this post was submitted on 10 Aug 2026
14 points (81.8% liked)
Fuck AI
7560 readers
840 users here now
"We did it, Patrick! We made a technological breakthrough!"
A place for all those who loathe AI to discuss things, post articles, and ridicule the AI hype. Proud supporter of working people. And proud booer of SXSW 2024.
AI, in this case, refers to LLMs, GPT technology, and anything listed as "AI" meant to increase market valuations.
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
The Turing machine, and the Turing test are two very different things. They've been both coined by Alan Turing. But a Turing machine doesn't have anything to do with AI. It's a computer that can do any possible maths calculation. But it requires infinite memory. And since nothing is infinite in the real world, it doesn't exist. (Einstein might disagree.)
The Turing test, or "imitation game" is a hypothetical experiment setup. At least it was hypothetical back in the day. It tells us whether some machine can imitate human dialogue well enough to fool someone into thinking they're talking to a human.
The "test" is so vague we had machines pass I many years before the current LLM AIs. But it is also always limited by the scope of the AI. Can you play a game of chess against a chessbot, where after the game you are unable to tell if it was a bot or a human? Can you chat with an LLM and think it is a human? Can you chat with a chessbot and think it is human. Can you have a face to face dinner at a restaurant with an LLM and think it is a human? Can you have a LLM drive a car and think it is human?
AI are able to fool us in their specific area of expertise but so far never in all areas at once.
I think so as well. It comes with some kind of implication on how the result of the imitation game is supposed to mean something. But it doesn't. There's no good/scientific way to see how something is able to fool a human observer and then conclude that thing is something... Or has specific attributes... I mean other than it's able to fool a human in this specific experiment setup. There's just a huge gap in reasoning after the experiment. I think it's mostly what it is, a spin on a party game.
And yes. So far all we've seen is narrow AI. That's the term for an artificial thing which can do some specific tasks it's been trained for. You can never teach ChatGPT to drive a car properly. You'd have to design a new AI for that. And that one will probably not be able to load the dishwasher... In fact I've never seen any robot be able to do simple things like load a dishwasher. They regularly tip over while doing it or put in just 3 plates as a demo... While I as a human have a broader bandwidth of things I'm capable of. And I can transfer my knowledge to entirely new things and/or successfully learn about them. While current AI is very limited in doing that.
I think there might be some prospect of AI becoming more universal. Maybe one day we'll see general (/strong) AI. But as of today that's utter science fiction. Nobody has a clue how to make AI learn all by itself about new concepts, without feeding it in in the factory. They'll have to figure out how to make it learn while running, and not take it down and release a new ChatGPT every half a year. Adapt and change, remember things... That's all huge, unsolved engineering problems. I suspect to become properly useful for a lot of real-world tasks we gotta merge robotics and AI some more... But maybe that's not strictly necessary, depending on how we define the concept. But it's hard to come up with an experiment setup to test for all the things humans can possibly do. Or choose the correct subset of tasks to make the comparison mean something. But I'm fairly positive it's going to take a decade or more before we're anywhere close. If that's even possible. It's a bit difficult to judge since we don't know how to design strong AI on a human level, what compute power it'd need etc.
Till then, we're stuck with chatbots which can mimick text and learn some from ingesting it, maybe oneday self-driving cars, chessbots, smart thermostats... I mean these things can solve tasks. (Some better than others.) But what does that mean? Probably not a lot, if you're looking at a smart thermostat and how that's able to get the temperature just right, but you expect it to somehow turn into Commander Data from Star Trek.
To add a bit to this: I don't see a reason to stay "skeptical" about this. It's an experiment. And it's not even hard to set it up. You can just try and learn the answer to it.
It's a bit tricky since some details are vague and there's some variations of it. It's more a party game than written down with mathematical precision, like other texts from Turing.
What I'd be sceptical of, is the implications of it. And claims from big companies who have a vested interest in promoting their products. It's not really influential science until it's been replicated by other (independant) researchers and been peer-reviewed.
I think you mean "Alan", friend
thx! corrected.