Gary Marcus

There's a couple problems with this whole line of attack. 1) Saying it louder ≠ saying it first. You can't claim credit for differentiating between reasoning and pattern recognition. 2) Saying X doesn't solve Y is pretty easy. But where are your concrete solutions for Y?

But I did say this stuff first, in 2001, 2012 etc? Not about louder. And no, I don’t know how to solve the problems, but I have pointed to specific directions that are finally getting some air (explicit operations over variables, in particular) that for a long time were dismissed

There's nothing new added to this conversation on account of deep vs not deep. Basic questions about the limits of mining associations (vs reasoning) have been plumbed far earlier and far deeper by Rubin, Robins, @yudapearl, no?

The basic question IMHO is symbol-manipulation - do we need it or not? Two entirely different classes of problems. No real causal reasoning with out it, but people like Hinton and LeCun dismiss it, and even ridicule it (eg https://sites.google.com/site/krr2015/home/schedule …)

If only "symbol manipulation" meant anything precise in algorithmic, statistical, or information-theoretic terms (contrast causal inference, for which @yudapearl et al provided precise distinctions). As it is, "symbol manipulation" is the AI/cogsci analog of intelligent design.