This is an excellent article. I ways, In already study this problem, but in the neural nets that make up our minds. Our neocortex is a neural net that, when trained, works well with cultural artifacts like language, logic and reason. There are other neural nets in the mind though, such as vision, that are far older than the neocortex and not so adept at cultural tools. The ones I focus on are the moral instincts and the older survival instincts. Neither is very good with language (but can be talked to). Neither tends to follow logic we can access any more than evolution, what created them, does. For that matter, even the insights from the neocortex, the flash of lightning, can be extremely difficult to convert to a "language" that can be retained in memory and communicated. (I wrote "When Barbara Explained Genius" to describe how to do that, but it needs re-writing... soon.)

One great question is if an AI solves a problem in a way humans don't, can we learn that method? ... I have my opinions.