It is sad that this model does not understand the brain's organ for human natural language. Easily over 90% of what people think and reason about is what they are thinking and reasoning about and can, because of this organ, communicate it bilaterally with others. This is typical of AI people who are ML/Vision people not ML/NLP people, or rather people without a truth bias one way or another.
That said, our models of intermodal brain computation in the 1980s were identical to the one put out here. They were wrongly based on conventional computer models. There were a few of us who believed in highly parallel discrimination/classification networks (Perceptron style) which over the years have now formed the basis for much of AI/ML/Vision/NLP success today.
What is new is that we now have observational evidence, through fMRI and such, of the massive processing that goes on in simple recognition and action tasks. Like within 10 seconds comprehending a simple sentence like "I feel good about that" the neural activity is distinct and uniformly localized in the intermodal brain across people.
The real problem we have is the same as the problem we had in the 1980s. The computation power to simulate what goes on is still way way beyond what we have been able to engineer. Physics, by the way, has been hampered the same way and much of what physics thinks they can compute still cannot be computed. (Think the weather..or for that matter, brain computation.)
Here is an article on the organ in the brain which is the human natural language organ that LeCun does not include in his boxology. But gives a huge box that is part of the more realistic models of reasoning and thinking based on the practical need for uniquely human communication. (As well as seeing and acting).
All that said LeCun is truly a good computational cognitive neuroscientist and every word out of his language communication is worth listening to, understanding, and having dialogues with. If only it was more complete.
https://medium.com/liecatcher/uniquely-human-7b14099b7462
He needs to understand why our society and AI has an Internet Court of Lies to handle the evil fundamental to how our brains think and reason.
We need Microsoft, Google, Facebook, Twitter, Amazon, and others to help with moving the AI fundamentally forward, not just reimagining what we have already imagined. Even though that is also incredibly important as our computing capability of the right kind has improved.
(I'm sorry, but when I read "objective function" I went nuclear. Idiots. We need "not-quite-objective functions")
