Collective Illusions of Understanding in NeuroAI.

Large language models and other deep neural networks often mimic human behavior well, which has led to strong claims about human and machine intelligence alignment. We argue that such claims reflect an illusory understanding of human mind and brain: Authors routinely equate predictive success with understanding of causal mechanisms-a form of confusing correlation with causation. The illusion is compounded by neglect of the importance of falsification through causal interventions. We speculate on some of the cultural factors that fuel collective illusions of understanding in neuroAI.