Interactive diagnostic · 5 minutes

Can you tell when AI is wrong?

Eight short problems ahead. On each one you can — but don't have to — consult an AI assistant. Some of its advice is correct, some deliberately faulty, following the design of the Wharton School experiment this test is a mini version of.

  • 8 problems, about 5 minutes

  • An AI advisor on every problem — some of its advice is wrong

  • After each answer you rate your confidence

  • Nothing is sent anywhere, no email — only you see your results

We measure three things: when you consult the AI, when you trust it, and whether your confidence matches how well you actually do.

Fair warning: the AI assistant always sounds sure. It's only sometimes right.

How this test was built

The test follows the protocol of the Wharton cognitive-surrender study: short problems with an intuitive wrong answer within reach, an optional AI assistant with shuffled advice quality, and a confidence rating after every answer. It is shorter than the original experiment and runs without incentives — treat the results as indicative.

Shaw, S. D., & Nave, G. (2026). Thinking—Fast, Slow, and Artificial: How AI Is Reshaping Human Reasoning and the Rise of Cognitive Surrender. Working paper, The Wharton School, University of Pennsylvania (preprint, SSRN No. 6097646; not yet peer-reviewed).