ps see reply to @tdietterich for operational definition of understanding.
Show this thread
Hinton’s “favorite Gary Marcus quote” argues that pure #deeplearning fails to understand language. I stand it by it: no current #AI system can understand a stories, articles, etc.
If you disagree w the quote, I’d like to hear your thoughts why.https://twitter.com/rsalakhu/status/1187518370276614145 …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
The issue is that "understanding" is a sliding scale. Obviously today's AI systems (which are virtually all hybrids) understand some things and not others. There is plenty of room for improvement!
-
operationally: answer questions about who did what where when why and how. what motivations were, what agents intended to achieve, what obstacles they faced, what sequences of events was etc. stuff we expect of any normal reader. discussion: http://rebooting.ai chapter 4
- 3 more replies
New conversation -
-
-
First, choose a story. Second, agree on what test you’d apply to a human to decide whether that human had understood the story after reading it. Third, apply the same test to an AI system. Second step essential so that there is agreement re what counts as story understanding.
-
@heuristicity and i proposed soem specific methods in AI magazine: https://nyuscholars.nyu.edu/en/publications/toward-a-comprehension-challenge-using-crowdsourcing-as-a-tool … (sorry it is paywalled)
End of conversation
New conversation -
-
-
I’m interested in hearing why symbol manipulation is better suited for the task ?
-
better representation of abstraction, freeer generalization, more reliable inference. see Doug Lenat’s Forbes article for best-case scenario, inferences re Romeo & Juliet that are far outside scope of what deep learning QA can do.https://www.forbes.com/sites/cognitiveworld/2019/07/03/what-ai-can-learn-from-romeo--juliet/ …
End of conversation
New conversation -
-
-
"each sentence has its own unique meaning" is both too fine-grained (some sentences mean the same as each other) and too coarse-grained (because of ambiguity and context). Some careful version of this claim might be true, but as written it is either vague, wrong or both.
-
it could indeed have been sharper, it was a brief quote drawn out of context from a brief popular essay. the whole of chapter 4 of http://rebooting.AI makes elaborates what we mean by understanding, and central chapters of algebraic mind spell out the prediction’s logic.
End of conversation
New conversation -
-
-
I would like to hear a good definition of 'understand'. All we have now is a vague 'you know it when you see it' sense.
-
see reply to
@tdietterich
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.