Transferability of deep learning successes to Theory of Mind

Determine whether any of the successes of deep learning in vision and language tasks carry over to Theory of Mind in artificial systems, enabling comparable progress in modeling and performance for inferring others’ mental states.

Background

The paper contrasts the substantial progress of deep learning in vision and language—where strong performance and brain–model correspondences have been observed—with the greater complexity and ambiguity of Theory of Mind. The authors caution that the conditions enabling success in vision and language (e.g., clear datasets, objectives, and neuroscientific comparison data) are not (yet) available for Theory of Mind, raising uncertainty about whether similar advances can be achieved.