This theoretical exploration compares erroneous outputs in AI systems with false perceptions in human cognition, revealing striking parallels in predictive processing mechanisms.

• Both humans and LLMs engage in predictive processes that can generate outputs disconnected from reality
• Human brains actively fill information gaps under uncertainty, similar to how LLMs generate completions
• Neural mechanisms for error correction exist in both systems but function differently
• Understanding these parallels offers insights for developing more reliable AI systems

For biology and neuroscience, this research provides a comparative framework to understand human cognition through the lens of current AI limitations, potentially informing both improved neural models and more human-like artificial intelligence.

I Think, Therefore I Hallucinate: Minds, Machines, and the Art of Being Wrong