[Animats]

This is fascinating. And important. We need better ways to see what neural nets are doing. At least for visual processing, we now have some.

This might be usable on music. Train a net to recognize a type of music, then run it backwards to see what comes out.

Run on the neural nets that do face popout (face/non face, not face recognition), some generic face should emerge. Run on nets for text recognition, letter forms should emerge. Run on financial data vs results ... optimal business strategies?

But calling it "inceptionism" is silly. (Could be worse, as with "polyfill", though.)

[userbinator]

Run on the neural nets that do face popout (face/non face, not face recognition), some generic face should emerge

https://en.wikipedia.org/wiki/Eigenface