Top AI researchers (Yoshua Bengio, Yann LeCun) are essentially cognitive scientists. By "cognitive science", I mean here general theories of cognition, not human cognition. If you watch any recent talk by Bengio (example), you recognise that it's a talk about cognitive science at least as much as it is about AI. From his talks, you could also roughly sense the type of problems these researchers are solving when they move to the level of thinking about cognitive science.

Theories of cognitive science and ML/DL form an "abstraction-grounding" stack:
general theories of cognition (intelligence, agency) ->
general theories of DNN working in runtime ->
interpretability theories for a concrete DNN architecture.