4 ms·

Is AI even there yet? Or are we using the term very loosely here?

NameError 3y ago
I'm not sure there's a good way to answer this question, people seem to have very different definitions of AI. The definition I learned in school was "AI is when machines do things that most people would say requires intelligence to do", which is (maybe deliberately) vague and prone to shifting over time. See https://en.wikipedia.org/wiki/AI_effect https://en.wikipedia.org/wiki/AI_effect - the term "AI" used to include things like chess engines (and even simpler search algorithms), but once computers are able to do those things for a while and people get used to it, the term "AI" tends to shift and only refer to things that machines can't do yet. It's admittedly a bit of an unsatisfying answer, but whether AI is here is more a question of your definition of "intelligence" than anything else.
ultra_nick 3y ago
Yes, we've had chess engines and spellcheck for years.
phatfish 3y ago
LLMs are definitely AI within the scope of text or even "conversation" generation. Just as AlphaGo is a Go playing AI within that scope. Obviously an AI that can have a human like text conversation has broader use than the Go playing AI. I find it curious that in sci-fi the idea of being human-like was considered the hardest task for an AI, and the walking encyclopedia which is never wrong bit was the easy part. Seems we did the hard part first with LLMs...
- ben_w 3y ago
  We already had search engines for our walking encyclopaedias, which the chat bots can use if we tell them to. The hard part is making the physical androids walk without falling over, but even that got much easier when someone realised they could be trained in simulations and what they learned would transfer to the real world.
wharvle 3y ago
LLMs are like a 10% boost (optimistically, for workers who are good at using them) for some legitimate jobs. They’re like a 1,000% boost if your job is spamming, advertising for scams, managing astroturf campaigns, or spreading misinformation. So. Hooray. So for most jobs that aren’t net-negative for society, no, not really, if “there” is “useful enough to seriously shake things up”.

I'm not sure there's a good way to answer this question, people seem to have very different definitions of AI. The definition I learned in school was "AI is when machines do things that most people would say requires intelligence to do", which is (maybe deliberately) vague and prone to shifting over time. See https://en.wikipedia.org/wiki/AI_effect https://en.wikipedia.org/wiki/AI_effect - the term "AI" used to include things like chess engines (and even simpler search algorithms), but once computers are able to do those things for a while and people get used to it, the term "AI" tends to shift and only refer to things that machines can't do yet. It's admittedly a bit of an unsatisfying answer, but whether AI is here is more a question of your definition of "intelligence" than anything else.

ultra_nick 3y ago

Yes, we've had chess engines and spellcheck for years.

phatfish 3y ago

LLMs are definitely AI within the scope of text or even "conversation" generation. Just as AlphaGo is a Go playing AI within that scope. Obviously an AI that can have a human like text conversation has broader use than the Go playing AI. I find it curious that in sci-fi the idea of being human-like was considered the hardest task for an AI, and the walking encyclopedia which is never wrong bit was the easy part. Seems we did the hard part first with LLMs...

ben_w 3y ago
We already had search engines for our walking encyclopaedias, which the chat bots can use if we tell them to. The hard part is making the physical androids walk without falling over, but even that got much easier when someone realised they could be trained in simulations and what they learned would transfer to the real world.

wharvle 3y ago

LLMs are like a 10% boost (optimistically, for workers who are good at using them) for some legitimate jobs. They’re like a 1,000% boost if your job is spamming, advertising for scams, managing astroturf campaigns, or spreading misinformation. So. Hooray. So for most jobs that aren’t net-negative for society, no, not really, if “there” is “useful enough to seriously shake things up”.