A formal economic model frames AI adoption as cognitive leverage — and concludes the riskiest period looks, from the inside, exactly like success.
A working paper applies financial economics to AI adoption, coining the term "cognitive debt" for the unverified reasoning obligations that accumulate when people use AI as a substitute for thinking rather than a complement to it. The model tracks two variables per person: cognitive capital (what you can do unaided) and cognitive debt (the reasoning you've outsourced and never validated). Like financial leverage, cognitive debt is self-reinforcing: productivity gains during calm periods lower subjective risk assessments, which encourages heavier AI substitution, which deepens the debt. The authors call the resulting inflection point a "cognitive Minsky moment" — borrowed from the economist Hyman Minsky, who argued that financial stability quietly generates the conditions for a crash.
Two findings stand out. First, the model predicts decentralized markets will systematically over-adopt substitutive AI relative to what's socially optimal, because individuals don't bear the full cost of their cognitive offloading — a classic externality problem. Second, and more unsettling: high-skill people who adopt AI most intensively could eventually erode their unaided cognitive capital below that of lower-skilled people who adopted less. The people most confident in their ability to leverage AI may be the ones most exposed when it fails.
The paper also names a "false-correction loop" — the tendency to respond to AI failures by deploying more AI. Anyone watching enterprise teams debug LLM outputs with other LLMs will recognize the pattern.