Gary Marcus

I have already reviewed the limits that i see, in http://rebooting.ai  w Ernie Davis. this thread isn’t about what the limits are, it’s about what limits practitioners acknowledge publicly. common attempt at defense is to allege limits are widely acknowledged. are they?

What I’m saying is that I suspect you have the reach to just publicly ask top people about some set of limits you consider representative. Then you’ll have at least a decent empirical representation of what real contributors to the discipline acknowledge about those limits.

IMHO, most of people with relevant opinions are well aware of Gary and his arguments. He is talking past some of them, because he is understood as saying "neural networks cannot do symbolic computation", which of course is not true.

If he'd ask for eg. more work at understanding the intersection between grammatical operations and perception and go into details, it might be more useful to the field and public debate.

but it still off the point of this thread, which is to assess how forthcoming advocates of ML are about limits. and how much they invite the inference that there are no limits, even if they know better.

I’d like to know more about what standard you are using to decide who exactly “advocates of ML” and what they collectively think. The vague sense that at least someone from a large vague set of people think X is hard to falsify.

self-identification & prominence (eg publications in the field). a prominent researcher recently told me that everyone in the field already in the field acknowledged all the limits that I point out; I doubt that they have done so publicly, & I am curious to see what happens here.

But it’s hard to identify what would change your mind. By criticizing the views of a general community without identifying more clearly who and what, you can just keep saying this forever. There will always be *someone* with an overly hype soundbite you can point to.