Also, functions in our heads are probably nonparametric. That is, we do not learn a set of parameters and then use it every time we encounter something. It doesn’t seem right to me. However, think about nonparametric functions. It’s like we learn some rules, heuristics and use it (and/or combine it with other functions) every time we encounter something. That makes more sense. For example, you might find this dumb but we can do classification with nearest neighbors up to certain accuracy for every data set :D. Again, maybe you do not find this interesting but think about it. We do no training at all but come up with this certain rule, certain knowledge that examples that are nearby should be similar. Instead of learning anything, why not store a sufficient amount of experiences and interpolate from them?
I don’t think we learn by trying to predict the world. We rather try to detect anomalies in the world. It is viable economically, and why would we want to predict the world anyway? We plan, yes, but this is different from predicting every single state of the world. I think we use the anomaly detector in a somewhat smart way to figure out what things can happen or extract the normal way from the anomaly detector. Anomaly detection may explain curiosity. I will think about that.
This morning, I stumbled upon a paper by Kenneth Stanley and his friends. He discusses the importance of having an open-ended environment for the development of artificial general intelligence. I couldn’t agree more. I think the idea is probably not new (though couldn’t give you the origination). Even Rick and Morty made a reference to this.
This is a generative adversarial soft decision tree. Red regions imply thinks its from , blue regions imply thinks its from .