Experimental evidence increasingly demonstrates that human brain data can be leveraged to train artificial intelligence models. In this context, neuroimaging data is both valuable, since it can improve model performance and could be prioritized for high-value steps in model training, and open-ended, since future neuroscience discoveries could retroactively identify new neural signals of interest in current neuroimaging datasets. The emergence of human brain data as a valuable and open-ended resource for artificial intelligence, including for foundation models, could translate into a new and strategic role for neuroscientists, and open new possibilities for the simulation of human cognition.
Maël Donoso (2026) studied this question.