are, on average, highly stable and difficult or impossible to change
after preschool age. Although important caveats remain—and will be discussed later in this paper—
research further shows that non-epistemic competence components can be used to predict life
outcomes. Machine learning models can potentially predict life outcomes better than current
statistical models, also using personal data collected in the first years of life. Although these
predictions may not always be accurate at the individual level, they may have a profound social
impact. As the importance of non-epistemic competence components has been widely recognised,
data on these are also increasingly collected in educational settings. The emerging technical
opportunities to sort individuals using data on non-epistemic competence components, therefore,
generate important challenges for education policy and the ethics of AI, in and beyond education.

The argument presented in this article illustrates the point that existing AI ethics frameworks
inadequately address developmental and society-wide concerns that are central to education. For
example, the proposed EU AI Act (EC, 2021), which focuses on safety, security, human rights, and
trustworthiness from an individual point of view, misses the long-term systemic challenges
identified in this article.

A moratorium is proposed because theoretical understanding in this area is still evolving, and there
are strong interests in using data on non-epistemic competence components in machine learning
models. A call for a moratorium may sound dramatic or alarmist but is intended to allow researchers
and policymakers time to stop and think about the potentially profound consequences of using
machine learning in this domain. Social classification and categorisation provide the foundations for
societies, and the effective use of data on personal characteristics imply a tectonic change in these
foundations.1 The combination of state-of-the-art machine learning approaches with data on
personal characteristics is potentially leading to a ‘Cambrian explosion’ in the social sciences and
psychological research. 21st century competences, and noncognitive, social and emotional skills are
rapidly becoming a central topic in education and policy, and AI-based systems that use data on
personal characteristics are now widely used, for example, in recruiting (Hunkenschroer & Luetge,
2022), counseling and guidance (Tuomi et al., 2021), and increasingly also in AIED (du Boulay,
2019; Joksimovic et al., 2020). If empirical research on personality psychology, life outcomes, and
economics of personality traits is accurate, and a small number of measurable personal qualities can
be used to characterise individuals, these qualities most probably are currently being recovered by
data-driven AI systems used to predict user behaviour on the internet. Even if the genie has already
been let out of the bottle, it is important to consider how the potential harms could be alleviated.

3
The resulting ethical challenges have little to do with technology design or intended use, for
example, algorithmic fairness (Kizilcec & Lee, 2022; Mitchell et al., 2021), unacceptable biases
(Baker & Hawn, 2021; Kirkpatrick, 2016), or explainability and transparency (AI HLEG, 2019;
Barredo Arrieta et al., 2020; Biran & Cotton, 2017; Mittelstadt et al., 2019). Ethical challenges
emerge here as a society-wide interaction of new information infrastructures and machine learning
systems that use data available on these infrastructures. Personal characteristics have always been
socially important; data-driven AI simply makes them more important than before as technology
allows us to use knowledge and data more efficiently. The ethical challenge, therefore, can not be
found from the characteristics of any specific AI system; instead, it is rooted in the general
predictive capacity of data-driven AI. In contrast to much research on how AI-based prediction can
go wrong, the present case shows what can happen if the system works as intended.

The paper is organised as follows. Section 2 discusses different conceptualisations of competence

and skill and suggests that 21st century competences and social and emotional skills should be
understood as non-epistemic competence components. Section 3 then focuses on the nature of these
non-epistemic components building on Roberts’ (2006) neo-socioanalytic model of personality.
Section 4 reviews research on the malleability of non-epistemic competence components and notes
the importance of the stable core personality for AI-based prediction. Section 5 discusses data-
driven AI and non-epistemic competence components and highlights some limitations of the
presented argument and current knowledge. Section 6 puts these discussions in the context of
education policy and the ethics of AI. The paper ends with a short concluding section with a
suggestion that a new framing may be necessary to realise the potential of future AIED for non-
epistemic learning.

2 STRUCTURAL CHANGE IN COMPETENCES AND THE BIRTH OF 21st CENTURY

SKILLS
The use of the term competence has proliferated in different contexts during the last decades. One
source of this popularity is its use in strategic management literature in the 1990s, from where it has
spread to many core competence frameworks used in formulating capabilities that are needed in
work and life. Such frameworks range from the OECD Core Competence Framework on financial
literacy (OECD, 2015) to the UNESCO Core Competency Framework for its staff (UNESCO,
2016), and the EU Key Competence Framework for Lifelong Learning (Council of the European
Union, 2018).

Despite the proliferation of the term, the nature of competences is often addressed only
superficially. For example, the recent UNESCO Competence Framework for Cultural Heritage
Management (UNESCO, 2020) explains parenthetically that the framework identifies areas of
competence as skills and knowledge. UNESCO’s Competency Framework for UNESCO employees
(UNESCO, 2016) uses a slightly broader definition that constructs competences as a set of related
knowledge, skills and abilities that resulting in essential behaviours expected from those working
for the organisation. OECD’s internal Competency Framework (OECD, 2014) separates technical
competences from core competences that are important across all OECD jobs. These, according to
OECD, include competences such as analytical thinking, achievement focus, flexible thinking,
teamwork and team leadership, diplomatic sensitivity, influencing, negotiation and strategic
thinking. According to the European Commission, “competences include more than knowledge and
understanding and take into account the ability to apply that when performing a task (skill) as well

4
as how – with what mind-set – the learner approaches that task (attitude)” (EC, 2018, p. 4). In
European Union literature, competence thus can often be understood as an intersection between
knowledge and skill related components and attitudes.

The literature on skills and competences often distinguishes technical domain-specific skills and
functional or generic skills. Skills have been associated with individuals, occupations and jobs, and
also argued to be distributed among persons and technology (Attewell, 1990). Competence, in turn,
has been conceptualised as a combination of cognitive, conative, and affective components (Snow
et al., 1996). Cognitive elements include various forms of knowledge and intelligence. The conative
dimension, in turn, captures motivational and intentional aspects such as attention, grit and
executive control of behaviour, whereas the affective dimension of competence captures the
emotional and temperamental aspects of competence.

Many AI-based systems now use extensive skill taxonomies built using natural language processing
of on-line job advertisements (e.g., Australian Government, 2019; CEDEFOP, 2019; ILO, 2020;
Squicciarini & Nachtigall, 2021). At the time of writing, the European Skills, Competences,
Qualifications and Occupations (ESCO) skills pillar includes some 14 000 skills (EC, 2022). The
widely used US government sponsored occupational information resource website O*Net OnLine
(O*NET, 2022) provides data on job-related basic content skills, basic process skills, such as
critical thinking and active learning, and cross-functional skills, including social skills, complex
problem solving skills and technical skills.

Simplifying this complex picture, we can depict competence as a combination of two qualitatively
different components. One is the epistemic component that includes knowledge, domain specific
skill and experience. This can be summarised using the term expertise. As new knowledge and
domain-specific skills can be learned and experience gained, this competence component has been
the traditional focus of education, vocational training, and intelligent tutoring systems in AIED.

To become realised as an expressed capability to get things done, this epistemic component needs to
be complemented with non-epistemic elements. These include the ‘attitudes,’ ‘dispositions,’ and
‘noncognitive’, ‘soft’ and ‘21st century’ skills that have also been called the “behavioural repertoire”
(Hoekstra & Van Slujis, 2003). More broadly, as discussed in more detail in the next section, they
also include cognitive aptitudes and capacities. These non-epistemic elements capture many of the
things commonly called transversal, generic, or core competences and skills, including analytical
and critical thinking, communication skills, emotional intelligence, leadership skills, curiosity,
openness to experience, grit, and learning skills. The list is long. It can be shortened by labelling it
as ‘21st century skills,’ or ‘social and emotional skills.’ In this article, an attempt is made to capture
the essence of such skills, and how they differ from the focus of mainstream 20th century education,
by terming these skills the non-epistemic components of competence.

Competent action often occurs in situations where the current material and cultural contexts can be
taken for granted. Historically, social institutions—including education, the organisation of work,
and stocks of knowledge and expertise—have been able to stabilise after new key technologies have
been introduced (Freeman & Louçã, 2001); in effect, becoming the invisible background where
skills and competences are distinguished. Domain-specific skills and knowledge, therefore, can
often be viewed as a reflection of the prevailing technological and material context (Tuomi, 2020;
for an example, see Kotamraju, 1999). In this sense, a car generates the skills of a car-mechanic, a
5
programmable computer generates the skills of a software programmer and computational thinking,
and an anvil creates a blacksmith. In the industrial era, the resulting skill sets were stable enough to
be associated with professions, areas of expertise, and educational qualifications. Occupational
skills, as listed in taxonomies—such as provided for the US on the O*Net website and for Europe
by ESCO—therefore, are conceptually mirror images of current technology. When technology
changes, many of these skills become obsolete, and policy focus shifts to ‘21st century
competences,’ now commonly understood as social, emotional, and transversal skills.

The capability of getting things done depends on the material and technical environment as well as
human collaboration and coordination. At one extreme, actor-network theory and science and
technology studies argued for the symmetry between human and non-human actors (Callon et al.,
1986).2 Research on socially distributed and situated cognition (e.g., Brown et al., 1989; Hutchins,
1995) and cultural-historical activity theory (e.g., Engeström et al., 1999; Leont’ev, 1978), in turn,
emphasised the interactions between human cognition, culture, and tools and technologies used in
action. In general, a sociological and organisation theoretic view on competence would highlight
the social and cultural constraints, norms, rules, and resources that are necessary for human action
and agency, whereas sociology of technology and science would align with Vygotsky (Vygotsky &
Luria, 1994), and point out that—beyond socially shared conceptual systems—the capability to get
things done depends on instruments, tools, and technology.

Both in activity theory and in general sociology, the cultural context provides systems of value that
make action meaningful and socially intelligible and useful. It is easy to see that in a world where
cultural contexts lose their stability and activities become connected across cultures, the importance
of communication, collaboration, and intercultural skills increases. Such a cultural change, of
course, is deeply linked to technical change and, in particular, communications technology. In
contrast to the classical sociological theories of modernisation, intercultural collaboration cannot
anymore be reduced to transactions across geographies (Durkheim, 1933), or even global flows of
information (Castells, 1996). To be able to act competently, the actors need to mobilise social
resources across cultural boundaries and systems of meaning. As a result of this process, and the
continuous innovation that results from it, also the technological and cultural contexts, previously
taken for granted, now become visible.

The above overview shows that there are important social, economic, and historical reasons for the
increasing attention to the non-epistemic components of competence. In this sense, the focus on 21st
century competences was a reaction to a global social transformation, commonly known as the
information or knowledge society, that described a disruptive historical period before recent
advances in AI. The following elaborates on non-epistemic competence components that can be
associated with personal characteristics. However, for education and policy it is important to note
that competence emerges at the intersection of changing social, technological, epistemic, and non-
epistemic components.

It is therefore no accident that the non-epistemic components have been characterised as ‘21st
century competences.’ Labelling these as skills, however, is bound to generate confusion as skills
are commonly understood as domain-specific or practice-oriented competence components, and
often associated with ‘know-how,’ and procedural and tacit knowing (Polanyi, 1967; Ryle, 1949).
Similarly, calling these person-related elements ‘competences’ contributes to a disregard of the

6
cultural and technological contexts, and makes it difficult to understand how the ongoing changes in
these contexts generate the need to address these non-epistemic components of competence.

In a graphical form, the domains of competence can be depicted as in Figure 1. The left-hand side
consists of expertise that can be defined as a combination of knowledge, skill and experience. The
right-hand side includes various dispositions, characteristics and generic capacities that are needed
to realise competent action. The realisation of action requires social and material resources and is
constrained and enabled by technology and culturally shared norms and values. In general terms,
the shift from 20th century competences towards 21st century competences can be viewed as a shift
from the left to right, accompanied by the increasing visibility of the changing cultural and material
contexts.

[INSERT FIGURE 1 ABOUT HERE]

3 NON-EPISTEMIC COMPETENCE COMPONENTS AS TRAITS, ABILITIES,

INTERESTS AND NARRATIVES
A useful model of the dimensions of personality can be represented as in Figure 2, derived from
Roberts’ neo-socioanalytic model of personality (Roberts, 2006). The four segments in the figure
correspond to the domains that make up the core of personality: traits, motives, abilities, and
narratives. These four domains, according to Roberts, subsume most of the important categories of
individual differences. In Figure 2 these different domains are also associated with common
constructs that are used to characterise these domains.

[INSERT FIGURE 2 ABOUT HERE]

The lower left segment of Figure 2 consists of abilities. These are often conceptualised and
measured as ‘intelligences.’ Much scientific debate exists on what these intelligences are and how
they should be modelled. A widely used model is Carroll’s three-stratum model (Carroll, 2005). In
this hierarchical model, the top-level construct of general mental ability, denoted as g, explains
variance across all types of cognitive ability tests, second-level constructs, such as fluid
intelligence, crystallised intelligence, broad visual perception, and processing speed, explain
variance across different task domains, and the bottom level consists of factors that are most
specific.

Personality traits are commonly defined as the enduring patterns of thoughts, feelings, and
behaviour that people exhibit (Roberts, 2006). These are most often measured using psychometric
tests that map personality to the Big Five factors of extraversion, agreeableness, conscientiousness,
emotional stability/neuroticism, and openness to experience (e.g., McCrae & John, 1992).

The third segment in Figure 2 is aimed to cover the domain of motives and the structure of interests.
For the purposes of this article, I use Holland’s theory of career choice as a basis for measuring
motives and interests (cf. Nauta, 2010)—which of course, a simplification.3 Holland’s character
description factors, Realistic, Investigative, Artistic, Social, Enterprising, and Conventional
(RIASEC), however, are widely used in educational and career guidance, counselling, recruiting
and occupational research. At present, RIASEC tests, or their variants, also underpin many public-
sector and commercial web-based career and educational guidance services in countries around the

7
world, including Singapore, Australia, the US, the UK, the Netherlands, Denmark, and in a
multitude of other countries (Tuomi et al., 2021).

In contrast to the neo-socioanalytical model that groups values and motives, Figure 2 implicitly
locates values in the narrative segment. Although the term is used in many ways in existing
literature, the assumption here is that values are a fundamentally social phenomenon, grounded in
making good-bad or right-wrong distinctions in the context of cultural systems of meaning.4
Research on cultural categorisation suggests that systems of values are culture specific (Douglas,
1966; Lakoff, 1987). Narrative accounts of personality, thus, rely on existing socially shared value
systems.5 In this interpretation, values are part of the tacit context in Polanyi’s (1967) original sense,
and the narrative elements of personality consist of accounts that, for example, explain who the
person is (McAdams, 1993; Taylor, 1989) and where the person is located in the culturally
constituted world (Eliade, 1991). These narratives have both an internal aspect, for example, self-
generated accounts of how the person came to be what she is, who she is, and where she is going,
but also external accounts about the person and her social roles. In Roberts’ neo-socioanalytic
model, internal assessment forms the source of identity, whereas external observation and
assessment forms the reputational aspect of personality.

4 STABILITY AND MALLEABILITY OF NON-EPISTEMIC COMPETENCE

COMPONENTS
The concept of personality rests on the assumption of stability across time. In everyday discourse,
we use the concept of personality to describe those generalisable and relatively stable characteristics
that differentiate individuals and make prediction, expectation, and explanation possible. The
sources of this stability have been actively debated over the centuries. In the last decades there have
been major shifts in these debates. The idea that personality can be characterised using a small set
of stable person-related traits was strongly rejected towards the end of the 1960s, when it was
argued that personality reflects situational factors, and personality ‘traits’ should be understood as
artifacts generated by personality tests (Mischel, 1968). In reaction to this situational view, in the
1990’s it was increasingly argued that five situation-independent universal factors (The Big Five)
consistently characterise personality across cultures (e.g., McCrae & Costa Jr., 1997). In the last two
decades, many empirical studies have argued that personality traits exist, are largely genetically
determined, but also change over the lifetime due to maturation, social pressure, and major life
events, such as becoming a parent (Almlund et al., 2011; Heckman & Kautz, 2012; Roberts &
DelVecchio, 2000; Specht et al., 2014).

An influential study by McGue et al. (1993), for example, showed that over 80 % of the observed
stability in personality characteristics in late adolescence and early adulthood can be explained by
genetic factors. Close to two thirds of the individual differences in openness to experience—the
most important Big Five factor for educational achievement along with conscientiousness—can be
explained by hereditary factors (Kankaraš, 2017, p. 15). The maturation of personality and the
resulting changes in the Big Five traits have a strong hereditary component, in particular for
agreeableness, conscientiousness, and neuroticism (Bleidorn et al., 2009). Although linkages and
causalities between hereditary factors, the environment, and personality characteristics are not well
understood, this means that also the possibilities for change in personality traits is constrained and
enabled by hereditary factors. After hereditary and early childhood influences, apparently only

8
limited space is left for individually differentiated personality change when children enter formal
education.

Although life events, environment, and, for example, education may influence the development of
personality, it is generally believed that personality traits have substantial predictive value across
time. For example, in his review on non-cognitive skills, Zhou (2016) suggests that the Big Five
traits have to be separated from non-cognitive skills because they are shown to be relatively stable
and not easily improved by training or education.

Similarly, large bodies of empirical research suggest that interest structures are relatively stable
(Nauta, 2010). As Holland’s influential RIASEC model was developed in the context of
occupational choice, studies on interest stability have mainly focused on adolescents and adults.
Research, however, indicates that there is moderate rank-order stability in the interest structure
already among elementary school children (Tracey & Sodano, 2008). Strong rank-order stability,
i.e. the ordering of people along this dimension, means that the relative future location of a person
on this scale can be well predicted using historical data.

Extant psychometric research suggests that the most stable personality factor is general mental
ability, or g. Although components of g, such as fluid intelligence and crystallised intelligence
change with age—the first peaking in young adulthood, and the second accumulating through the
life-time—the relative order of people on these scales remains fairly stable (Heckman & Kautz,
2012). The reasons for this stability are not clear, but genetic factors and early childhood
experiences apparently play an important role. Research suggests that about half of the variance in g
has genetic origin (Plomin & Spinath, 2002). Large bodies of empirical literature have shown that
cognitive capacity, temperament, and, for example, social capabilities show continuity and
predictability from the first months of age, and sometimes already from prenatal developmental
stages (Bornstein, 2014). Some cognitive capacities, such as language performance at eight years,
can be predicted by brain activity of newborn infants (Molfese, 2000).

There are different forms of consistency of personality that need to be distinguished. Consistency
can mean continuity, for example time invariance or predictable patterns of maturation.
Consistency, however, can also mean relative stability and rank-order consistency. Strong rank-
order stability has been observed in many empirical studies on personality traits (Roberts &
DelVecchio, 2000). Although recent research has started to focus on the more fine-grained facets of
personality that underpin the general Big Five factors and related cognitive abilities (e.g.,
Rammstedt et al., 2018), and alternative models, such as the Big Six (Ashton et al., 2004), may
better describe personality across cultures, there is a general agreement that personality
characteristics are relatively rank-order stable after the early childhood (Bornstein, 2014).
Individuals can, therefore, be sorted using information on traits, abilities, and interests, and future
orderings of individuals along these characteristics can be predicted using this information.

For educators, a critical question is to what extent educational interventions can impact the level or
development of personality characteristics. For example, there exists much work on Social and
Emotional Learning (SEL) and mindset interventions (Abrahams et al., 2019; Chernyshenko et al.,
2018; Durlak et al., 2011; Taylor et al., 2017). In policy-related literature, non-epistemic
competence components are usually defined as malleable skills. The term skill, in itself, suggests
that it can be developed and that the result is useful and valuable, and the additional attribute

9
malleable emphasises that these skills, indeed, can intentionally be changed. At the same time, the
existing evidence on stability and predictability of many capacities that could well be labelled social
and emotional skills and 21st century competences, is considered by definition largely irrelevant.

Summarising the vast literature on personality change in terms relevant for the present analysis,
there is now strong evidence that there exists a core personality that is stable enough to make
predictions across time and environmental conditions. McGue et al. (1993) showed that, on average,
over 80% of the variance in this stable component was associated with genetic factors. They also
showed that, although also hereditary factors play a role, about half of the change in the ‘residual’
elements of personality can be associated with environmental influences. This particular
longitudinal twin-study focused on changes in early adulthood but is consistent with the claim that
core personality is formed by the first years of childhood. Reversing the logic commonly used in
discussions on 21st century skills that focus on the residual and potentially changing ‘noncognitive’
parts of personality, for AI-related prediction, the stable core and the non-epistemic competence
components associated with it are a more natural starting point.

From a practical point of view, the changing social demand for education puts educators in a
challenging position. For example, the top three most requested skills in CEDEFOP’s job market
monitoring system include (1) working in teams, (2) planning and scheduling events, and (3)
accessing and analysing digital data (CEDEFOP, 2022). These can in a straightforward way be
mapped to, for example, the taxonomy of behavioural, emotional, and social skills recently
proposed by Soto et al. (2022). Although research suggests that social contexts and environmental
influences have an impact on measured personality characteristics, the induced changes seem to be
temporary. There is a long history of claims of the effectivity of “brain training” interventions (Katz
et al., 2018), but, despite many popular claims, the evidence remains particularly weak for short-
term interventions such as growth mindset (Moreau, 2022).

Discussion on the malleability of life-determining personal characteristics is bound to create

controversy. Education is traditionally based on the fundamental belief that it can change and
develop individuals.6 For epistemic competence components, this clearly is the case. It also seems
intuitively clear that life experiences and maturation can change personality. It would be natural, for
example, to expect that studying abroad could increase openness to experience. Recent research,
however, suggests that whereas students with high openness often choose to study abroad, this does
not seem to have noticeable impact on their personality (Niehoff et al., 2017; Nissen et al., 2022).

The three segments of Figure 2: abilities, traits, and motives and interests, provide key elements of
the stable core personality. It should be noted that the fourth ‘narrative’ segment has a very different
dynamic. The stories we tell about ourselves and others have a fundamental impact on how we
interpret the world and make sense of it, and how we act. The socialising function of education
(Biesta, 2015) can in many ways be understood as a process that generates shared inter-generational
stories, understanding, and systems of value that make social life possible. The creation of new
stories and reframing existing ones, therefore, may be an important source of personality change.
Culturally shared narratives do not only enable us to make sense of ourselves, but also define the
social relevance and meaning of concepts such as ability, personality traits, and motives and
interests. In this sense, the narrative component of personality may both be seen as the most
malleable component of personality and its most fundamental layer on top of which abilities, traits
and motives are constructed. From a point of view of sociology of knowledge and social learning

10
theories, however, narratives are fundamental elements of the prevailing systems of knowledge. The
narrative component of personality in Figure 2 should therefore be located among the epistemic
competence components.

5 AI AND NON-EPISTEMIC COMPETENCE DATA

For the present analysis, it is not necessary to make strong claims about the possibility of useful
educational interventions in the non-epistemic domain. In fact, education theorists and practitioners
agree that education has important social functions that are not epistemic (e.g., Biesta, 2010).
Instead, the argument is that whatever regularities, stabilities, and continuities there may exist in
human personality, current machine learning systems will be able to find them and use them for
prediction. Whereas empirical studies on personal characteristics have struggled to find statistically
valid associations that predict future life outcomes, AI-based models can skip most concerns about
statistical validity and conceptual and theoretical immaturity. The approach in machine learning is
different. The development and training of machine learning models stops when the system works
well enough, and its predictions can be generalised. At that point, the system has generated an
internal representation of the domain in question and a predictive model that embeds as much
information on relevant distinctions it can. There is no guarantee that the representation would be
based on the Big Five, Big Six, or any other proposed factors or their facets.

Many studies on personality factors, for example, have been based on collecting words used to
describe personality, and clustering them into a manageable set of factors. In contrast, common
representations of input data in AI natural language applications based on machine learning usually
represent words in 300 or 500 -dimensional spaces (e.g., Mikolov et al., 2013). Psychometric
research is strongly limited by its aim to find a relatively small set of factors that can have intuitive
human interpretations. AI systems are not restricted by this requirement.

Data published by the OECD Survey on Social and Emotional Skills (OECD, 2022), for example,
can in a straightforward way be used to generate models of social and emotional skills that with
high probability make more accurate predictions than the best existing factor models. Whereas the
intent of the OECD study is to fit collected data into a predefined descriptive theoretical frame,
machine learning can use these data to construct a new implicit theory of the domain. Although
survey instruments used to collect these data are based on existing theories and hierarchical models
—thus implicitly supporting the chosen theoretical model—machine learning systems can easily
complement such data from other sources. After a model that associates patterns of personality
characteristics with life outcomes is trained, it can predict these outcomes also for individuals.
There remain many technical challenges, for example, in transferring models trained with specific
datasets in new application domains and complementing and augmenting their data, but these are
typical engineering challenges in machine learning.7

Borghans et al. (2008) list five limitations of current evidence on predictive validity that may have
underestimated the impact of personality traits for socioeconomic outcomes including job
performance, health, and academic achievement. First, the optimal level of a trait is not necessarily
the maximal one, and the curvilinear relationships are not well represented by correlation
coefficients. Second, the top-level factors of Big Five are probably too blunt an instrument to
capture relationships between personality and outcomes, and more narrowly defined facets of
personality can predict outcomes better. Third, personality is often measured using self-report
questionnaires that are not very reliable. Fourth, interaction effects, for example, between cognitive
11
abilities and personality traits may be important. Fifth, standard measures of predictive power, such
as effect size and variance explained, may not be the most useful ones, for example, when causal
effects are studied. Machine learning systems can avoid all these sources of potential error and
could therefore have more predictive power than existing research.

Many data-driven AI systems are now used, for example, to predict school drop-out (Del Bonifro et
al., 2020) and to select applicants for jobs and job-related training (Aspan, 2020; HR Research
Institute, 2019; Hunkenschroer & Luetge, 2022). Although there is no guarantee that a prediction
would be correct for a specific individual, by definition the predictions will be statistically robust.
For example, state-of-the-art generative language models in AI, such as the famous GPT-3 from
OpenAI (Brown et al., 2020), can use data on personality characteristics as an input and, with a
relatively minor retraining, write the expected life story of the person. If this can be done using data
from a very early age, the personal and social consequences deserve further study. Similarly, if
preschool children can be sorted based on their ability to work in teams, capacity for problem
solving and planning, and characteristics such as creativity, curiosity, ability to work under pressure
and willingness to constantly learn new things, it should be clear that data on personality has social
consequences.

An extreme claim that data collected in the first years of life can predict life outcomes can only be
made with important caveats. First, despite the broad agreement that social and emotional skills,
personality traits, and cognitive abilities have a clear impact on various life outcomes (Heckman &
Kautz, 2012; Kankaraš, 2017), there is very little robust scientific evidence to support or reject these
claims. Smithers et al. (2018) reviewed 554 publications that have reported the relationship between
childhood noncognitive skills, such as attention, self-regulation, and perseverance, and later
outcomes. About 40% of these empirical studies were judged to present better-quality scientific
evidence, 21% were classified as studies that provided weak evidence, and 38% classified as
providing poor evidence as there was effectively no attempt to control confounding (Smithers et al.,
2018, p. 869). The observed effects have been heterogeneous, and the 95% prediction interval
includes negative, null, and positive effects. Overall, this meta-analysis of existing research, that
also considers the methodological quality of the reported results, suggests that there is some
evidence supporting a role for non-cognitive skills in better academic achievement, but the reported
effects are highly heterogeneous, and a true null effect of noncognitive skills on outcomes cannot be
ruled out.

This, of course, does not mean that non-epistemic competence components would have a negligible
impact on life outcomes. It simply means that researchers have not been able to produce good
scientific evidence for that. This also means that machine learning may be able to drill through the
prevailing conceptual confusion and data limitations and find the true personal characteristics that
predict future outcomes.

Second, it is always possible to question the indicators used to measure outcomes. Given, for
example, that many successful entrepreneurs are school dropouts, educational achievement is not
necessarily the best indicator for a positive outcome. In employment contexts, Bartram (2005), for
example, suggested a criterion-based Big Eight model of work-related competence that was based
on analysing factors that underpin on-the-job performance as judged by the superiors of the
employee. In this approach, the question to ask is What factors can best predict job performance?,
instead of the conventional What performance personality trait X predicts? A theoretically more

12
robust approach might be to link outcome measures to capabilities that are important for
development and well-being; for example, by using the capability-based approach as a starting point
(Sen, 1993).

Third, reported experiments with machine learning have so far been unable to make accurate
predictions about life outcomes. A large experiment in scientific mass collaboration, the Fragile
Families Challenge (Salganik et al., 2020), used data from birth to age nine from 4,242 US families
to predict outcomes at age fifteen. From the 160 competing teams of machine learning experts, no
one was able to make very accurate predictions about the selected six outcomes, including a child’s
grade point average and grit. Although the Fragile Families data used in this experiment is highly
selective and do not include detailed data on non-epistemic competence components or
sophisticated measures for the outcome variables, the result supports the view that life events, in
general, are difficult to predict. As life increasingly has a digital representation, it can be expected,
however, that prediction accuracy is higher when online trace data can be used for prediction.

Based on existing research, it is possible to make contradicting claims; for example, that non-
epistemic competence components are malleable skills amenable to interventions, to a large extent
stable but malleable in some critical early stages of development, or largely fixed at birth. It is easy
to select one alternative over the others based on personal beliefs and anecdotal evidence. In this
article, I have emphasised predictability and generalisability that underpin the concept of
personality as a relatively stable phenomenon. Whether this stability at the end can be reduced to
the Big Five personality domains or its lower-level facets remains an empirical question. It is highly
probable that lower-level facets can more effectively be used for prediction, and it is also probable
that contextual factors not accounted for in conventional personality instruments, such as motives,
incentives, and social roles (Almlund et al., 2011; Borghans et al., 2008; Roberts, 2006), have an
important role. For the developers of data-driven AI systems, these are practical engineering
challenges. When the system makes accurate enough predictions, it works. Somewhat
paradoxically, however, there is no proof that past success will accurately predict the future in
individual cases as the predictions will always be based on the statistical characteristics of past data.

6 DISCUSSION
For education policy, it is important to know what factors influence social, economic, and
educational outcomes. There is now extensive evidence that social and emotional skills and
personality characteristics predict educational and broader life outcomes. Social and emotional
skills are closely related to personality, and interventions aimed at the development of social and
emotional skills are therefore associated with personality change. Although prediction does not
mean that life outcomes would be determined by personal characteristics, the existing associations
are statistically significant.

A core functionality in machine learning systems is prediction. This is not only prediction as an
outcome; prediction also underpins the adaptation and learning of these systems. Learning in these
systems is purely associative and training these systems is based on reducing errors in prediction.8
Instead of artificial intelligence, these systems could therefore better be called “artificial instincts”
(Tuomi, 2018a).

Despite such fundamental differences between machine learning and human learning, these systems
are in many domains able to outperform human cognitive capabilities (Zhang et al., 2022).

13
Although not based on existing research, it can be expected that data-driven AI systems will be able
to model personal characteristics in novel ways that lead to better predictions than traditional
methods. Most importantly, the data do not have to be collected from personality surveys. Although,
for example, the infamous Cambridge Analytica system (Isaak & Hanna, 2018) used a personality
test based on the Big Five factors as an entry point to user data, the implicit connections among
social network users became the main source of social impact.

From an ethical point of view, wrong predictions about individuals are an important problem,
already widely debated also in educational contexts (Baker & Hawn, 2021; Holmes & Porayska-
Pomsta, 2022). From a social point of view, social groupings and categorisations are important, and
they also underpin concepts of fairness, equality, equity, discrimination, and equal treatment
(Tuomi, 2023). The ethical challenge of using personality characteristics in machine learning,
therefore, is not as much about unfair algorithmic decision-making at the individual level as it is
about society-wide amplification of social sorting in many domains of life, including education.
Information and communications technologies generated what now is called the knowledge and
information society, and AI will have similar tendencies to reorganise the foundations of society,
including its systems of knowledge creation, learning, and education. Such a social impact is
orthogonal or extraneous to the conventional ethical concerns based on assessing individual risks.
Therefore, it remains outside the scope of, for example, the proposed EU AI Act (Smuha, 2021) and
checklists aimed to support the developers of ethical AI applications (e.g., AI HLEG, 2020).
Highlighting the potential adverse impact of AI-enabled profiling, Yeung (2019) notes that:

These observations alert us to the collective and cumulative impacts of

contemporary applications of data-driven technologies which, when
undertaken systematically and at scale may, over time, seriously erode and
destabilise the social and moral foundations that are necessary for flourishing
democratic societies in which individual rights and freedoms can be
meaningfully exercised. (Yeung, 2019, pp. 37–38)

Education is a building block of societies. The turn toward non-epistemic learning and competence
development is motivated by profound changes in the ways societies organise their activities. AI
will play an important role in shaping this change, because prediction and expectations are
fundamental factors of social life. The quest for the development of 21st century competences may
look attractive in the post-industrial knowledge society, but we may want to think where this road
leads in a world where machine learning systems are widely used.

The discussion presented in this article on the potential uses of AI in education, highlights the point
that a proper analysis of risks and social consequences may require studying long-term structural
impacts. Structural change is often disruptive and its consequences inherently unpredictable—and
difficult to study with methods traditionally used in empirical psychology, economy, and sociology.
Somewhat paradoxically, as data-driven AI relies on historical data, it will have considerable
difficulties in predicting the impact of the social change that it influences. Future-oriented thinking
and imagination, therefore, is an important element in detecting and assessing challenges and
opportunities, but also in creating anticipatory capabilities that allow the society and its members to
appreciate the emerging challenges and opportunities (Miller, 2018; Poli, 2017).

14
The first metric intelligence tests were created in 1905 by Alfred Binet and Théodore Simon when a
minister of public instruction in France tasked a special commission to identify pupils in need of
specialised education programs. The need to sort children based on their mental ability emerged as
France had in the previous decades introduced compulsory education for all children and extended
the role of central government in education.

Since then, research on intelligence and personality traits is known for the many heated
controversies it has generated. Whereas Binet was looking for ways to improve education (Cicciola
et al. 2014), the history of IQ testing is also associated with eugenics and claims about scientific
racism (e.g., Fraser, 1995). Similarly, any claim that non-epistemic competence components are
difficult to change after children enter formal education, is bound to create controversy. Although
this potential for controversy exists, the main message of this paper is that, to the extent difficult to
change individual differences exist, machine learning systems can profoundly amplify the social
consequences of these differences. Ethics of AI in education, and beyond, therefore, also needs to
revisit the ethics of education (Holmes et al., 2021).

Empirical research on non-epistemic competence components has been struggling with

measurement challenges, methodological issues, and lack of coherent conceptualisations. It is
possible to use existing research selectively to make contradicting claims about research findings.
One reason is that the lack of shared coherent theoretical frameworks makes it difficult to
synthesise existing research results. This is the case for social and emotional skills, noncognitive
skills, soft skills, and 21st century competences. Therefore, there is a need for further conceptual
work in this area. In contrast, machine learning systems can avoid these conceptual and empirical
challenges, as they do not need domain-specific theory. Machine learning systems simply extract
regularities from the data used for their training, and optimise predictions based on criteria given by
the system designers. This is both their strength and their weakness.

Binet’s research was driven by the belief that comparatively lower-ability students can benefit from
their own classrooms and specialised education. A similar belief underpins mastery learning
approaches that underpin many influential AIED applications.9 These applications build on the
assumption that—given enough time and practice—students with different abilities can achieve
similar levels of mastery. The expectation that AI can transform education by personalising learning
(see for example European Parliament, 2021) is often at least implicitly based on the idea that AI
can tailor instruction at the individual level. These systems have typically been based on models of
human cognition and domain-specific knowledge, although there has been increasing interest in
supporting the development of non-cognitive 21st century skills, emotional and social skills, and, for
example, metacognition and self-regulated learning. This shift towards non-epistemic learning has
been driven both by the new technical possibilities of data-driven AI and the recognition of the
importance of non-epistemic components of competence.

On the background of this analysis, I recommend a moratorium on the use of data on non-epistemic
competence components in data-driven AI systems. A moratorium is a temporary halt and
suspension of activity, intended to give time for adequate consideration. Temporarily postponing the
commercial and public sector uses of AI systems that rely on personal characteristics is proposed.
This policy decision is important for facilitating research and debate on the society-level impact of
AI use in education. It is possible that many existing uses of AI implicitly or explicitly rely on data
on personal characteristics. The proposed moratorium would help to surface such uses and allow the

15
current and future users to consider possible consequences—including adverse effects that can be
avoided.

7 CONCLUSION
The observations presented above, of course, do not mean that we should stop studying non-
epistemic competence components, their development, or AI as a tool to detect and support non-
cognitive learning. We are, indeed, in the 21st century. AI-supported non-epistemic learning is a
highly promising area of possible future uses for AI in education.

A new framing of the problem, however, might be useful. Vygotsky’s empirical work in the 1920’s
on the role of artifacts and tools in cognition (Luria & Vygotsky, 1992) suggested that technology
can augment human thinking. AI has successfully been used to support students with special needs,
and from the point of view of unique personality, each of us is special.

For the use of AI in education (AIED), the proper unit of analysis for research and development of
future AIED could well be a hybrid person-technology complex. If it is difficult to change person-
related 21st century competences, it may well be possible to augment and complement social and
emotional skills with AI-based systems when needed. We live, think, and learn in a technological
context, and currently AI technologies are changing this context. In many ways, technology has
become a part of our personality and, perhaps, should be studied as such.

16
REFERENCES

Abrahams, L., Pancorbo, G., Primi, R., Santos, D., Kyllonen, P., John, O. P., & De Fruyt, F. (2019).
Social-emotional skill assessment in children and adolescents: Advances and challenges in
personality, clinical, and educational contexts. Psychological Assessment, 31(4), 460–473. ht-
tps://doi.org/10.1037/pas0000591
AI HLEG. (2019). Ethics Guidelines for Trustworthy AI. Brussels. European Commission. https://
ec.europa.eu/digital-single-market/en/news/ethics-guidelines-trustworthy-ai
AI HLEG. (2020). The Assessment List for Trustworthy Artificial Intelligence (ALTAI). Brussels.
European Commission. https://doi.org/10.2759/002360
Almlund, M., Duckworth, A. L., Heckman, J., & Kautz, T. (2011). Personality psychology and eco-
nomics. In E. A. Hanushek, S. Machin, & L. Woessmann (Eds.), Handbook of the Economics
of Education (Vol. 4, pp. 1–181). Elsevier. https://doi.org/10.1016/B978-0-444-53444-
6.00001-8
Ashton, M. C., Lee, K., Perugini, M., Szarota, P., de Vries, R. E., Di Blas, L., Boies, K., & De
Raad, B. (2004). A Six-Factor Structure of Personality-Descriptive Adjectives: Solutions
From Psycholexical Studies in Seven Languages. Journal of Personality and Social Psycho-
logy, 86(2), 356–366. https://doi.org/10.1037/0022-3514.86.2.356
Aspan, M. (2020, January 20). A.I. is transforming the job interview—And everything after. For-
tune. https://fortune.com/longform/hr-technology-ai-hiring-recruitment/
Attewell, P. (1990). What is Skill? Work and Occupations, 17(4), 422–448. https://doi.org/
10.1177/0730888490017004003
Australian Government. (2019). Reskilling Australia: A data-driven approach. Canberra. Depart-
ment of Employment, Skills, Small and Family Business.
Azevedo, R., Mudrick, N. V., Taub, M., & Bradbury, A. E. (2019). Self-regulation in computer-as-
sisted learning systems. In The Cambridge Handbook of Cognition and Education (pp. 587–
618). Cambridge University Press. https://doi.org/10.1017/9781108235631.024
Baker, R. S., & Hawn, A. (2021). Algorithmic bias in education. EdArXiv. https://doi.org/
10.35542/osf.io/pbmvz
Barredo Arrieta, A., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., Garcia,
S., Gil-Lopez, S., Molina, D., Benjamins, R., Chatila, R., & Herrera, F. (2020). Explainable
Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward re-
sponsible AI. Information Fusion, 58(June 2020), 82–115. https://doi.org/10.1016/
j.inffus.2019.12.012
Bartram, D. (2005). The Great Eight Competencies: A criterion-centric approach to validation.
Journal of Applied Psychology, 90(6), 1185–1203. https://doi.org/10.1037/0021-
9010.90.6.1185
Berg, J., Osher, D., Same, M. R., Nolan, E., Benson, D., & Jacobs, N. (2017). Identifying, Defining,
and Measuring Social and Emotional Competencies. Washington, D.C. American Institutes
for Research.
Biesta, G. (2010). Good Education in an Age of Measurement: Ethics, Politics, Democracy. New
York. Routledge. https://www.routledge.com/Good-Education-in-an-Age-of-Measurement-
Ethics-Politics-Democracy/Biesta/p/book/9781594517914
Biesta, G. (2015). What is education for? On good education, teacher judgement, and educational
professionalism. European Journal of Education, 50(1), 75–87. https://doi.org/10.1111/
ejed.12109
Biran, O., & Cotton, C. (2017). Explanation and justification in machine learning: A survey. IJCAI
2017 Workshop on Explainable Artificial Intelligence (XAI), 8–13. http://www.cs.columbi-
a.edu/~orb/papers/xai_survey_paper_2017.pdf

17
Bleidorn, W., Kandler, C., Riemann, R., Angleitner, A., & Spinath, F. M. (2009). Patterns and
sources of adult personality development: Growth curve analyses of the NEO PI-R scales in a
longitudinal twin study. Journal of Personality and Social Psychology, 97(1), 142–155.
https://doi.org/10.1037/a0015434
Borghans, L., Duckworth, A. L., Heckman, J. J., & ter Weel, B. (2008). The economics and psycho-
logy of personality traits. Journal of Human Resources, 43(4), 972–1059. https://doi.org/
10.3368/jhr.43.4.972
Bornstein, M. H. (2014). Human infancy…and the rest of the lifespan. Annual Review of Psycho-
logy, 65(January 2014), 121–158. https://doi.org/10.1146/annurev-psych-120710-100359
Brown, J. S., Collins, A., & Duguid, P. (1989). Situated cognition and the culture of learning. Edu-
cational Researcher, 18(1), 32–42.
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal,… Amodei, D. (2020). Lan-
guage Models are Few-Shot Learners. ArXiv:2005.14165 [Cs]. http://arxiv.org/abs/
2005.14165
Callon, M., Law, J., & Rip, A. (1986). Mapping the dynamics of science and technology: Sociology
of science in the real world. Houndmills, Basingstoke. The Macmillan Press Ltd.
Carroll, J. B. (2005). The Three-Stratum Theory of Cognitive Abilities. In Contemporary Intellec-
tual Assessment: Theories, Tests, and Issues (pp. 69–76). The Guilford Press.
Castells, M. (1996). The Information Age: Economy, Society and Culture: Volume I: The Rise of the
Network Society. Cambridge, MA. Blackwell Publishers.
CEDEFOP. (2019). Online job vacancies and skills analysis: A Cedefop pan-European approach.
Luxembourg. Publication Office of the European Union.
CEDEFOP. (2022). Skills-OVATE. Cedefop: European Centre for the Development of Vocational
Training. https://www.cedefop.europa.eu/en/tools/skills-online-vacancies
Chernyshenko, O. S., Kankaraš, M., & Drasgow, F. (2018). Social and emotional skills for student
success and well-being: Conceptual framework for the OECD study on social and emotional
skills. Paris. OECD. https://doi.org/10.1787/db1d8e59-en
Cicciola, E., Foschi, R., & Lombardo, G. P. (2014). Making up intelligence scales: De Sanctis’s and
Binet’s tests, 1905 and after. History of Psychology, 17(3), 223–236. https://doi.org/10.1037/
a0033740
Council of the European Union. (2018). Council recommendation of 22 May 2018 on key compet-
ences for lifelong learning (2018/C 189/01). Brussels. The Council of European Union.
https://eur-lex.europa.eu/legal-content/EN/TXT/PDF/?uri=CELEX:32018H0604(01)
Del Bonifro, F., Gabbrielli, M., Lisanti, G., & Zingaro, S. P. (2020). Student dropout prediction. In
I. I. Bittencourt, M. Cukurova, K. Muldner, R. Luckin, & E. Millán (Eds.), Artificial Intelli-
gence in Education (pp. 129–140). Springer International Publishing. https://doi.org/
10.1007/978-3-030-52237-7_11
Douglas, M. (1966). Purity and Danger: An Analysis of Concepts of Pollution and Taboo. London.
Routledge & Kegan Paul.
du Boulay, B. (2018). Intelligent tutoring systems that adapt to learner motivation. In S. D. Craig
(Ed.), Tutoring and Intelligent Tutoring Systems (pp. 104–128). Nova Science Publishers.
du Boulay, B. (2019). Escape from the Skinner Box: The case for contemporary intelligent learning
environments. British Journal of Educational Technology, 50(6), 2902–2919. https://doi.org/
10.1111/bjet.12860
Duckworth, A. L., & Yeager, D. S. (2015). Measurement matters: Assessing personal qualities
other than cognitive ability for educational purposes. Educational Researcher, 44(4), 237–
251. https://doi.org/10.3102/0013189X15584327
Durkheim, E. (1933). Division of Labor in Society. New York. The Free Press.
Durlak, J. A., Weissberg, R. P., Dymnicki, A. B., Taylor, R. D., & Schellinger, K. B. (2011). The
impact of enhancing students’ social and emotional learning: A meta-analysis of school-based
universal interventions. Child Development, 82(1), 405–432. https://doi.org/10.1111/j.1467-
8624.2010.01564.x

18
EC (2018). Annex to the Proposal for a Council Recommendation on Key Competences for Lifelong
Learning (COM(2018) 24 final). Brussels. European Commission. https://ec.europa.eu/educa-
tion/sites/education/files/annex-recommendation-key-competences-lifelong-learning.pdf
EC (2021). Proposal for a Regulation of the European Parliament and of the Council: Laying down
harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain
Union legislative acts. COM(2021) 206 final. Brussels. European Commission. https://eur-
lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:52021PC0206&from=EN
EC (2022). The ESCO classification. European Commission, DG Employment, Social Affairs and
Inclusion. https://esco.ec.europa.eu/en/classification
Eliade, M. (1991). The Myth of the Eternal Return: Cosmos and History. Princeton, NJ. Princeton
University Press.
Engeström, Y., Miettinen, R., & Punamäki, R.-L. (1999). Perspectives in Activity Theory. Cam-
bridge. Cambridge University Press.
European Parliament (2021). Artificial intelligence in education, culture and the audiovisual sector,
19 May 2021. https://www.europarl.europa.eu/doceo/document/TA-9-2021-0238_EN.html
Fraser, S. (1995). The Bell Curve Wars: Race, Intelligence, and the Future of America. New York.
Basic Books.
Freeman, C., & Louçã, F. (2001). As Time Goes By: From the Industrial Revolutions to the Inform-
ation Revolution. Oxford. Oxford University Press.
Giddens, A. (1984). The Constitution of Society: Outline of the Theory of Structure. Berkeley, CA.
University of California Press.
Harley, J. M., Lajoie, S. P., Frasson, C., & Hall, N. C. (2017). Developing emotion-aware, ad-
vanced learning technologies: A taxonomy of approaches and features. International Journal
of Artificial Intelligence in Education, 27(2), 268–297. https://doi.org/10.1007/s40593-016-
0126-8
Harré, R., Clarke, D., & De Carlo, N. (1985). Motives and Mechanisms. London. Methuen.
Heckman, J. J., & Kautz, T. (2012). Hard evidence on soft skills. Labour Economics, 19(4), 451–
464. https://doi.org/10.1016/j.labeco.2012.05.014
Hoekstra, H. A., & Van Slujis, E. (2003). Managing Competences: Implementing Human Resource
Management. Assen. Koninklijke Van Gorcum.
Hogan, R. (1996). A socioanalytic perspective on the five-factor model. In J. S. Wiggins (Ed.), The
Five-Factor Model of Personality: Theoretical Perspectives (pp. 163–179). Guilford Press.
Holmes, W., & Porayska-Pomsta, K. (Eds.) (2022). Ethics of Artificial Intelligence in Education.
Taylor & Francis.
Holmes, W., Porayska-Pomsta, K., Holstein, K., Sutherland, E., Baker, T., Buckingham Shum, S.,
Santos, O. C., Rodrigo, M. T., Cukurova, M., Bittencourt, I. I., & Koedinger, K. R. (2021).
Ethics of AI in education: Towards a community-wide framework. International Journal of
Artificial Intelligence in Education. https://doi.org/10.1007/s40593-021-00239-1
HR Research Institute (2019). The 2019 State of Artificial Intelligence in Talent Acquisition.
https://www.oracle.com/a/ocom/docs/artificial-intelligence-in-talent-acquisition.pdf
Hunkenschroer, A. L., & Luetge, C. (2022). Ethics of AI-enabled recruiting and selection: A review
and research agenda. Journal of Business Ethics, 178(4), 977–1007. https://doi.org/10.1007/
s10551-022-05049-6
Hutchins, E. (1995). Cognition in the Wild. Cambridge, MA. MIT Press.
ILO (2020). The feasibility of using big data in anticipating and matching skills needs. Geneva. In-
ternational Labour Office.
Isaak, J., & Hanna, M. J. (2018). User data privacy: Facebook, Cambridge Analytica, and privacy
protection. Computer, 51(8), 56–59. https://doi.org/10.1109/MC.2018.3191268
John, O. P., & Srivastava, S. (1999). The Big-Five Trait Taxonomy: History, Measurement, and
Theoretical Perspectives. In O. P. John & L. Pervin (Eds.), Handbook of Personality: Theory
and Research (Second edition, pp. 102–138). The Guilford Press.

19
Joksimovic, S., Siemens, G., Wang, Y. E., Pedro, M. O. Z. S., & Way, J. (2020). Editorial: Beyond
cognitive ability. Journal of Learning Analytics, 7(1), 1–4. https://doi.org/10.18608/
jla.2020.71.1
Kankaraš, M. (2017). Personality matters: Relevance and assessment of personality characteristics.
Paris. OECD. https://doi.org/10.1787/8a294376-en
Kankaraš, M., & Suarez-Alvarez, J. (2019). Assessment framework of the OECD Study on Social
and Emotional Skills. Paris. OECD. https://doi.org/10.1787/5007adef-en
Katz, B., Shah, P., & Meyer, D. E. (2018). How to play 20 questions with nature and lose: Reflec-
tions on 100 years of brain-training research. Proceedings of the National Academy of Sci-
ences, 115(40), 9897–9904. https://doi.org/10.1073/pnas.1617102114
Khorrami, M., Khorrami, M., & Farhangi, F. (2022). Evaluation of tree-based ensemble algorithms
for predicting the big five personality traits based on social media photos: Evidence from an
Iranian sample. Personality and Individual Differences, 188, 111479. https://doi.org/10.1016/
j.paid.2021.111479
Kirkpatrick, K. (2016). Battling algorithmic bias: How do we ensure algorithms treat us fairly?
Communications of the ACM, 59(10), 16–17. https://doi.org/10.1145/2983270
Kizilcec, R. F., & Lee, H. (2022). Algorithmic fairness in education. In W. Holmes & K. Porayska-
Pomsta (Eds.), Ethics in Artificial Intelligence in Education. Taylor & Francis.
Kotamraju, N. P. (1999). The birth of web site design skills: Making the present history. American
Behavioral Scientist, 43(3), 464–474.
Lakoff, G. (1987). Women, Fire, and Dangerous Things: What Categories Reveal about the Mind.
Chicago. University of Chicago Press.
Latour, B. (1996). On actor-network theory: A few clarifications. Soziale Welt, 47(4), 369–381.
Leont’ev, A. N. (1978). Activity, Consciousness, and Personality. Englewood Cliffs, NJ. Prentice-
Hall.
Luria, A. R., & Vygotsky, L. (1992). Ape, Primitive Man, and Child: Essays in the History of Beha-
vior. Hemel Hempstead. Harvester Wheatsheaf.
Matcha, W., Gašević, D., Jovanović, J., Uzir, N. A., Oliver, C. W., Murray, A., & Gasevic, D.
(2020). Analytics of learning strategies: The association with the personality traits. Proceed-
ings of the Tenth International Conference on Learning Analytics & Knowledge, 151–160. ht-
tps://doi.org/10.1145/3375462.3375534
McAdams, D. P. (1993). The Stories We Live by: Personal Myths and the Making of the Self. Guil-
ford Publications.
McCrae, R. R., & Costa Jr., P. T. (1997). Personality trait structure as a human universal. American
Psychologist, 52(5), 509–516. https://doi.org/10.1037/0003-066X.52.5.509
McCrae, R. R., & Costa, P. T., Jr. (1999). A Five-Factor Theory of Personality. In L. A. Pervin &
O. P. John (Eds.), Handbook of Personality: Theory and Research (pp. 139–153). Guilford
Press.
McCrae, R. R., & John, O. P. (1992). An introduction to the Five-Factor Model and its applications.
Journal of Personality, 60(2), 175–215. https://doi.org/10.1111/j.1467-6494.1992.tb00970.x
McGue, M., Bacon, S., & Lykken, D. T. (1993). Personality stability and change in early adulthood:
A behavioral genetic analysis. Developmental Psychology, 29(1), 96–109. https://doi.org/
10.1037/0012-1649.29.1.96
Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations
in vector space. ICLR Workshop Proceedings. https://doi.org/10.48550/arXiv.1301.3781
Miller, R. (Ed.). (2018). Transforming the Future: Anticipation in the 21st Century. Routledge.
Mischel, W. (1968). Personality and Assessment. New York. Wiley.
Mitchell, S., Potash, E., Barocas, S., D’Amour, A., & Lum, K. (2021). Algorithmic fairness:
Choices, assumptions, and definitions. Annual Review of Statistics and Its Applications,
2021(8), 141–163. https://doi.org/10.1146/annurev-statistics-042720-125902

20
Mittelstadt, B., Russell, C., & Wachter, S. (2019). Explaining Explanations in AI. Proceedings of
the Conference on Fairness, Accountability, and Transparency, 279–288. https://doi.org/
10.1145/3287560.3287574
Molfese, D. L. (2000). Predicting dyslexia at 8 years of age using neonatal brain responses. Brain
and Language, 72(3), 238–245. https://doi.org/10.1006/brln.2000.2287
Moreau, D. (2022). How malleable are cognitive abilities? A critical perspective on popular brief
interventions. American Psychologist, 77(3), 409–423. https://doi.org/10.1037/amp0000872
Nauta, M. M. (2010). The development, evolution, and status of Holland’s theory of vocational per-
sonalities: Reflections and future directions for counseling psychology. Journal of Counseling
Psychology, 57(1), 11–22. https://doi.org/10.1037/a0018213
Niehoff, E., Petersdotter, L., & Freund, P. A. (2017). International sojourn experience and personal-
ity development: Selection and socialization effects of studying abroad and the Big Five. Per-
sonality and Individual Differences, 112(July 2017), 55–61. https://doi.org/10.1016/
j.paid.2017.02.043
Nissen, A. T., Bleidorn, W., Ericson, S., & Hopwood, C. J. (2022). Selection and socialization ef-
fects of studying abroad. Journal of Personality, published online 13 March 2022 in
early view. https://doi.org/10.1111/jopy.12712
OECD (2014). Competency Framework. OECD. https://www.oecd.org/careers/competency_frame-
work_en.pdf
OECD (2015). OECD/INFE Core competencies framework on financial literacy for youth. Paris.
OECD. https://www.oecd.org/daf/fin/financial-education/Core-Competencies-Framework-
Youth.pdf
OECD (2022). Data from the Survey on Social and Emotional Skills. Organisation for Economic
Co-operation and Development. https://www.oecd.org/education/ceri/social-emotional-skills-
study/data.htm
O*NET (2022). O*NET OnLine. National Center for O*NET Development. https://www.oneton-
line.org/
Plomin, R., & Spinath, F. M. (2002). Genetics and general cognitive ability (g). Trends in Cognitive
Sciences, 6(4), 169–176. https://doi.org/10.1016/S1364-6613(00)01853-2
Polanyi, M. (1967). The Tacit Dimension. New York. Anchor.
Poli, R. (2017). Introduction to Anticipation Studies. Cham. Springer International Publishing.
Porayska-Pomsta, K., Alcorn, A. M., Avramides, K., Beale, S., Bernardini, S., Foster, M. E.,
Frauenberger, C., Good, J., Guldberg, K., Keay-Bright, W., Kossyvaki, L., Lemon, O., Ma-
demtzi, M., Menzies, R., Pain, H., Rajendran, G., Waller, A., Wass, S., & Smith, T. J. (2018).
Blending human and artificial intelligence to support autistic children’s social communication
skills. ACM Transactions on Computer-Human Interaction, 25(6), 35:1-35. https://doi.org/
10.1145/3271484
Rammstedt, B., Lechner, C. M., & Danner, D. (2018). Relationships between personality and cog-
nitive ability: A facet-level analysis. Journal of Intelligence, 6(2):28. https://doi.org/10.3390/
jintelligence6020028
Roberts, B. W. (2006). Personality development and organizational behavior. Research in Organiz-
ational Behavior, 27(2006), 1–40. https://doi.org/10.1016/S0191-3085(06)27001-1
Roberts, B. W., & DelVecchio, W. F. (2000). The rank-order consistency of personality traits from
childhood to old age: A quantitative review of longitudinal studies. Psychological Bulletin,
126(1), 3–25. https://doi.org/10.1037/0033-2909.126.1.3
Roberts, B. W., Walton, K. E., & Viechtbauer, W. (2006). Patterns of mean-level change in person-
ality traits across the life course: A meta-analysis of longitudinal studies. Psychological Bul-
letin, 132(1), 1–25. https://doi.org/10.1037/0033-2909.132.1.1
Ryle, G. (1949). The Concept of Mind. London. Hutchinson.
Salganik, M. J., Lundberg, I., Kindel, A. T., Ahearn, C. E., Al-Ghoneim, K., Almaatouq,… McLa-
nahan, S. (2020). Measuring the predictability of life outcomes with a scientific mass collab-

21
 oration. Proceedings of the National Academy of Sciences, 117(15), 8398–8403. https://
doi.org/10.1073/pnas.1915006117
Sen, A. (1993). Capability and well-being. In M. C. Nussbaum & A. Sen (Eds.), The Quality of Life
(pp. 30–53). Clarendon Press.
Sen, A. (2009). The Idea of Justice. Cambridge, MA. The Belknap Press of Harvard University
Press.
Smithers, L. G., Sawyer, A. C. P., Chittleborough, C. R., Davies, N. M., Davey Smith, G., &
Lynch, J. W. (2018). A systematic review and meta-analysis of effects of early life non-cog-
nitive skills on academic, psychosocial, cognitive and health outcomes. Nature Human Beha-
viour, 2(11), 867–880. https://doi.org/10.1038/s41562-018-0461-x
Smuha, N. A. (2021). Beyond the individual: Governing AI’s societal harm. Internet Policy Review,
10(3). https://doi.org/10.14763/2021.3.1574
Snow, R. E., Corno, L., & Jackson, D. N. (1996). Individual differences in affective and conative
functions. In D. C. Berliner & R. C. Calfee (Eds.), Handbook of Educational Psychology (pp.
243–310). Simon & Schuster Macmillan.
Soto, C. J., Napolitano, C. M., Sewell, M. N., Yoon, H. J., & Roberts, B. W. (2022). An integrative
framework for conceptualizing and assessing social, emotional, and behavioral skills: The
BESSI. Journal of Personality and Social Psychology, 123(1), 192–222. https://doi.org/
10.1037/pspp0000401
Specht, J., Bleidorn, W., Denissen, J. J. A., Hennecke, M., Hutteman, R., Kandler, C.,…Zimmer-
mann, J. (2014). What drives adult personality development? A comparison of theoretical per-
spectives and empirical evidence. European Journal of Personality, 28(3), 216–230. https://
doi.org/10.1002/per.1966
Squicciarini, M., & Nachtigall, H. (2021). Demand for AI skills in jobs: Evidence from online job
postings. OECD Science, Technology and Industry Working Papers No. 2021/03. Paris.
OECD. https://dx.doi.org/10.1787/3ed32d94-en
Stetsenko, A. P. (1995). The role of the principle of object-relatedness in the theory of activity.
Journal of Russian and East European Psychology, 33(6), 54–69.
Taylor, C. (1989). Sources of the Self: The Making of the Modern Identity. Cambridge. Cambridge
University Press.
Taylor, R. D., Oberle, E., Durlak, J. A., & Weissberg, R. P. (2017). Promoting positive youth devel-
opment through school-based social and emotional learning interventions: A meta-analysis of
follow-up effects. Child Development, 88(4), 1156–1171. https://doi.org/10.1111/cdev.12864
Tracey, T. J. G., & Sodano, S. M. (2008). Issues of stability and change in interest development.
The Career Development Quarterly, 57(1), 51–62. https://doi.org/10.1002/j.2161-
0045.2008.tb00165.x
Tuomi, I. (2015). Epistemic literacy or a clash of clans? A capability-based view on the future of
learning and education. European Journal of Education, 50(1), 21–24. https://doi.org/
10.1111/ejed.12101
Tuomi, I. (2018a). The impact of artificial intelligence on learning, teaching, and education:
Policies for the future. Luxembourg. Publications Office of the European Union. https://
doi.org/10.2760/12297
Tuomi, I. (2018b). Vygotsky meets backpropagation: Artificial neural models and the development
of higher forms of thought. In C. P. Rosé, R. Martínez-Maldonado, U. Hoppe, R. Luckin, M.
Mavrikis, K. Porayska-Pomsta, B. McLaren, & B. du Boulay (Eds.), Artificial Intelligence in
Education. Proceedings of the 19th International Conference, AIED 2018, London, UK, June
27-30, 2018, Part I, pp. 570-583. Springer. https://doi.org/10.1007/978-3-319-93843-1_42
Tuomi, I. (2020). The use of artificial intelligence (AI) in education. Brussels. European Parliament,
Policy Department for Structural and Cohesion Policies. https://bit.ly/3lCMotK
Tuomi, I. (forthcoming). A framework for socio-developmental ethics in educational AI. Proceed-
ings of the 56th Hawaii International Conference on System Science. Maui, January 3-6, 2023.

22
Tuomi, I., Hautamäki, A., & Ketamo, H. (2021). Työmarkkinatieto digitalisoituvassa Suomessa
[Labor market knowledge in the era of digitalization in Finland]. Helsinki. Prime Minister’s
Office, Finland. http://urn.fi/URN:ISBN:978-952-383-271-8
UNESCO (2016). UNESCO Competency Framework. UNESCO. https://en.unesco.org/sites/
default/files/competency_framework_e.pdf
UNESCO (2020). Competence Framework for Cultural Heritage Management. https://bangkok.un-
esco.org/content/competence-framework-cultural-heritage-management
Vygotsky, L., & Luria, A. (1994). Tool and symbol in child development. In R. van der Veer & J.
Valsiner (Eds.), The Vygotsky reader (pp. 99–174). Blackwell Publishers.
Yeung, K. (2019). A study of the implications of advanced digital technologies (including AI sys-
tems) for the concept of responsibility within a human rights framework. Council of Europe
study DGI(2019)05. Strasbourg. Council of Europe.
Zhang, D., Maslej, N., Brynjolfsson, E., Etchemendy, J., Lyons, T., Manyika, J., Ngo, H., Niebles,
J. C., Sellitto, M., Sakhaee, E., Shoham, Y., Clark, J., & Perrault, R. (2022). The AI Index
2022 Annual Report. Stanford, CA. AI Index Steering Committee, Stanford Institute for Hu-
man-Centered AI, Stanford University.
Zhou, K. (2016). Non-cognitive skills: Definitions, measurement and malleability. Background pa-
per prepared for the 2016 Global Education monitoring Report. Paris. UNESCO.

ENDNOTES

23
 1 Classification and categorisation are central topics, for example, in sociology of knowledge, sociocultural
psychology, research on cultural cognition, as well as in mainstream social theory. Here I simply note that the use of
data on personal characteristics in machine learning systems can structure societies in novel ways.
2 Latour (1996), however, emphasises that actor-network theory has no model of human competence.
3 The conceptualisation of motives varies widely across different disciplines, also within the different traditions of
psychology. For example, in activity theory motive is the culturally constituted object of activity (Harré et al., 1985;
Stetsenko, 1995).
4 It can be argued that values become increasingly culturally determined as a child develops “higher forms” of
thought (Tuomi, 2018b). In economics, values are often understood as preferences (Sen, 2009), and in sociological
literature, typically associated with norms and social legitimation structures (e.g., Giddens, 1984).
5 All personality characteristics, of course, have this narrative element. Hogan (1996, p. 173), for example, argued
that personality traits are “categories that people use to evaluate one another.”
6 The history of education is, of course, more complicated as there have been many pedagogic traditions. For
example, in critical pedagogy and activity theory social change plays an important role.
7 Data-driven AI models require that training data and data used for predictions have similar structure. It is not
possible to directly use the OECD data to predict outcomes if data on outcomes is not available or if input data are
structurally different from data used for model training. In practice, the OECD data would need to be agumented.
8 More accurately, data-driven AI systems define a “loss function” and learning occurs by reducing this loss until the
system optimally generalises predictions to examples that it has not seen before.
9 I have collaborated on another article that further elaborates on this point, in this issue of the European Journal of Ed-
ucation.

View publication stats