ChatGPT a false promise?

in agi •  2 years ago 


image.png

https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html

Been thinking a lot about AI lately. This Noam Chomsky article is interesting. I have a great deal of respect for the man, but I think he misses the mark pretty badly.

In particular, he uses the example of how a child, with no actual instruction, intuitively develops and understanding of grammatical rules. And contrasts that with ChatGPT being utterly different and comically shallow.

Yet from what I understand of deep neural nets, especially ones like AlphaGO, the child learning from nothing model seems very, very similar to what the AIs are doing. Being exposed to a broad set of examples, it infers what things are right in a given situation, but can't readily explain how it knows.

I read a metaphor recently that rings true. The discussions about whether AIs are now or will soon be "thinking" is very similar to discussions about whether a submarine can swim. The answer depends on how you define swimming, but the whole question is mostly a rhetorical distraction. The real issue is if it can accomplish the things that we do through swimming, and the answer for submarines is yes, and much better/longer/deeper/faster. Whether it's swimming or not isn't really the relevant.

Likewise, I literally don't care if the AIs are "thinking". I care if they're coming up with answer (and questions!) that are insightful and clever, and this long-term skeptic is quickly being won over to the yes side.

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!