To gear up for your return, tryLe Grand Continent. The ideas and analyses you won’t find anywhere else. But you’ll see them everywhere soon. In your inbox from eight euros per month
Collective human cognition, which could be succinctly defined as the product, at the scale of a society, of the computational power and the entire set of emotional and behavioral processes inherent to our neurobiological structures, is it seeing its regeneration mechanisms destroyed by AI?
Since 2021, a report by François du Cluzel presents the “cognitive war” as a sixth dimension in its own right, on par with land, sea, air, space, and cyber space. In 2024, NATO publishes its first development agenda whose multi-domain operational planning integrates cognitive warfare alongside the five other dimensions of warfare . At the same time, the mathematical community is undergoing an existential crisis concerning its role, its future, and its “added value” in a world where AI is now capable of solving problems at the frontier of our knowledge (such as the Erdős conjecture ) and of producing publishable articles in the most prestigious journals . On one side, Fields medalist Terence Tao envisions a world of “happy collaboration” between humans and AI to produce ever more mathematics. On the other, the newly crowned Fields medalist Jacob Tsimerman sees AI as the end of mathematics (even of humanity). Fatalistic, he announces he will quit research to join OpenAI as soon as this prize, intended to accelerate the careers of young researchers, is awarded: for him, the only choice would be to use AI to have a positive “inside” impact.
In this context, the American researcher Nolan Lovett recently published an article in which he reflects on how artificial intelligence “empties” our “collective cognition” . He lays out a series of predictions if the current trajectory continues with the same incentive mechanisms and the same regulations. According to him, in the short term AI gives the impression of augmenting human cognition and productivity, but it is actually a mirage, because in the medium and long term it will degrade our collective cognition, prevent its regeneration, and ultimately cause humanity to regress. The phenomenon, he explains, resembles overfishing: in the short term there are more fish and the illusion of progress, but in the medium and long term we end up with none. Ultimately, he warns, we will no longer possess brains capable of understanding Poisson’s law without AI. He draws a parallel between the depletion of this “collective cognition” and that of natural resources theorized in 1968 by Garrett Hardin in his Tragedy of the Commons.
Collective cognition as a common good
Hardin indeed describes how rational individuals acting independently, in their own interest, inevitably exhaust the collective resources we all depend on. He calls this “the tragedy of the commons,” defined as resources accessible to all (non-excludable) but whose use by one person reduces the amount available to others, particularly rivals or competitors. He distinguishes two levels of “common resources”: the planet’s carrying capacity, and natural resources. With a modern eye, one can be critical of the first level proposed. His fear of overpopulation and his virulent critique of natalist policies that, in his view, “make the cost of reproduction borne by society as a whole” have aged poorly.
However, his second level of “commons,” namely natural resources, attracts more attention, and it is to this that Lovett ties the idea of “collective cognition.” The regeneration of this commons would be imperiled precisely by AI, which would destroy the cognitive friction spaces and times from which it arises. It eliminates early-career jobs during which expertise sharpens itself through direct contact with the rough realities of professional practice in business. It also shortens the often painful learning periods required to acquire a new skill through trial-and-error and the intense friction of schooling and university, by providing immediate, tailored solutions. AI would thus be responsible for a “tragedy of the common cognition.” To support the idea that collective cognition is a commons, the author assigns it three key characteristics: it depends on everyone, it is non-exclusive, and it withers once its regenerative capacities are compromised.
First, collective cognition depends on everyone. An organization seeking to hire relies on a talent pool created in common. Hence the problem of the free rider: each organization benefits from recruiting experts from this pool but has no incentive to bear the cost of developing new “talents,” since its competitors, who would not have invested resources, would profit as well. Second, collective cognition is non-excludable. No organization can easily prevent its rivals from benefiting. The mobility of experts and the multitude of channels for transferring knowledge ensure that those who do not feed the pool will nonetheless draw from it. In 1965, Olson showed that such a configuration rewards the free rider. Finally, third, collective cognition withers if there is no longer the capacity to regenerate it. This is arguably the most important point.
Just as common natural resources are depleted by overuse and harvesting beyond their regenerative capacity, collective cognition disappears through the outright destruction of its regeneration mechanisms. The arrival of AI in the workplace destroys both jobs and the mechanisms that regenerate collective cognition. While the concern is legitimate, focusing only on potential job destruction at the expense of these regeneration mechanisms risks overlooking the full iceberg. Recent work shows a systematic elimination of early-career positions, which many observers and policymakers treat as a simple “market adjustment” driven by automation waves, akin to previous cycles, without considering the impact on the regeneration of collective cognition.
Elinor Ostrom and polycentric governance
Is the tragedy of the common cognition inevitable? In 1990, Elinor Ostrom, who would later win the Nobel Prize in economics in 2009, argued that it is not and proposed an approach grounded in the observation of communities that exploit common resources without exhausting them. She derives from this a governance framework based on polycentric governance that ensures the sustainability of the commons. This governance rests on eight key principles:
- Clear boundaries: everyone knows which resources they may use and up to what limits.
- Proportional equivalence allowing adaptation to different local specifics.
- Rules arising from collective choices: those affected by the rules are the ones who can modify them.
- Active monitoring: the resource and its users must be actively monitored and the monitors must report to the community dependent on the resource.
- Graduated sanctions: punishments appropriate to the different possible transgressions concerning the resource.
- Conflict-resolution mechanisms: the community dependent on the resource must have venues and processes to arbitrate disputes quickly and at low cost so as to resolve them before they escalate.
- Minimal recognition of the right to organize: external authorities must recognize the users’ right to create their own institutions.
- Nested levels: for large resources, governance must be organized across different embedded levels (local, regional, national). This is polycentric governance.
Nolan Lovett applies these principles to the “tragedy of the common cognition” and derives three nested levels of action:
- The organizational level: each organization can, for example, set up “IA-free learning spaces,” or a recruitment onboarding process that includes gradual access to all AI tools.
- The level of professional bodies (Medical Association, Bar, Chamber of Industry, etc.): for example, in continuing education or the issuance of certifications or licenses to practice, requiring a period of practice without AI or the passing of tests without AI assistance.
- The level of public authority: for example, subsidies or tax credits to companies that maintain employment.