I know that feedback was introduced at least back in 80’s by Teuvo Kohenon in Self-Organization and…

The difference however is previous methods did not traverse hierarchies consisting of many layers of abstraction. That is why this is entirely new although Hofstadter has some intuition how perhaps it might work.

Not Hofstadter’s strange loop is in the domain of infinite recursion. There is nothing about Deep Learning processes that assume infinite computation. Rather it will always be bounded finite computation.