The pseudoscience label is overused in the article and answers to comments here. Science would not ever move forward without exploring hypotheses, which are not science yet. And even what is the scientific body is not expected to be the truth forever. It's an approximation. Science uses gradient descent in a sense, but needs to be pushed out of local minimum from time to time.

We are yet to see scientific method to be adjusted to account for the consciousness phenomena. Until that time many things will remain pseudoscience. Science does not advance in algorithmic way. Logic is needed only when scientists check the knowledge.

As for "gradient-based function optimization" part of AGI solution - of course it is necessary part of AGI solution. Any AGI should be capable of it one way or another. Even simple things like Kalman Filter or least square estimation can be emulated by simple neural networks, let along more serious AGI solutions. The argument of function optimization being opposite is shallow. Hardly AI let alone AGI can operate as one way street. So your argument does not hold even on educated guess intuitive level.

I am not defending Mr LeCun. Just want to point out that in my opinion without a paradigm shift or two (related to consciousness phenomena, and especially introspection-based research) modern science plays whac-a-mole game, labeling opponents pseudoscientists instead of trying to think out of the box and make real progress.

PS. Beauty in Dall-E is in a way natural. Crystals and plants are perceived as beautiful because the observer (we) see the result of some function optimization, but do not know what function that is when we look at those. Plants may for instance optimize solar radiation intake, and the resulting leaf forms can be result of genetic constraints. We call that beauty. Dall-E pictures without us are just pixels.