Gary Marcus

Do you really think you are so unique or outstanding in calling for "new models that incorporate symbol-manipulation and innateness" & lots of other ideas, too, or is it just that you pitch this as a critique of DL?

i have been doing it for 25 years, against enormous resistance, some of which has been on display today.

When a thing is overhyped, pointing out that fact may well hasten its fall. But that hastened fall is probably a good thing.

What exactly would be "falling"? Machine Learning is here to stay (foreseeably) in areas where it has had success (NLP, vision, etc.). I don't see Gary's approach of pointing out the limitations and overblown rhetoric as being remotely threatening to the field of AI as a whole.

I think Gary Marcus is doing a great service to the field of AI. There is more to AI than detecting statistical regularities in carefully labeled data. Plus accuracy olympics totally ignore robustness which is very worrying if you are in a self driving car.

We need to “evangelize the limitations” or we will have a new AI winter!! We should promote the strengths but acknowledge the weaknesses!!!!

I insist Dr. Marcus. Your paper "Deep Learning: A Critical Appraisal" is on its way to become a cult classic like the original "Marvin Minsky & Seymour Papert (1969). Perceptrons"... I have no doubt about this! Great service to the field...