A recent paper titled “Your brain on ChatGPT” has generated a lot of controversy and dispute. The much-discussed findings from a team based at MIT were reported to indicate that generative AI use for essay writing induces damaging “neural and behavioral consequences” when measured with brain scanning technologies.

Though the central claim that generative AI dependence may induce “cognitive debt” among student users has been widely debated, the far-reaching circulation of the results in the press and on social media shows one clear thing: a contemporary scientific, public and media fascination with locating learning processes and outcomes in the brain. Such findings have only been possible to produce using new “neurotechnologies”.

Neurotechnologies are now being used in a wide range of studies in the field of educational neuroscience, as my colleagues Jessica Pykett, Dimitra Kotouza and I have explored in a new article in Pedagogy, Culture and Society. These brain technologies include brainwave sensing devices, neuroimaging, brain-computer interfaces, brain simulations, neural network modelling, and even neurofeedback and neurostimulator technologies.

For scientists in this domain, these new brain technologies promise breakthroughs in how the neural correlates of learning are understood, with significant implications for translation into pedagogic practices or educational policies. The MIT study is one spectacular case of a rapid growth in neurotechnology-based scientific studies in education that are intended to transform how we understand the neural correlates of learning.

Neurotechnologies in education

Neurotechnologies such as brain imaging have existed for decades, but in the last twenty years there has been an explosion of neurotech development, especially as new instruments have been devised for the collection and analysis of neural data. Recent developments in artificial intelligence are even leading to a synthesis widely described as “NeuroAI”, accompanied by grand claims of potential transformations in the scientific understanding of the human brain. At the same time, there remain significant scientific, ethical and regulatory debates about the impacts of new neurotechnologies on scientific knowledge production and human subjects.

Rising interest in the brain sciences in education over recent decades stems from the availability of new neurotechnologies, alongside strong support from international organizations including the OECD and UNESCO. Although educational neuroscience has long faced barriers to translation into practice, new technologies and data are believed to make its findings more relevant for teaching and policy. According to a recent special issue editorial on the current development and future potential of neurotechnologies in education:

Neurotechnology comprises a range of techniques that offer indications about the operation of the brain separate from its manifestation in the kinds of behaviour that educators typically monitor to track students’ progress in learning. Its use is therefore predicated on the assumption that the way that learning is implemented in the brain will be relevant for educators…. These technologies might directly reflect physiological markers of brain function, such as in the brain’s electrical discharges (EEG) or its oxygenated blood flow (fNIRs); they may reflect body markers of the operation of the sympathetic autonomic nervous system, often indexing emotional processes (electrodermal activity); or they may detect subtle behavioural markers reflecting attention processes or memory retrieval (eye-gaze). Together, these measures can offer a window on students’ engagement in the classroom, their current knowledge, and the nature of learning as it unfolds.

Central to the promise of neurotechnologies in education is real-time recording and imaging of brain data, plus the mobilization of “machine learning algorithms to decode brain activation patterns” and other embrained processes.

Reading across dozens of scientific articles, reports and conference papers for our project, we wanted to figure out in our analysis how neurotechnologies are being used to develop new conceptualizations of what is widely termed the “learning brain”. How in other words is the learning brain being algorithmically decoded via neurotechnologies? That meant looking closely at the specific neurotechnological “setups” through which the learning brain is made visible and legible, as they are described in published neuroscientific studies.

Our conceptual orientation in this study, drawing on sociology of science approaches as we report in the article, was that scientific technologies are not merely tools that scientists use to make discoveries. Rather, technologies such as scientific instruments of measurement and analysis themselves mediate and shape the objects of study, affecting how phenomena become visible to and knowable by scientists: “Without the presence of technologies such as fMRI, the phenomena to be studied in cognitive neuroscience could not become present to scientists”. Investigating this technical mediation of scientific knowledge production for us meant thinking about how the learning brain is itself produced or fabricated from digital data collected and analyzed with neurotechnologies.

The learning brain, we have argued, is being conceived in terms of “brain facts” that are assembled out of neural information through highly-data-intensive and computational methods. That’s why we refer in the article to a “neuro-informationalization” of the learning brain as a “digital epistemic object” of scientific attention. What this means is that the scientific setup of investigation—the instruments used, the institutional priorities underpinning the study, the types of information collected—all play a part in constituting the findings, and in how the learning brain is rendered as authoritative scientific knowledge in published results, findings and presentations.

Moreover, once the learning brain has been rendered in this way, it can become the basis for “neurogovernance”, by which we mean interventions based on knowledge about the brain that are intended to impact in various ways on neural capacities. Perhaps the clearest example of educational neurogovernance is the use of brain stimulation for direct cognitive enhancement, though it also includes less invasive interventions such as the use of commercial brain-training platforms based on claims of neuroscientific expertise or other forms of “brain-awareness” teaching.

The study, in other words, was designed to examine the scientific setups involved in generating knowledge about the learning brain and to explore the implications in terms of proposed interventions based on the knowledge produced with neurotechnologies.  

Kinds of learning brains

Our key observation as we examined educational neuroscience papers was that the “learning brain” does not necessarily appear as a single coherent object of analysis and description. Instead, what we documented was several kinds of learning brains that have been produced with neurotechnologies. Different experimental setups produce quite distinctive conceptualizations of what a learning brain is, and this leads to different proposals for translation into pedagogic or policy intervention.

In our analysis we surfaced four key kinds of learning brains:

A plastic learning brain imprinted by social influences that is legible through neuroimaging technologies. The plastic learning brain is most clearly illustrated by studies exploring the impact of socioeconomic disadvantage on neural and cognitive development and capacity. Such studies have led to proposals for early years educational intervention, particularly in maths and literacy, to support those children who, it is claimed, have measurable biomarkers of poverty in the brain that can be detected with neural scanning technologies.

A synchronized learning brain characterized by “brain-to-brain coupling” between teachers and students. Recent “hyperscanning” research deploys mobile neurotechnologies to study the neural aspects of interactive learning in real-world classroom contexts rather than the artificial setting of the lab, focusing specifically on “interbrain” coherence between teachers and students. This is leading to proposals for pedagogic interventions that enhance interbrain cognition.

An attentive learning brain understood via brainwave monitoring neurotechnologies. Commercial companies and nonprofit research centres have developed neurotechnology applications that are intended to measure student attention and engagement, and then to provide neurofeedback to improve attentional states. Here, the attentive learning brain is known through consumer neuroheadset devices in terms of brainwave oscillations, and then targeted for automated forms of neurofeedback. For example an MIT project has developed a pair of wearable neurofeedback spectacles for students that delivers a “haptic nudge” to prompt re-engagement if the device senses lapses of attention.  

A computational learning brain conceived through neurocomputational brain modelling methods. Drawing on fields of neurocomputation and connectomics, educational neuroscience has begun to use mathematical instruments, theories and AI to study, map, model and conceptualise the brain in terms of “connectivity”, neural “circuitry” and “information integration”. The implication of the neurocomputational understanding of the learning brain is that educational AI applications could then be used “to provide adaptive, in-the-moment learning and teaching support, for example, pedagogical feedback to learners, or advice to teachers on how to support individual learners in context”.

Our intention in the article was not to dispute the findings of educational neuroscience, nor to suggest that these different kinds of learning brains demonstrate some methodological or conceptual shortcomings. Rather, we were interested in how the proliferating claims of educational neuroscience as a source of knowledge about the neural substrates of learning gives this domain growing authority as the basis for educational interventions.

By conceiving the learning brain in terms, variously, of plasticity, synchronicity, attention and neurocomputations, educational neuroscience can become the basis of a wider range of pedagogic and policy development. The increasing visibility and legibility of the learning brain makes it increasingly governable and manageable. And that in turn can also lead to proposals for acting upon learning brains for purposes that are political and economic rather than merely pedagogic.

The OECD, for instance, sees brain training as beneficial for the development of a “brain economy”:

The modern global Brain Economy puts a premium on individual brain skills, but even more importantly, it demands the networking of individual brains to communicate and work together; this requires social intelligence resting on interpersonal skills and social perception and related abilities. The jobs of the future, for which we must prepare today, will increasingly value an individual’s cognitive, emotional, and social brain resources. … Thus, this is a critical time to affirm the primacy of human and brain capital—that is, people—in the twenty-first century workplace and at both the production and consumption ends of the economy.

Here we can see how aspects of plasticity, interbrain networking and attentive engagement could all be invoked in pedagogic and policy proposals to support the political and economic project of building a so-called brain economy. A learning brain could quite easily become a capitalized learning brain.

The “diminished” learning brain?

When the article was peer reviewed, one of the reviewers gave away their professional identity (though not their personal identity) as an educational neuroscientist. Their feedback on points of accuracy was extremely helpful, given the article is broadly critical in orientation. But we were unable due to space constraints to respond fully to one point of feedback: might there not also be other kinds of learning brains in addition to the four we identified?

Perhaps another kind of learning brain is now becoming visible through studies such as the analysis of the effects of ChatGPT on students’ brains. When the paper was published by researchers from MIT, it was widely reported as showing that generative AI negatively impacts the brain. While the authors were much more cautious about their findings than press and social media coverage suggested, they did offer the concept of “cognitive debt” as a long-term consequence of LLM reliance with significant educational implications.

The MIT study itself was carried out using an electroencephalograph (EEG) brainwave scanning device called AttentivU—the very same wearable device that the lead author of the study previously developed to monitor and nudge student attention—built on the Enobio Neuroelectrics headset and BioSignal Recorder application. In other words, the study was highly leveraged by neurotechnologies and enacted through methods that have become increasingly commonplace in educational neuroscience. It represents a contemporary instantiation of how educational neuroscience is extending brain-based explanations to a growing range of learning processes and outcomes, enabled by neurotechnology advances in data collection and analysis.

What it proposes, however cautiously, is another kind of learning brain: the diminished learning brain, affected by an accumulation of cognitive debt induced by overdependence on technologies like AI. The study makes the diminished brain and its cognitive debt into digital objects of knowledge and potentially subsequent educational interventions.

Indeed, there is growing debate about “technology-induced cognitive diminishment” among those who have studied and theorized the implications of neurotechnologies and neuroAI. Some now argue there is “empirical evidence that ICTs can contribute to impairments in various cognitive and affective processes”. The clear implication of brain diminishment research would be regulatory programs and remedial pedagogies of brain protection.

While a scientific assessment of these claims is beyond the scope of our research, it is clear from the MIT study and related conceptualizing that specific neurotechnology setups are now being designed to make the diminished learning brain visible and legible as an object to intervene in. Indeed, due to the MIT paper and its reception, the diminished brain is already an object of significant public attention and media controversy.

The remaining question is what kinds of neurogovernance interventions may be proposed and devised to act upon the diminished learning brain, such as for protection, harm-reduction or enhancement, and with what purposes and consequences.

The article, “Learning brains: educational neuroscience, neurotechnology and neuropedagogy”, published in Pedagogy, Culture and Society, is an outcome of a research project grant awarded by The Leverhulme Trust on the rise of data-intensive biology in education, and is available open access.