I agree that statistical ML is only good for interpolation, but I also feel that the set of human behaviors that are reducible to interpolation is larger than w

I agree that statistical ML is only good for interpolation, but I also feel that the set of human behaviors that are reducible to interpolation is larger than we realize. Even high order human behaviors like academic research or art criticism contain huge chunks of pattern recognition.