Wharton coins 'cognitive surrender' as users let AI decide

A short piece in Business Insider puts a name on something a lot of people have started doing quietly: handing small decisions, and then larger ones, to a chatbot. The name comes from Wharton researchers Steven Shaw and Gideon Nave, who call it 'cognitive surrender' — the moment a person adopts an AI's output as their own with minimal scrutiny, overriding both intuition and deliberation.

The numbers from their study, reported by TNW, are the part worth sitting with. Across 1,372 participants given a chatbot that sometimes returned a confident wrong answer, people accepted the wrong answer 80% of the time. They also rated their own confidence 11.7% higher than participants who reasoned without AI. When the model was right, accuracy jumped 25 percentage points above baseline; when it was wrong, accuracy fell 15 points below the baseline of people working alone. So the assistance is real, and the failure mode is real, and the user does not notice the difference.

Shaw and Nave frame this as a 'Tri-System Theory' extension of Kahneman, where AI-assisted cognition sits as a third system alongside fast intuition and slow deliberation. The consumer side of the story is already moving faster than the academic one. Business Insider points to financial writer Dominic Frisby, who wrote on Substack that a chatbot's relationship advice felt more useful than what a human friend had offered, and to an app called Moot, launched earlier in 2026, that submits user dilemmas to a panel of five AI personas — The General, The Sage, The Skeptic, The Diplomat, The Architect — who debate and then vote.

The honest caveat is that this is one study with a controlled chatbot, not a measurement of how people behave with their actual ChatGPT history, and the reporting does not tell us how durable that 11.7% confidence bump is once a user gets burned. What it does suggest is that the interesting design problem for the next wave of decision apps is not making the answer better, it is making the user think again before taking it. Friction, dissent, and visible disagreement may be the features that age well.