We developed a method called Brain2Model Transfer (B2M) that uses human brain activity, recorded via EEG or invasive intracranial electrodes, as a teacher signal for AI models. When we align model representations to neural data, models train faster and generalize better, consuming less data to achieve equivalent performance to brain-less learning. We tested B2M on two proof-of-concept tasks, and, in both tested cases B2M-trained models required less data and learned faster.
