Researchers tested ChatGPT on the ancient "doubling the square" problem, which has long debated whether mathematical knowledge is innate or learned. Since the problem's solution isn't easily found in text data, scientists predicted ChatGPT's success would indicate learned ability. The AI generated a flawed solution, similar to Socrates' student, suggesting a "learner-like" approach. This implies AI might operate within a "zone of proximal development," improvising solutions rather than recalling them directly from training data, highlighting the need for critical evaluation of AI-generated proofs.

Prepared by Olivia Bennett and reviewed by editorial team.