Gary Marcus

advocates of #machinelearning, I am told that you all know that (current) #ML is limited. fair enough. but which limits are you willing to *publicly* acknowledge?https://twitter.com/NotSimplicio/status/1173373706674085888 …

Watch anyone of my talks of the last 4 years. Many even have the title "the power and limits of deep learning" More obvious answer: limitations are pretty much everything everyone in the field is doing research on: self-sup learning, model-based control, reasoning, causality...

agree with @ylecun on much of this list, and tried (among other things) to explain many of them (sometimes with slightly different terms) to the general public in http://rebooting.ai , because few seem to be widely understood or appreciated.

@yudapearl has of course done a great job bringing causality to the forefront of public attention. but overall i think there is a mistaken impression that all we need is more data and more compute, when in fact bolder steps are required.

There is a subtle but important difference between a) all we need is more data and more compute (which I don't know if any expert believes) and b) all we need are general methods like search and learning that leverage data and compute (which is what Richard Sutton argues).

There are theoretical impediments that even "general methods" cannot circumvent. If you search for a common point of two parallel lines, you can have the best search-and-learning method in the world and you won't find one. ML confronts such impediments to reaching GAI. #Bookofway

That's the whole debate, isn't it? Is the human ability to learn things like mathematics, geometry, formal logic, informal reasoning, and science the result of some ensemble of general search and/or learning processes operating in the brain, or something else?