Gary Marcus

The problem with #machinelearning in a nutshell, via extensional vs intensional logic: A: {2, 4, 6, 8, 10, ... X} B: {2x: x ∈ N} There is no way for a statistical, asymbolic machine to arrive at B from A, no matter how large you choose X @GaryMarcus @filippie509 #ai

This isn't a problem with machine learning, but rather with low bias approximators. They fit the data well and can interpolate but have no mathematical reason to extrapolate. A lot of work, including in deep learning, deals with yielding better extrapolation by introducing bias.

This. I'd like to add: bias which can often be learned, which humans did. I don't think that humans were born with hard-coded reasoning capabilities. We learned them from data (observations and interactions). When we humans deal with the problem above, we use years of experience.

you might want to read Bonatti’s recent Science paper or Gervain’s newborn work that follows up my infant rule work. or to think about what evolution does to brains over time. (see my book The Birth of The Mind.)

Ok, but how does that change the argument? Evolution is machine learning with a longer time span, e.g. an optimization process. It is still machine learning. The bias is learned by evolution.