We are interested in question such as, how does the mind combine simple concepts to produce infinitely new meanings, draw analogies across diverse situations, or use abstract structure to support conceptual and social cognition? How are these abilities made possible by neural mechanisms in the human brain? And in what ways can they be implemented in artificial neural networks? We study these problems jointly in minds, brains and machines using behavior, fMRI, cognitive computational models, and interpretability techniques in neural networks.

Emergent relational abilities in recently developed large language models present a new opportunity to test hypotheses about how neural networks can achieve these feats and develop novel hypotheses for biological networks. We also believe that the methods of cognitive science are critical for shedding light on black-box AI systems and addressing pressing problems in AI interpretability and alignment.