Genetics for ‘equality’? The politics of knowledge production in educational genomics

Abstract

Educational genomics is an emerging field of research that analyses associations between vast samples of human DNA and educational outcomes. I trace how this field navigates a series of old and new methodological problems and political controversies, while attempting to distance itself from the elitist, eugenic, and racist history of genetics in education. Moving away from genetic determinism, its multidisciplinary approach embeds knowledge from the social sciences selectively. In particular, I highlight how microeconomic methodologies and concepts have become salient not only in educational genomic explanations and hypotheses but also in scientists’ political understanding of ‘equality’, reframing past political debates and reimagining governance applications of genetic knowledge. While controversies on the biologisation of social hierarchies persist, the associated debates on human difference, now framed around producing educational ‘equality’, are premised on educational genomics’ contribution to predicting, valuing, and enhancing social human capital.

Introduction

Since the Human Genome Project's completion in 2003, genomic science has enabled the translation of life into data, fuelling a new bioinformatic economy (Sunder Rajan, 2006), and has itself undergone conceptual, methodological, and technological transformations. With advances in high-performance computing and machine learning, studies have moved from seeking single causal genes to identifying multiple genetic associations in vast bio-datasets that can be analysed at high speed (Cambrosio et al., 2014). Social researchers today describe a ‘postgenomic’ era, where epigenetic gene expression makes it untenable to distinguish the genome from its (molecularised) environment (Griffiths and Stotz, 2013: 68; Meloni, 2016: 193). Simultaneously, promises proliferate about the groundbreaking possibilities of genomic prediction beyond health, medicine, and food production, to social welfare and education (Conley and Fletcher, 2017; Plomin, 2018). Among these, the genomics of education, or ‘educational genomics’ (Kovas et al., 2016), has been proposed as a new, and necessary, contribution to educational research, promising to offer predictive tools akin to precision medicine (Asbury and Plomin, 2014), to outline educational reforms, and to address ‘social inequality’ (Harden, 2021a). These proposals have received policy and media attention (Asbury, McBride, and Rimfeld, 2021; Harden et al., 2022; Serious Science, 2017; Wintour, 2013), along with criticism by other geneticists (Bird, 2021; Henn et al., 2021). By the time of writing, educational genomics has become a highly active, though contested, multidisciplinary field, bringing together behaviour geneticists, economists, data scientists, educational psychologists, and social scientists.

As sociogenomics, which investigates the genomics of social or ‘behavioural’ difference and stratification more broadly, has received scrutiny from social scientists (Bliss, 2018; Burt 2023) and bioethicists (Meyer et al., 2023a; Parens, 2021), so has the contemporary sociogenomic gaze on education, particularly as it is the most studied social variable of genomic studies. Bioethicist Daphne Martschenko and colleagues have initiated what she calls ‘adversarial collaboration’ with researchers in educational genomics, aiming to develop reflective and ethical scientific practice (Martschenko, Trejo, and Domingue, 2019). This work has underlined a series of risks in light of the turbulent history of genetics and eugenics of intelligence: attributing educational performance differences to genetic ‘ancestry’, commercialising prenatal and neonatal genetic screening for ‘educational attainment’ (Meyer et al., 2023b), genetic discrimination by insurance providers, and uncertainties around genetic prediction for learning difficulties. Meanwhile, researchers’ arguments for the use of educational genomics to improve education while avoiding these risks have been criticised for appealing to neoliberal governance approaches (Panofsky, 2015, 2021) and for diverting attention away from alternate, social-structural explanations of educational inequities (Roberts and Rollins, 2020). Researchers in the genomics of education have more recently collaborated with some of their critics to produce a ‘consensus report’ on ethics in social and behavioural genomics led by the bioethics institute Hastings Centre (Meyer et al., 2023a). While participants began from similar principles of social justice, the report reveals a lack of complete consensus on critical ethical issues in the field, as, for example, the question of whether comparisons between ancestry groups are scientifically valid and ethically permissible.

This article engages with these debates, not so much to offer an alternative bioethics approach, but to explore how researchers frame the value of educational genomics, how that framing is related to currently dominant knowledge claims and epistemic assumptions in this field, and how these are intelligible in the contemporary political-economic context. I particularly draw inspiration from the question of how the life sciences may embed in their forms of knowledge production epistemic assumptions, epistemologies, and ethico-political positions dominant in the political-economic regimes in which they emerge, as part of a broader understanding of how bioscience is shaped in contemporary capitalism (Birch and Tyfield, 2013; Sunder Rajan, 2006). As a result, this article does not comment on the public understanding of genomic science, or on uses of educational genomics that contravene the recommendations of researchers, as, for example, the genetic testing of embryos (Au, 2022).

The use of ‘big data’ bioinformatic infrastructures and computational techniques in educational genomic research can produce a ‘qualitatively different kind of knowledge’ (Stevens, 2017: 172), but this knowledge is not merely technologically determined. Intertwined political, economic, and technological contexts have historically played a role in the development of biological knowledge (Meloni, 2016). Knowledge production is itself also a site of political contestation, which, as Panofsky (2014: 141) has elaborated in his history of behaviour genetics, can mobilise particular rhetorical strategies as ‘an approach to building the symbolic and material resources for securing scientific credibility and recognition, or scientific capital’. Historically contextualising educational genomic research then demands attention simultaneously to the material conditions and infrastructures of research; to conceptual, methodological, and technological shifts in knowledge production; and to ethical-political and political-economic conceptions of valuable knowledge and its imagined beneficiaries.

Here, I do not analyse the economic structures, infrastructures, and networks that underpin the genomics of education. This analysis is developed elsewhere (Williamson et al., 2024). I pay attention, instead, to how the currently politically dominant episteme of governing populations affects knowledge production on education, as a privileged site of state intervention on bodies, subjects, and the reproduction of labour-power (Foucault, 2008). Neoliberal economic governance is relevant to genomics, in that it understands educational subjects and populations as investable bodies (human capital) and seeks to intervene at the level of ‘cells, molecules, genomes, and genes’ (Pierce, 2013: 12). As Pickersgill (2018, 2020) also notes, enhancing and economising human ‘brains’ is a major contemporary policy concern, with interventions, from early years to ‘brain training’ programmes, invested with expectations of how the ‘imagined biological’ might be rendered more ‘plastic’ and be better optimised for the generation of wealth. In the case of genomics in education, economising the ‘imagined biological’ involves the production of knowledge on how biology, and genes, are linked to economic outcomes, as well as how they can be manipulated to enhance such outcomes. Eugenics—the science of intervening in human biological reproduction to enhance what are thought to be desirable human abilities—is a conspicuous precedent to such manipulation, immediately raising ethical and political opposition. Educational genomic research finds itself in the middle of this conflict, compelled to defend its purpose.

In this article, then, I show how the legacies and dilemmas of the genetics of education (and its implications for discourses on intelligence) are negotiated in this contemporary context. Educational genomics, as a multidisciplinary field, renegotiates epistemic (epistemological, ontological, methodological) assumptions of behaviour genetics in its data-intensive bioinformatic approach. Seeking to move away from genetic determinism, it embeds knowledge from the social sciences, but does so selectively. In particular, genomic studies of education adopt methods, questions, interpretive concepts, and concerns of mainstream economic science. This resonates with the broader move towards predictive and probabilistic reasoning in genomic science, with educational genomics framed as useful for managing molecular-level risk and potential in education—an education understood economically as ‘investment’ in human capital. Simultaneously, past political debates over the pathologisation and biologisation of social hierarchies are reframed through the promise that educational genomics will contribute to educational ‘equality’ by predicting, valuing, and enhancing social human capital.

These observations are informed by critical discourse analysis of research, policy, and media articles that used, reviewed, or critically analysed genomics methods in education, covering the period 2005–23. Starting with publications by researchers identified through previous network mapping, along with two other members of our project's research team, we snowballed through bibliographies to identify influential studies. Searches produced a corpus of over 100 articles and 4 books, as well as recordings of conference presentations, media appearances, and podcasts. After initial note taking and discussion, we established emergent themes, which informed further note taking around the domains of actors and networks, knowledge and discourses, and methods and technologies of educational genetics. This article is additionally informed by seven interviews with key researchers in the genomics of education, which, while not analysed here, helped us understand dominant perspectives in the field.

After the analysis has been situated in the shift from educational genetics to educational genomics, the article's structure illuminates and historically contextualises the analytical ‘flashpoints’ Gulson and Baker (2018: 161) have identified in the ‘new biological rationalities’ seeking to reconfigure Western systems of education. These are (a) strategies and tactics to authorise genomic data as a superior source of truth about educational performance, (b) epistemic premises regarding gene-environment interplay and their reshaping in the context of data-centric and microeconomic approaches to educational genomic research, and (c) disciplinary political conceptions of biological and social inequalities associated with the value of genomics in education and of education itself.

Educational genetics and genomics

Scientific attempts to bring a genetic lens to educational and economic outcomes have long been entangled in political conflict over racialised and class hierarchies, not least through the history of eugenics (Lowe, 1998). Post-war behaviour genetics sought to reposition itself away from eugenics, scientific racism, and World War II war crimes (Panofsky, 2014), and become aligned with a ‘liberal-democratic framework’ endorsing genetic diversity, individuality, and uniqueness (Meloni, 2016). By the 1970s, however, conflict erupted again over the genetic basis of differences in IQ scores between ‘black’ and ‘white’ populations (the ‘race and IQ controversy’) and its implications for education. In response to this conflict, the scientific community of behaviour genetics became ‘bunkerised’ around a biological view on the difficulties of disadvantaged children (Panofsky, 2014, 2015).

Research on the genetics of education in the 1970s, 1980s, and 1990s, using twin and adoption study methods, linked educational outcomes to the heritability of both cognitive abilities (Willerman, Horn, and Loehlin, 1977) and economic status (Taubman, 1976). The statistical concept of heritability—the ratio of genetic variance to total variance in a phenotype—was used to indicate the degree of trait inheritance in a given population. Scientists have long specified that heritability ratios are not measures of genetic determinacy or causality and are specific to the study population, varying in different social or historical environments (e.g. Harden, 2021a: 122; Plomin, 2018: 7; Turkheimer, 2016: 25). The association between genetic variance and phenotypic variance in a population does not necessarily imply causation. Yet the concept was repeatedly used as evidence of genetic determinacy, especially in response to criticism by left-wing biologists like Lewontin, Rose, and Kamin (1984) and others. Bypassing methodological contestation—which remains relevant and which I discuss in more detail later—and through a proliferation of heritability studies, respected figures like Sandra Scarr (1992) and David Rowe (1994) asserted that families and parenting have little influence on children's school outcomes. Intellectual provocation, Panofsky (2014: 163) argues, served to consolidate twin and adoption methods, studies of heritability, and genetic determinism in the field.

After the sequencing of the human genome, the genetics of cognitive abilities engaged in a string of candidate gene studies that failed to consistently replicate. Many behaviour geneticists then moved towards more thoroughly exploring gene-by-environment interactions (GxE) in studies of IQ, including interactions with socio-economic status (Turkheimer et al., 2003), concluding that intelligence and school performance are, to an extent, malleable (Sauce and Matzel, 2018). Nonetheless, heritability measures continue to be misused as proof of genetic causality (see Keller, 2010; Moore and Shenk, 2017 for extensive criticisms). Even prominent figures, like the behaviour geneticist Robert Plomin (2018: 10), state, for example, that ‘performance on tests of school achievement is 60 per cent heritable on average. That is, more than half of the differences between children in how well they do at school is due to inherited DNA differences.’

Meanwhile, a statistical approach emerged in genomics, enabled by expanding biobanks and fast, low-cost sequencing and genomic analysis technologies: genome-wide association studies (GWAS). Using complex statistical methods and vast population samples, GWASs screen for variations in the human genome (single nucleotide polymorphisms, or SNPs) that are statistically associated with a given ‘trait’. Each significant association, which can be in thousands of SNPs, is then weighted by the strength of its relationship to the trait to construct a polygenic index (PGI), which indicates the proportion of variance in the trait predicted by the collective variance in SNPs. While clinical traits have concerned genomics in medicine (see Tabery, 2023, for a historical account), the field of sociogenomics (Robinson, Grozinger, and Whitfield, 2005) and its sub-field of genoeconomics (Benjamin et al., 2012) have been searching for genomic associations with social and economic variables.

The largest-scale genomic studies of education emerged from collaboration between economic scientists and behaviour geneticists of education who set up SSGAC (the Social Science Genetic Association Consortium), now the biggest bio-dataset aggregator and innovator in sociogenomic methods. Seeking to understand ‘human difference’ in socio-economic outcomes, they selected ‘educational attainment’ (EA) as a target phenotype to help identify the genetic correlates of ‘economic behaviour’ and stratification (Beauchamp et al., 2011). Note that phenotype selection has long been debated in behaviour genetics, largely between human and animal researchers (Panofsky, 2014: 59, 86–91), and continues to be in question (Ramus, 2023). The debate revolves around whether biologically proximal phenotypes are explored, in order to more easily generate hypotheses about mechanisms. Even though genoeconomists, based on economic psychology, hypothesise that economic outcomes are affected by genetically determined personality traits, they have been unable to test these hypotheses by conducting GWASs on such traits, due to lack of consistent data across datasets. Selecting EA as the target phenotype was therefore a pragmatic decision ‘dictated by which variables are consistently measured across a large-enough number of data sets that the joint sample size will yield reliable results’ (Benjamin et al., 2012: 655). Evidently, SSGAC's genomic studies in education have been explicitly economic. They stand in contrast to longer-established behaviour geneticists of intelligence like Robert Plomin and his collaborators (e.g., Krapohl et al., 2014; von Stumm and Plomin, 2021), who use PGIs of EA as a proxy for ‘intelligence’ and ‘human abilities’—especially since GWASs of IQ currently only explain roughly 5% of the variance in IQ (Savage et al., 2018). As I discuss later, the interpretation of statistical associations between EA and SNP variance remains an epistemic, ethical, and political question.

SSGAC's approach to the genomics of education has gained significant international authority through large-scale organisational networking, the consolidation of core concepts and explanations, and methodological innovations in the analysis of digital bioinformation (Williamson et al., 2024). Large-scale genomic studies of EA, with samples reaching up to 3 million, now mobilise an international network of scholars; research centres; biobanks across (mostly) North America, Europe, Australia, and New Zealand; and consumer genomics companies — predominantly 23andMe. Statistically normalising international data on ‘years of education’—a process subject to continual refinement (Mills and Tropf, 2020: 572)—is used to represent a common international measure of educational attainment, which enables ever-growing sample sizes as biobanks expand across the world.

However, GWASs have, so far, not replicated the heritability estimates of twin studies for any phenotype. This is the so-called ‘missing heritability’ problem, which some behaviour geneticists attribute to insufficiencies in GWAS methodology (Matthews and Turkheimer, 2022). Others consider its solution only a matter of developments in bioinformatics and expanding data volumes, along with increasing methodological sophistication and computational power to sequence the whole genome and detect rare SNP variants (Harden, 2021a: 124; Zuk et al., 2014). Nevertheless, by 2022, it began to be admitted that with expanding sample sizes ‘there are clearly diminishing returns’ (Okbay et al., 2022: 444). In 2013, with a GWAS sample of 126,559, the PGI predicted 2% of variance in EA (Rietveld et al., 2013). With a sample of 293,723, 3.2% (Okbay et al., 2016); with a sample of 1.1 million, 11–13%; and, most recently, with a sample of 3 million, 12–16% (Okbay et al., 2022).

Discussions of these findings in popular science books and magazine articles feature warnings about inevitable scientific progress: ‘A DNA revolution is coming to our schools, and teachers need to be ready for it’ (Asbury, 2023: n.p.); ‘The ability to use genetic information to predict responsiveness to different treatments and policies is in our immediate future’ (Conley and Fletcher, 2017: 169). Genomic advances have been described as offering an evidence-based, metric-driven approach to inform educational policy (Visscher, 2022), or as capable of addressing social and educational inequalities against the racist eugenics of the past (Harden, 2021a). Critics argue that seemingly measured claims on the genetics of education still naturalise social hierarchies (Bird, 2020). Many highlight the still widespread measurement of intelligence for classifying pupils, which, founded on the notion of ‘natural’ abilities, has continued to reproduce racialised social hierarchies, as it was designed to do by its eugenic inception (Martschenko, 2023; Roberts, 2015). Educational genomics scholars sometimes respond combatively to critics, writing about genetic ‘denial’ (Visscher, 2022: 2) or ‘blank slate theories’ (Asbury and Plomin, 2014; Benjamin et al., 2012; Kovas et al., 2016; Plomin, 2018) — phrases famously popularised by Steven Pinker's defence of evolutionary and genetic interpretation in psychology. Panofsky's (2014) extensive account of controversy in the field from the 1950s to the 21st-century attributes such combatative attitudes predominantly to the micropolitics of building the scientific capital of behaviour genetics. Below I draw attention, additionally, to political and epistemic convergences with currently dominant models of governing education.

Authorising truth claims: The imagined genetic in economised education

Truth claims in educational genomics fundamentally rest on the concept of polygenicity, which serves to unify in a single statistical prediction the incremental predictive power of thousands of SNP ‘hits’ in massive DNA datasets. Polygenicity is the outcome of GWAS's data-mining approach to genomic research, described as ‘unbiased’, as ‘hypothesis-free’ (Mills and Tropf, 2020: 556), and as producing results ‘completely reliable in the sense of test–retest reliability’, ‘free from the measurement errors that typically affect data from psychometric tests and self-reports’ (von Stumm et al., 2020: 2). ‘Complex traits’ like EA are only marginally correlated with each of thousands of SNP variants identified by GWASs. Yet the assumption SNP correlations can be added to constitute genetic prediction of ‘outcomes’ or ‘traits’ in the form of PGIs enables scientifically conceptualising these traits as ‘highly polygenic’ (e.g. Harden and Koellinger, 2020: 569). Even though PGIs are recognised as a ‘constructed, a pragmatic solution introduced when the number of SNPs became too large to be considered as separate variables in a regression analysis’ (Janssens, 2019: R147), many scientists used them discursively to represent genes themselves and genetic causality.

With GWAS methodologies, the language of ‘heritability’ has been displaced by the equally ambiguous language of ‘prediction’ (Belsky et al., 2016; Malanchini et al., 2020; Plomin, 2018; von Stumm and Plomin, 2021). SSGAC, which, as a consortium, engages bioethicists in reviewing the remit and reception of their research, has been most cautious in using such language. They have sought to prevent interpretive slippage by publishing explicatory FAQ documents alongside their influential GWAS of EA studies (Lee et al., 2018; Okbay et al., 2022):

When we and other scientists say that SNPs … ‘predict’ certain outcomes … we do not mean that the presence of an allele guarantees an outcome with certainty, or even with a high degree of likelihood. (Okbay et al., 2022: 7, in ‘Supplementary Data 1’)

Others, despite pleading caution on the difference between correlation and causation, simultaneously claim genes cause differences in outcomes. As Paige Harden puts it in her much-discussed book The Genetic Lottery, ‘genes’ ‘cause behavior’ (2021a: 109) and thus ‘DNA [may] be a systematic force in a child's life, putting its thumb on the scale at every point in an educational trajectory’ (ibid.: 70). Most audaciously, Asbury and Plomin (2014: 12) have speculated that ‘the technology will soon be available, for example, to use DNA “chips” to predict strengths and weaknesses for individual pupils and to use this information to put personalized strategies in place for them’. The strongest statements about the predictive and causative potential of ‘DNA’ are then being used to advocate for genomics-informed educational interventions.

In such statements, PGIs are reified as ‘genes’—as observable biological matter. ‘Polygenicity’ comes to represent what could be called the imagined genetic (cf. Pickersgill, 2018), a discursive reduction of molecular complexity. This reduction has been said to create a new ‘biological reality’ instead of merely describing it (Janssens, 2019: R148). PGIs, as the socially and statistically produced ‘reality’ of the imagined genetic, allow statistical data on socio-economic status and educational outcomes to be treated as explainable by genetic information ostensibly discovered within individuals. This reductive conceptualisation of PGIs as ‘genes’, in turn, authorises the formulation of policy proposals. Behaviour geneticists Kathryn Asbury and Robert Plomin (2014) favour a model of ‘personalised education’ informed by PGIs. Schools are advised to ‘draw out natural ability and build individual education plans for every single child’ (ibid.: 11–12). Others support using genetic predictors for school streaming (Visscher, 2022: 2) and for predicting learning difficulties (Asbury, McBride, and Rimfeld, 2021), or have proposed using genomic findings to construct control variables and inform interventions (Harden, 2021a: 192).

While these proposals, as I discuss later, are often politically and ethically opposed to one another, their proponents converge on understandings of the purpose of education and, in turn, of educational research. This is most evident by the critical contrasts they draw between genomic and social (i.e. non-genomic) educational research. The predictive power of GWASs is claimed to be superior to predictive models that centre on socio-economic status (Asbury, 2023; Asbury, McBride, and Rimfeld, 2021) and to surpass ‘the effects on EA estimated from policy changes or randomised intervention studies’ (Visscher, 2022: 1). But while in the past such criticisms concluded that educational reforms were futile (Panofsky, 2014: 161), today they conclude that interventions ‘fail’ because they are not informed by genomics. Plomin (Serious Science, 2017: 06:43–07:02), for example, remarks, ‘In education people just use their hunches.… I think they're going to have to deal with evidence and prove empirically, scientifically that their programs work, and then I think they will take genetics seriously.’ Harden (2021b: n.p.) is even harsher:

Ignoring genetics dooms much work in the social sciences to failure, wasting massive amounts of time and money and opportunities to improve people's lives.… Given the glaring flaw at the heart of much academic research, it's small wonder that most interventions don’t work, and that the body of research on which those interventions rest has been summarised as unreliable.

The value of educational research is then assessed in terms of its ability to ‘work’—to predict and improve performance outcomes cost-effectively. The often less-instrumental aims and ethos of social educational research are left out of the discussion. Promoting genomic educational research and its scientific capital is premised on its alignment with an evidence-based ‘what works’ agenda in education policy. In the UK, this has involved direct collaboration of scientists with the UK government's ‘what works’ policy network (Asbury, McBride, and Rimfeld, 2021). This agenda's exclusive focus on outcomes has been criticised for its reluctance to consider social contexts outside education (Lewis, 2017) and for evading public contestation on the purpose of education (Biesta, 2007)—a form of neoliberal depoliticisation (cf. Burnham, 2001) of education.

Educational genomics’ data-centric, probabilistic, predictive approach to genetics and education is then entangled, and resonates, with dominant educational policy ideas and practices. The scientific lens through which ‘biological reality’ is reconceptualised through PGIs is not merely driven by the datafication and molecularisation of the genetics of ‘complex behaviour’, or even by mere faith in the ‘truth, objectivity, and accuracy’ (boyd and Crawford, 2012: 663) of big genomic data. This lens is already directed towards particular objects and methods by the questions it seeks to answer, aimed at predicting and intervening in educational and economic outcomes. Such an orientation deprioritises reflection on the constructed nature of polygenic prediction or on the multiple meanings of education as a social and relational activity.

Epistemic premises: GxE, methodological atomism, and ancestry

Approaching education in terms of outcomes, educational genomics answers predominantly economic questions. However, the production of knowledge about the relationship between biology, education, and the economy is premised on epistemological assumptions about how knowledges about each can be combined. The disciplinary culture of behaviour genetics and sociogenomics has long been criticised for establishing a knowledge hierarchy that displaces social environments to a downstream position in the causal chain of explanation, or for ‘genetic reductionism’ (Bird, 2020; Bliss, 2018: 76–83; Coop and Przeworski, 2022: 850; Lewontin, Rose, and Kamin, 1984; Panofsky, 2014: 138–64). Tensions around this knowledge hierarchy are also evident in key current epistemic controversies characterising the field, which concern conceptions of the ‘environment’, gene-centred interpretation, and the question of ‘ancestry’ in educational genomics. As I detail in this section, however, it would be too simple to suggest that this knowledge hierarchy merely prioritises biology. Most educational genomics researchers attach importance to ‘gene-environment interactions’ (GxE), which understand genes as expressed within environments. The social lens required to understand genes in ‘environments’, is, however, dominated by the contemporary microeconomic conception of the social, which entails methodological individualism. The imagined genetic is placed at the root of individual agency, partitioned from social environments. This approach is particularly troubling when educational genomics comes to consider ‘ancestry’ as a dimension of its research.

Debates on the meaning of ‘environment’, the discernibility between ‘nature’ and ‘nurture’, and approaches to GxE have a long history in genetics. Tabery (2014) dates the debate back to the 1930s, between Ronald Fisher and Lancelot Hogben over eugenics. For Fisher, GxE was conceptualised as a statistical tool to partition genes from environment, which Tabery names a ‘biometric’ view dominant in population genetics. Lancelot Hogben supported instead a ‘developmental’ conception, whereby GxE denotes an always already biosocial ‘third class of variability’, the ‘combination of a particular genetic constitution with a particular kind of environment’ (Tabery, 2008: 720). Tabery (2008, 2014) notes the political dimension of this debate, in relation to not only eugenics but also Richard Lewontin's later intervention in the ‘race and IQ controversy’ in the 1970s. The debate continued in the 1990s, when Gottlieb (1995) criticised behaviour genetics’ unidirectional approach to GxE for failing to explore complex developmental mechanisms, and Turkheimer, Goldsmith, and Gottesman (1995) responded that GxE was necessarily premised on a conceptual partitioning of genes and environment. As is also clear in Tabery's account, the debate has extended into the 21st century (Bird, 2020; Feldman and Ramachandran, 2018; Keller, 2010; Moore and Shenk, 2017), with advocates of GWAS tending to favour the partitioning-biometric approach to GxE.

The partitioning GxE approach has not been superseded by a new ‘biosocial’ ontology as anticipated by social studies of neuroscience and epigenetics. Such studies have argued that notions of the plastic brain have begun to ‘allow nurture into nature’ (Meloni, 2016; Pitts-Taylor, 2016: 3). Yet the anticipation that debates over genetic and biological determinism will be displaced by post-genomic explanatory schemas that view genetic code as part of complex, ‘nonhierarchical networks’ (Rose, 2007: 130) by now seems optimistic. In contemporary sociogenomics, and by extension in educational genomics, the concern to partition genes from environments persists, and is evident in articulations of the social environment as an obstacle to modelling the causal effects of genes: as ‘geographic, ancestral, and/or socioeconomic confounding’, a form of ‘bias’ (Trejo and Domingue, 2018: 187) that can inflate results (Abdellaoui et al., 2022). It is acknowledged ‘some of the predictive power of the polygenic score reflects environmental amplification of the genetic effects’ (Lee et al., 2018: 1116). To avoid this, GWAS designs attempt to statistically control for ‘assortative mating’ (the tendency of individuals with similar phenotypes to mate with one another) and associated ‘population stratification effects’, which cause alleles to differ systematically because of how social geography and demographic history interact with ancestry (Blanc and Berg, 2020). Using ‘principal components analysis’ (PCA), another statistical technique invented in the eugenics era by Karl Pearson, genomic data is reduced to fewer, uncorrelated variables to create data clusters that, plotted spatially, are thought to represent genetic and geographic distances between populations. Yet PCA's ability to dispense with social confounds is questioned (Coop and Przeworski, 2022). For Zaidi and Mathieson (2020), it cannot capture the demographic history of recent millennia, so ‘polygenic scores are [still] biased in that they recapitulate environmental structure’.

For the majority of education genomic scientists, genetic effects are nevertheless isolatable by combining GWAS with family study design (sibling comparisons and parent-sibling trios) in GxE studies. Such studies are thought to ‘[peel] apart genetic and environmental pathways for intergenerational transmission’ (Harden and Koellinger, 2020: 571). The approaches and results of such studies vary, but, broadly, they explore how different levels of environments (family characteristics, schools, school districts) interact with EA PGIs to derive mechanisms of outcomes (e.g. Cheesman et al., 2022; Zhou et al., 2024). Developmental plasticity is acknowledged to mediate genetic variation, but is typically treated as another variable. By measuring ‘variance’ PGIs for ‘plasticity’ in response to educational reforms or environmental stressors (Johnson, Sotoudeh, and Conley, 2022: 1066), studies aim to identify ‘genotypically plastic individuals’ and design ‘precision’ educational interventions.

Some sociogenomics scholars raise concerns with this approach to environments: a lack of social, cultural, and historical complexity (Herd, Mills, and Dowd, 2021; Mills and Tropf, 2020), limiting analyses to behaviour, family, and school influence (Boardman, Daw, and Freese, 2013; Herd et al., 2019). More than that, others have questioned how this interpretive chain links imagined genes (rather than PGIs) to educational environments and outcomes. Matthews and Turkheimer (2021, 2022) have relentlessly argued that not only are such hypothetical GxE mechanisms untestable by means of GWAS, but the underlying biological mechanisms are almost impossible to identify. This is not for the lack of trying. GWASs of EA have explored whether causal genes for educational outcomes are located in the brain. In the 2018 GWAS, ‘biological annotation analysis’ was used to ‘identify the tissues or cell types where the causal genes are strongly expressed’ (Lee et al., 2018: 1123). The most significant SNPs, the analysis stated, ‘implicate genes involved in brain-development processes and neuron-to-neuron communication’ (ibid.: 1112). However, non-localisable ‘broadly expressed’ SNPs accounted for most of the heritability in EA. The PGI for EA could not be proven to represent a particular kind of genetic ability. The 2022 GWAS (Okbay et al., 2022) moved away from bioannotation analysis, as the increasing number of predictive SNPs makes it even harder to identify biological causal mechanisms. For Turkheimer, attempting to identify molecular genetic explanations for social outcomes via GWAS is based on an erroneous methodological atomism: ‘Educational attainment and divorce are not discernible entities at a genetic level of analysis’ (Turkheimer, 2016: 27).

Despite the lack of a mechanistic explanation that would validate a methodologically atomist approach, however, PGIs of EA are used to represent imagined genetic intellectual ability. This representation is linked to such GxE hypotheses’ unidirectionality, per Gottlieb's (1995) criticism. Researchers are concerned with ‘how the genome works in different environments’ (Gaysina, 2016: n.p.)—treating the genome as an ontologically and causally prior essence (Edelenbosch, Kupper, and Broerse, 2013). Even in GxE studies on how different socio-economic backgrounds are associated with measures of EA heritability (Woodley of Menie et al., 2021), and how historical sociopolitical changes have modified the predictive power of EA PGI for women across generations, PGIs are assumed to represent ‘cognitive ability’ (ibid.: 110) or ‘genetic indications of potential’ that are ‘actualized’ or not (Herd et al., 2019: 1076). Indeed, for Plomin (2018: 162) PGIs ‘can only be interpreted causally in one direction – from the polygenic score to the trait … [because] nothing in our brains, behaviour or environment can change inherited differences in DNA sequence’. This is what Meloni (2016: 1) has described as the modern idea of ‘hard heredity’, reproduced intact. With PGIs treated as primary biological matter, causality is linear from (imagined) ‘genes’ to ‘outcomes’, via environments. ‘Social’ concerns are thus firmly focused on whether human ‘potential’ is actualised in the form of measurable educational outcomes. For this question to be asked, molecular genetic causal mechanisms have to be imagined linking DNA to human abilities, and, in turn, to social phenomena like EA.

This unidirectional and atomist approach is compounded by a dominant methodological individualism, to shape an interpretive terminology of GxE in which ‘genes’ represent a pre-social psychology at the core of individual human agency. For example, Belsky et al. (2018: E7275) employed the increasingly popular concept of environmental transmission of genes by ‘genetic nurture’ (Kong et al., 2018) to interpret how mothers’ PGIs were better predictors of their children's EA than the children's own PGIs. ‘Genetic nurture’ suggests it is not mothers but mothers' genes that create nurturing environments for their children's educational outcomes. Genes are also imagined to select or create environments that enhance educational outcomes via children's choices at school and ‘parents responding to their children's genetic differences’ (Plomin, 2018: 6), a concept known as ‘niche-picking’ (Johnson, Deary, and Iacono, 2009: 476). Those genetically selected environments, which can include methods of screening and streaming, then ‘transmute invisible DNA into visible academic credentials’ (Harden, 2021a: 145–8). The ‘environment’ is thus the space where genetically determined pre-social subjects come to interact with one another, create environments, and realise their genetic ‘potential’ if provided with freedom of choice and ‘opportunities’. ‘Heritability is maximized when people are free to choose their own experiences, partly based on their genetic propensities’, Malanchini et al. (2020: 236) argue. ‘As such, heritability can be considered an index of equal opportunities.’

Far from a novel observation, this gene-centred and individualist epistemic discourse has been the object of political debate within late 20th-century biology. Lewontin, Rose, and Kamin (1984: 59–60), for example, connect the notion that ‘the gene is ontologically prior to the individual and the individual to society’ to the adoption of ‘economic concepts’ for biological analogies like ‘cost benefit analysis, investment opportunity costs, game theory’. Beyond a mere error, however, the political and economic influence of methodological individualism and atomism on the human sciences is hard to overestimate: not only for its centrality to von Mises's, Friedrich Von Hayek's, and Karl Popper's attacks against the influence of ‘collectivist’ social concepts like ‘class’ and ‘capitalism’ (Udehn, 2014: 111–18, 203), but also for its dominance in microeconomics and economic policy since the late 20th century. As a result, this epistemic approach offers an important point of resonance between educational genomics and microeconomic science, which, as I show later, eases interpretive transitions between statistical genetic associations and social mechanisms that are useful to economic theories of human capital. These mechanisms and theories are also relevant to sociopolitical theoretical questions of how economic and social hierarchies come into existence.

This preference for methodological individualism in social explanation, along with the methodologically atomist imaginative reduction of PGIs to ‘genetic potential’, can be highly problematic in light of the most crucial methodological problem in genomics: the concept of ancestry. In educational genomics, ‘ancestry’ classification is a concern primarily because ‘non-European ancestry’ DNA is excluded from study samples, so EA PGIs are not portable to those populations (SSGAC, 2021: 18). Over-representation of ‘European ancestry’ DNA has affected all genetics research since the beginnings of DNA discovery in the late 1980s (Mills and Rahal, 2020; Reardon, 2017: 46–69). But the contexts of data collection, rather than levels of representation, are more concerning. For example, in the US federal forensic database of non-consensually collected DNA ‘non-European ancestry’ is over-represented, reflecting racialised patterns of policing and access to medical care (Jabloner, 2021).

Classifying populations according to continental ancestry is also in question. The mapping of human DNA using geographic and ethnicity identifiers (the International Hap Map, the 1000 Genomes project) was controversial from the start (Gannett, 2001; Reardon, 2017: 70–93), but scientists have increasingly used shortcuts to categorise populations, grouping ‘ancestries’ by continent (Panofsky and Bliss, 2017) and attributing epidemiological differences to such ‘ancestry’, molecularising and biologising ‘race’ (Bolnick, 2008; Fullwiley, 2007; Lee, 2015; Smart et al., 2008). Ancestry classifications, derived by PCA (see above), have become systematically embedded as factual properties of DNA in biobanks used for educational genomic research. Besides a few exceptions (e.g. Giangrande and Turkheimer, 2022), most educational genomics researchers agree that, while ‘race is not a biological variable’ (Herd, Mills, and Dowd, 2021: 420), ‘the notion of continental ancestry having distinct genetic signatures is a biological reality’ (Conley and Fletcher, 2017: 111) correlated with socially defined racial categories. Harden (2021a: 91), for example, affirms that ‘racial groups differ in genetic ancestry, and so differ in which genetic variants are present and how common those variants are’. This issue was also characterised by ambivalence in the recent ethical consensus statement on educational genomics as a research area of ‘heightened concern’ (Meyer et al., 2023a). Despite scientists’ eventual agreement that ‘the genetics community should move as quickly as possible away from continental-level genetic ancestry groupings and labels’ (ibid.: S38), there was no consensus on whether ancestry group comparisons could one day become ‘scientifically valid’, and whether such ‘validity’ would render them ethically justifiable and valuable. This genomic approach to ‘ancestry’ has been criticised as a new ‘race science’ (Roberts, 2011) that must be scrutinised for its own culturally conditioned lens (Morning, 2014). As Lee (2015: 149) warns, ‘Genomics beckons the notion of race as naturalized by genes and the body becomes the ground from which racial difference can be discovered, excavated, and held up for inspection.’

The dominant epistemic assumptions of educational genomics discussed in this section—the separation of ‘genes’ from ‘environment’, methodological individualist understandings of GxE interactions, the methodological atomist reduction of PGIs to ‘genetic potential’, continental ‘ancestry’— are not founded on denying biopsychosocial developmental realities or the social construction of race. As Turkheimer, Goldsmith, and Gottesman (1995: 148; emphasis added) wrote in response to Gottlieb's (1995) criticisms, ‘One chooses linear additive models not because of a belief that the underlying relationships are in fact linear, but simply because they work, albeit rather imperfectly, for some useful scientific and practical purposes.’ As we have seen already, educational genomic research has repeatedly framed itself as seeking to solve problems of contemporary educational and economic governance. Its epistemic approach resonates with microeconomic epistemological and interpretive orientations, as well as assumptions about the social order and the purpose of education in it. These, however, come into conflict with ethical and political issues concerned with the history of eugenics and the previously dominant rhetoric in behaviour genetics that naturalised purported ‘race’ and class differences in intelligence. I turn to these debates below.

Genetics for ‘equality’? Race, class, and human capital

Inevitably, at the heart of the controversy over educational genomics are the provocative questions it seeks to answer: Are socio-economic inequalities, educational policies, and school environments secondary to, and separable from, ‘human endowments’? Are there biological mechanisms that explain why some children do better than others? And do genes explain even a small proportion of the ‘underachievement’ of disadvantaged children? What is the purpose of posing educational questions through a gene-centred lens, when doing so can open avenues for racialised and class-based discrimination? This section brings into relief how microeconomic human capital models influence not only educational genomic research questions and social interpretations pertaining to social hierarchies, but also researchers’ ethical discourse of translating educational genomic research at the service of ‘equality’. I discuss how this approach, along with the liberal political positions that inform many researchers’ stances on scientific freedom, enable speculation on socio-economic and racial hierarchies in innate intellectual ability.

The question of how educational genomics might deal with the variable of ‘ancestry’ is urgent. On 14 May 2022, before going on a racist shooting spree at a Buffalo supermarket, Payton Gendron cited the Lee et al. (2018) paper in a white supremacist essay he posted online. In response to the events and to hostile social media commentary, researchers in sociogenomics and bioethics wrote in Scientific American to set the record straight (Wedow, Martschenko, and Trejo, 2022). After explaining how the paper had been misappropriated, they acknowledged that ‘genetics has been used time and time again in service of white supremacy’ and pointed out that existing regulatory mechanisms for research ethics did not incentivise researchers to consider and mitigate any ‘social risks’ beyond individual risks to research subjects. Yet, the authors added, ‘we are not advocating for academic censorship here’, betraying a long-running concern in behaviour genetics that research may be stifled by external criticism (Panofsky, 2014), but also a dominant liberal stance in the field in favour of scientific freedom along with the belief it can deliver truth.

After the ‘race and IQ’ controversy of the 1970s, and the Bell Curve controversy in the 1990s, comparing genes and traits, including IQ, between socially defined ‘races’ has been avoided by most, though not all, researchers. Still, there is no consensus on the notion that comparing traits between genetically defined ‘ancestries’ is an invalid scientific endeavour (Meyer et al., 2023a), despite critics describing it as ‘inflammatory’ (Martschenko, Trejo, and Domingue, 2019: 7). For example, Conley and Fletcher (2017: 102–12) argue it would be methodologically near impossible to compare the polygenic indices of ancestry groups, since differences in genetic ancestry can correlate with racialised physiognomic differences and their social effects. Yet they also state ‘for scientific accuracy’ that ‘it could be that genetic variants that are not equally distributed by continental ancestry do, in fact, affect nervous system functionality in ways that we do not yet understand’ (ibid.: 110). Such enquiry is encouraged: ‘Social genomics reveals hidden dynamics of race that belie our intuitions. We cannot be afraid to look.’ Harden (2021a), besides endorsing scientific freedom, also contemplates future research findings of ‘ancestry’ differences:

What if, next year, there suddenly emerged scientific evidence showing that European-ancestry populations evolved in ways that made them genetically more prone, on average, to develop cognitive abilities of the sort that earn high test scores in school? (ibid.: 91)

This conjecture demonstrates how the epistemic premises examined in the previous section are operant in linking DNA to the ‘cognitive abilities’ of racialised groups. Social-geographic history and stratification disappear in the genetic concept of ‘ancestral evolution’ so that genes are the unidirectional driving force of human difference. ‘Ancestral evolution’ is divided by continent to connote skin colour. These genes—presumably derived by GWAS—are then imagined as the potential cause of ‘cognitive abilities’ and hence ‘test scores’. Harden then proposes an idea of ‘genetics-proof’ anti-racism (ibid.: 91): ‘ancestry’ differences ought to be discovered in the name of inclusion, and intervention should be tailored to address ancestry-linked genetic disadvantages. In other words, this ‘antiracism’ would be based on the possibility that racialised social disparities are caused by inherent ancestry-based deficiency—an idea that, for critics, ‘flirts with centuries of scientific racism’ (Henn et al., 2021: n.p.). In these examples, mainstream researchers disavow racial biologisation at the same time as they speculate on the existence of ancestral hierarchies in innate cognitive ability. Under such conditions, the inclusion of ‘non-European ancestry’ DNA data in educational attainment GWAS samples threatens to reopen the ‘race and IQ controversy’ in a far more devastating way, allowing the use of such data as evidence of racial ‘biological reality’.

For Harden (2021a: 235), dealing with such ‘differences’ need not be evaluative, but is a means to delivering ‘equality’—or to ‘reduc[ing] inequality of outcome’. This is possible, she contends, by using PGIs of EA to identify interventions that can maximise outcomes for genetically ‘unlucky’ pupils. Harden's ‘genetic lottery’ metaphor, drawn from liberal political theorist John Rawls's ‘theory of justice’, has been criticised for misrecognising systematic social inequities (Martschenko, 2021) and the non-random distribution of genes (Coop and Przeworski, 2022: 848). Despite facing such criticism, Harden positions herself on the left among scientists in educational genomics. She opposes as ‘eugenic’ proposals for using PGIs to classify pupils for ‘precision education’ (Asbury and Plomin, 2014; Plomin, 2018) and streaming (Visscher, 2022), which aim to enhance individual performance of the strongest students. Indeed, these views are becoming less and less popular among scientists, and Asbury has more recently shifted her proposals towards using genomic data to identify those at risk of poor performance and inform school funding decisions (Asbury, McBride, and Rimfeld, 2021). Yet, in both Harden's and Asbury's proposals to help the weakest students, introducing the precision medicine model into education brings along with it a clinical language and methodology that frames educational inequalities in terms of a deficit model—individual or group deficiencies for diagnosis, treatment, and intervention.

The case of genomic medicine is illustrative of how such medicalised propositions for promoting equality and inclusion can play out in practice. Not only has representation in genomic samples failed to improve medical access for racialised populations, but US policies of racial inclusion in medical research has actually encouraged the return of racial discourse through the notion of personalised medicine (Bliss, 2012; Duster, 2015; Tabery, 2014). ‘Race’, in this context, becomes an ‘epigenetic marker’ for categorising patients (Prainsack, 2015: 33). Driven by the pharmaceuticals industry, precision medicine has pushed research on plasticity and epigenetics not towards alleviating environmental harms but towards predicting risk and biologically stratifying individuals (Tabery, 2014)—a form of intervention that has been described as ‘epigenetic biopolitics’ (Mansfield, 2012) and even ‘biosocial determinism’ (Pitts-Taylor, 2019). These observations additionally suggest that the issue to address lies not so much in problems of data inclusivity or biological determinism, but in what social science knowledges, and governance and economic aims, inform the role of ‘environments’ in GxE studies.

Unlike personalised medicine, scientists in educational genomics are against its commercialisation for individual benefit, contra Nikolas Rose's (2007) description of new genomic science as geared towards a bioeconomy of individual self-enhancement. The SSGAC, for example, both warns against and leads research that would help discourage the private use of EA PGIs in embryo screening (Meyer et al., 2023b). The ambition of using advanced bioinformatics to complete the ‘puzzle’ of how DNA contributes to the creation of individuals capable of ‘success’ in the labour market is instead shaped by the prism of social ‘human capital’ governance (Becker, 1964) through state investments. Most directly, genoeconomic approaches use EA as an operationalisation of human capital (Benjamin et al., 2012). Malanchini et al. (2020) make those links explicit:

Educational attainment is a measure of human capital and is indicative of the skills of a population. As countries’ economies gradually shift away from mass production towards becoming knowledge economies, governments are eager to increase the skills and welfare of the population through educational attainment.… Higher levels of educational attainment are associated with higher employment rates, better job prospects and higher earnings. (ibid.: 229)

Seeking to refine their formulae, economists of human capital have turned to educational genomics to incorporate genetic and prenatal factors (Almond, Currie, and Duque, 2018) using a model proposed by James J. Heckman (2007: 13250) that locates ‘the origins of human inequality’ in the interaction between genetic and environmental human capital inputs and investments. Under this frame, many educational genomics scholars have organised and participated in scholarly events at Heckman's Center for the Economics of Human Development at the University of Chicago. In their own research, Madole and Harden (2023: 49) aspire to ‘integrate whole-genome measures from family members’ into human capital research to understand ‘how individual differences in constitutional factors influence treatment outcomes’. Heckman's approach is frequently cited by researchers (Asbury and Plomin, 2014; Harden, 2021a; Papageorge and Thom, 2020) and appears to inform research questions and interpretive approaches, embodying normative assumptions about the purpose of education and approaches to inequality, along with a discourse of rationalising education spending according to a logic of cost-efficiency measured in terms or relative social mobility.

The promise that genomic knowledge can deliver equality hinges on proving that GWAS identifies genetic ‘inequality’ independently of social inequality. Belsky et al. (2018) claim their research proves PGIs for EA are ‘not confounded’ by parental levels of education (Belsky et al., 2018), although, as Lee et al. (2018) reveal, when demographic characteristics such as income and parental education are controlled for, the PGI for EA predicts only 4.6% of variance, down from 11.4%—the headline finding. The common assertion that PGIs for EA provide explanatory power equivalent to the proportion explained by socio-economic status (Asbury, 2023; Harden, 2021a: 68) presumes these are independent from each other, overstating GWAS findings (Fletcher, 2022). An even stronger case has been made, however: even if socio-economic status predicts EA more reliably than PGIs (see Morris, Davies, and Davey Smith, 2020), or confounds them, ‘socioeconomic status is heritable’ (Asbury and Plomin, 2014: 128); an ‘uncomfortable truth’, taken to suggest genetic mechanisms might be causing ‘socioeconomic status’ in the first place.

Although the latter ‘supra-causal’ view of genes (Madole and Harden, 2023) comes from scientists on the right of the political spectrum, the hypothesised mechanism of such causality has been widely explored, and is evident in genomic studies linking PGIs of EA to regional deprivation and social mobility. These studies use microeconomic methodologically individualist interpretive concepts (endowments, preferences, risk-taking, choices). A widely cited study reports PGIs predict low EA along with adverse social and health outcomes in ex-coal-mining areas of Britain (Abdellaoui et al., 2019: 1339). It then examines the PGIs of those who have emigrated out of coal-mining areas and reports that ‘people with a genetic predisposition to higher cognitive abilities are leaving these regions’. Drawing on the understanding of migration as an independently chosen investment in human capital (Becker, 1964), the study portrays geographies of deprivation as driven or compounded by genetic cognitive strengths. Another strand of research links the heritability of ‘economic preferences, including risk-taking and financial decision-making’ (Fletcher, 2018: 13) to EA and socio-economic status. For example, Barth, Papageorge, and Thom (2020) have suggested the relationship between EA PGI and income might be explained by ‘genetic endowments’ that favour risk-taking and stock ownership— a kind of economic intelligence. Other microeconomic GxE research on EA PGIs and regional levels of social mobility (Fletcher and Jajtner, 2023) moves away from genetic determinism, yet firmly locates the problematic of ‘inequality’ in the interaction between genetic and environmental human capital inputs and investments. Discourses of this social and genetic ‘lottery’ participate in the human capital understanding of education as investment for producing ‘returns’ to individuals and ‘growth’ to states—the other side of its disposability when there is ‘little to no evidence of any benefit of graduating from college for low-ability individuals’ (Heckman, Humphries, and Veramendi, 2018: S227).

Conclusion

Post-genomic science and neoliberalism have been said to converge towards the calculability of individual risks and ‘a rhetorical insistence on the liberating virtues of privatising health insurance and dismantling welfare states’ (Rouvroy, 2008: 254). As I have shown, the convergences previously highlighted between educational genomics and the ‘neoliberal’ concerns of individualisation, tweaks, and nudges (Panofsky, 2015, 2021) are not accidental. Educational genomic studies have been motivated by, and participate in, an academic discourse that is influenced by the economic science of human capital development. The theory of human capital has long been considered paradigmatic of the neoliberal turn in economics and its associated governmentality (Foucault, 2008: 226), turning homo oeconomicus from bearer of labour-power to ‘entrepreneur’ of their human capital. But, as has also become clear with educational genomics, the governmentality of human capital does not merely relegate responsibility from society and states to individuals but also governs social human capital. As Foucault (ibid.: 227) already understood in his lecture of 7 March 1979, ‘the political problem of the use of genetics arises in terms of the formation, growth, accumulation, and improvement of human capital.’

Indeed, for today's educational genomics scholars, responsibility for educational genetic risk and potential should not be privatised but socialised, while discouraging a eugenic engagement with biological reproduction. The debate, instead, revolves around the politics of social human capital investments: whether to precision-invest to enhance those who are best ‘endowed’ (Plomin), or whether to precision-invest towards uplifting those out of genetic ‘luck’ (Harden). Currently, Harden's political perspective towards improving equality of outcomes appears to be the most influential. Her attempt to distance herself from the elitist and racialised social ‘eugenics’ of school streaming and selection for intelligence allows the field—at least the proportion of it aligned with her views—to position itself as enlightened vis-à-vis ‘the science of terrible men’ of the past (Harden, 2021b). However, this approach to producing educational ‘equality’, as well as associating economic and racialised disparities with ‘genes’, is also far from a model of welfare that would provide a broad range of resources to economically disadvantaged groups. As Heckman asserted in a recent interview (Rand and Heckman, 2023), his research shows that early years education programmes in disadvantaged neighbourhoods like Educare Chicago are unwarrantedly expensive. His cost–benefit analysis has instead identified minimal cost interventions that ‘work’, not in terms of equalising access to resources, but only in terms of health, crime, and social mobility statistics relative to the child's parents—similar demographics to those that PGIs of EA are claimed to predict (e.g. Abdellaoui et al., 2019).

Responsibility for such precision interventions may be socialised, but individual ‘educational investments’ nevertheless operate in the imaginative interpretations of how genes are transformed into human capital by creating or choosing facilitatory ‘environments’. Methodological individualism narrativises genes as carriers of ‘cognitive potential’ that strengthen individual agency towards socio-economic ‘success’. Rather than outright determinism, this narrative of the imagined genetic is premised on, and motivates, the relentless effort to partition genes and environment, setting aside biosocially orientated challenges, as well as the known difficulty of isolating social stratification effects. Such partitioning, in turn, enables the promise of genomics for ‘equality’—or, more precisely, social mobility: that imagined individual genetic potential can be facilitated or enhanced, and that plasticity itself can be an object of genetic prediction in order to tailor interventions and investments. Precision interventions to ‘improve people's lives’ (Harden, 2021b: n.p.) are a promise of enhancing outcomes in contemporary economically rationalised educational systems. It is in this context that discourses about ‘ancestral’ differences in cognitive abilities can be figured as ‘inclusivity’ in science, similar to the new forms of racialisation in medical genomics research.

Within the longer history of behaviour genetics’ engagements with education, then, the rise of data-intensive, predictive genomic science tends to subsume the controversial genetics of intelligence under contemporary economic rationalities. Genetic fatalism turns into the management of plasticity, and cost-effectiveness entails not abolishing but fine-tuning precision interventions to replace ‘wasteful’ educational programmes. Promoting an economic rationalist view of education is part of the broader investment of neoliberal regimes in the depoliticisation of social institutions (Burnham, 2001), including education (Ball, 2021), by pushing further away from the democratic agenda the question of what education is for.

The dominant trend in educational genomics is to bracket messy biosocial, cultural, and political realities so as to devise genomic methodologies that ‘work’ in such terms. The risk is that methodological choices can come to be treated as realities through governance and commercial applications of research. Unapproved and unethical applications are therefore attributable not only to public misapprehension of science. Science should not take as given the ontologies, variables, concepts, and questions predefined by the concerns of economic governance in contemporary capitalism. Truly challenging the past of behaviour genetics, then, involves deeply questioning the ‘culturally conditioned lens’ (Morning, 2014) of this methodological bracketing. This would entail not only pursuing methodologies that move beyond the partitioning of nature and social history, of biological bodies and their social relationships. It demands imagining education—and human activity and its subjects—open-endedly, away from classificatory conceptualisations that readily submit to the instrumentalist gaze of contemporary governance.