Degenerative AI?

Paulo Coutinho

Generative systems, widely adopted without intellectual discipline, could erode the cognitive capacities education once labored to build – knowledge, character and ethics – psychologists, educators and neuroscientists argue. The issue is not that machines will outthink us. It is that we may gradually outsource the very effort that makes thinking possible.

Psychologist Daniel Wegner’s theory of “transactive memory” helps frame the risk. He contends that humans naturally distribute knowledge across social networks: “We remember who knows rather than what is known.” Later studies showed that when people expect information to be digitally stored, they recall fewer details themselves, meaning that “external memory” alters “internal memory.” Generative AI extends this principle from facts to reasoning.

Other scientists argue that deep learning requires effortful processing – struggle is not incidental to understanding; it is the mechanism by which durable mental schemas form. When a system drafts essays, solves equations or synthesizes arguments instantly, it reduces cognitive load. For experts, that may increase productivity. For novices, it may short-circuit the formative stage in which mental architecture is built.

Neural circuits strengthen through repeated activation, neuroscientists assert. If students routinely outsource composition, translation or analysis, those pathways will not disappear, but they may thin. “Selective atrophy” is plausible, particularly in working memory and sustained attention.

Nicholas Carr argued in his seminal book “The Shallows” that digital environments fragment concentration and discourage deep reading. Generative AI intensifies that dynamic. It does not merely deliver information faster – it performs structured thought. With fine irony, the temptation is not only to skim but to delegate.

Technological pessimism has a poor track record. Calculators diminished mental arithmetic without destroying mathematics. Writing reduced memorization while expanding philosophy. Philosophers Andy Clark and David Chalmers’ “extended mind” thesis suggests that tools can become genuine components of cognition. The question is not whether minds extend into technology, but whether internal capacities are maintained alongside that extension.

The deeper risk may be cultural rather than neurological.

Renowned French sociologist Pierre Bourdieu described education as “the transmission of cultural capital,” the acquisition of taste and judgment. If learners rely on generated summaries rather than primary texts, they may acquire information without developing discernment. A society fluent in prompting yet weak in discrimination would not be unintelligent, but it might be intellectually thinner.

Marshall McLuhan warned that media reshape habits of mind. Generative AI, as a medium of synthetic reasoning, may recalibrate expectations about effort itself. When friction disappears, mastery can appear inefficient and convenience becomes normative.

Still, degeneration is not inevitable. Used deliberately, AI can function as a “Socratic partner,” offering feedback, challenging assumptions and expanding access to expertise. Under demanding institutional standards, it could elevate rather than diminish intellectual performance.

The decisive variable is not the machine; it is the culture surrounding it. If education continues to require independent reasoning, defense under constraint and sustained engagement with difficulty, cognitive resilience will endure. If effort quietly becomes optional, dependency will deepen.

Artificial intelligence will not hollow out the human brain on its own. But a civilization that abandons disciplined struggle may discover that intelligence, like muscle, weakens when it is no longer required.

The same could be said about democracy – which also weakens when citizens outsource judgment and abandon disciplined engagement.