Work

From comparison to transformation: studying minds that learn from machines

Year: 2026
Type: preprint
Abstract: For seventy years, minds and machines have been studied by comparison: can machines model, imitate or outperform us? In bounded domains this comparison has saturated, and machines now produce certified knowledge that no person can inspect; a complete Lean formalization of Fermat’s Last Theorem runs to 13.5 million lines. We argue that the object of study must move from the performance gap to the change in minds that learn from machines. We propose Reverse Cognition: a search for faithful, explicitly limited representations through which a person can acquire a certified structure and keep it once the machine is withdrawn. In the public formalization, only 5.7% of the theorems beneath the four steps of the human proof belong to a single step. We define audit criteria for representational routes, machine operations that could make such objects followable, predictions that separate appropriation from offloading, and a study sequence to test them.
Language: English
Open Access status: green
OpenAlex ID: W7233585408
Cites:
Cited by:
Related to: