The speaker expects a system trained to match human input-output behavior to converge toward a similar causal structure, but says this may fail when a computation runs well on a brain but poorly on a GPU, or the reverse.

“But ultimately, it's going to converge to a structure that is similar to yours, to a similar causal structure. But there might be conditions when that is not the case. For instance, maybe certain things run well on a brain, but they don't run well on a GPU or vice versa.”