ChatGPT's 'Prefrontal Cortex Problems': A Curious Experiment in AI Cognitive Testing
2025-01-12
The author administered a series of cognitive tests, including the clock drawing test, to ChatGPT, revealing symptoms akin to those exhibited by humans with prefrontal cortex damage, such as poor spatial organization and planning deficits. While ChatGPT can programmatically generate correct clock images, it consistently fails when directly drawing or describing them textually. This leads the author to ponder AI cognitive abilities, supervisory mechanisms, and the ethical risks of endowing AI with higher cognitive functions. The conclusion is that current AI models struggle with human tasks, prompting suggestions for AI governance and legislation.
Read more