Gary Marcus

@GaryMarcus I would like in fact to hear a solid and testable proposal based on your 20 years of insisting how ML is not on the right track (as you said in an earlier tweet in this week), instead of another essay.https://twitter.com/GaryMarcus/status/1200613819518599169 …

The solid proposal is that we need to search a different part of model space, as described in detail in 2001, where I made several specifics proposals about important challenges that the ML community simply has refused (for the most part) to engage with. I can't do it on my own.

@GaryMarcus "search a different part of model space" is not a testable proposal, nor your whole book from 2001 is. I wanted to hear what that different part is, in equations or pseudo code.

ps 2001 book proposed many specific problems ranging from generalizing low frequency default morphology to acquiring universally quantified one to one mappings to tracking identity and properties of individuals over time. community really has not engaged in any of these.

This is a red herring. Problem proposals are not solution proposals.

Problem proposals are a key part of the scientific method though: through them we present falsifiable hypotheses in order to try to verify them. Lots of work goes into the precise definition of a hypothesis, which is perhaps where a little more detail is needed in Gary's problems

Yes of course. I'm more concerned with the subtle switch of argument, though. What starts as a request for a model description (*any* model description, at this point) quickly becomes a debate about whether others are focusing on the right problem