A new study suggests artificial intelligence systems may not replicate human brain function as closely as previously assumed, challenging a foundational premise in neuroscience research. Scientists found that AI models capable of accurately predicting brain activity can still rely on internal computational processes that the human brain does not employ, according to findings published in the field of Science. The discovery indicates that a model's predictive accuracy alone may not confirm it mirrors biological neural mechanisms, raising concerns about how such systems are used to draw conclusions about human cognition.
Researchers caution that these findings could affect the reliability of AI as a tool for neuroscience, particularly when models are treated as functional stand-ins for biological brain activity. The study underscores a growing debate over whether AI's predictive success reflects genuine mechanistic similarity or merely surface-level correlation. Scientists involved say the results call for more rigorous validation when interpreting AI outputs as evidence of how the brain works, and could prompt a reassessment of assumptions underlying a wide range of studies that depend on artificial models to explain neural processes.
Independent reporting synthesized from primary wires and global correspondents.