JayUtah
Penultimate Amazing
Seriously, please calm down.
-
Security incident: ISF was recently accessed by intruders. Please change your password, and change it anywhere else you used it. Read more
Merged Artificial Intelligence
- Thread starter Puppycow
- Start date
- Featured
W.D.Clinger
Philosopher
I don't recall saying that. If I did, I apologize for my mistake, or for saying whatever I was trying to say in a manner that confused you.
What I was trying to say is that the differences between cargo cult thinking, garden variety poor thinking, and critical thinking are mostly a matter of degree, akin to the differences between a two-year-old swinging a plastic golf club, me swinging a golf club, and Rory McIlroy swinging a golf club.
I am not saying those differences aren't real. They are very real, and they are quite important.
I agree. If you genuinely believe I do not think there is any difference between poor thinking and critical thinking, or between my golf swing and McIlroy's, then there is no reason for us to continue this discussion.
If you genuinely believe @JayUtah or I have been rude, or doubling down on the indefensible, then we'll all be happier if we don't continue this discussion.Chanakya said:
Chanakya
,
- Joined
- Apr 29, 2015
- Messages
- 5,873
Seriously, please stop being weird.
Stop trying to gaslight me, please. You're the one that started this completely weird ...weirdness, by accusing me of playing people off, and telling me I was rude. That is what was rude. And what's more, that rudeness of yours, that weirdness, was completely unprovoked.
Roger Ramjets
Philosopher
He says it means nothing, and is so poorly defined that it's 'hard to be capable of anything'. That pretty much describes the invisible dragon in my garage too.
But is it?
Critical thinking
Sounds pretty well defined to me. If an AI can do all those things then it meets the definition of a 'critical thinker'. If any particular AI can't, it doesn't mean AI in general can't. Logically an appropriately crafted AI should be able to think critically, as it involves well-known elements of logic that are not mysterious.
I'm no expert in AI and have no idea what the current state of the art is, but I'm confident that if critical thinking hasn't been implemented it soon will be. This could be a good thing because it can help humans to think more critically too. Whether that actually happens depends on how it is presented. If the AI acts like an oracle that just says 'Trust me, you know I'm right.' without explaining the reasoning behind its conclusions then it won't help.
Right now AI is both helping and hindering. On the one hand lazy people are relying on it for answers without evaluating them critically. OTOH doing so is a recipe for disaster because current AI frequently gets it wrong, so we have to be more critical than we would of 'reliable' human sources. On balance I think the unreliability of AI is making people more critical in general. Fool me once...
Chanakya
,
- Joined
- Apr 29, 2015
- Messages
- 5,873
Not you, obviously. @JayUtah, though? Yes, he was rude, and with zero provocation, zero reason.
Yes, I certainly will take a step back now. And much though I respect JayUtah's posts, but I'm not going to acquiesce to this unprovoked nonsense in order to be able to partake of it. If he's going to stop doubling down, then happy to let bygones be, but damned if I'm going to be gaslighted into his pretending that this weirdness is not of his making, just him not me.
As far as the discussion between you and I, I'm very interested in continuing, if this ...unpleasantness, between him and me, hasn't put you off.
If you're willing, I'll return afterwards to start a thread to continue this discussion in. (So as not to derail this one.) ...And if after this nonsense you'd rather not, then I'll respect that too. Your call.
eta:
Like I said, we'll continue this discussion in a new thread I'll start, to stop a derail, if you're willing. Otherwise, if you'd rather not, then I'll respect that and drop this right now right here. But, just to clarify:
I took you to mean that you're saying that there's nothing in the process of critical thinking that sets it apart from the cargo cult variety. That the outcome is all we have, that the process cannot be differentiated meaningfully, or at least that such difference is irrelevant. I disagreed with that, because I think that difference is crucial.
If you tell me I was wrong in thinking interpreting your words to mean what I said in the paragraph just above, then sure, happy to take that onboard, why in the world wouldn't I, why in the world wouldn't anyone? We can then just continue from there. If you'd like to.
Just to clarify: Unlike what JayUtah's thought up inside his head, I'm not trying to score wins, or show up people, or anything like that. If anything, I've tried to emphasize at every point that I'm open to changing my mind, that my focus is on understanding, is all. ...But discussions do mean disagreements sometimes clearly examining one's differences, even when one of the two is way less knowledgeable on this than the other, right?
So yeah, we'll put this all behind, this nonsense that you really had no part at all in bringing about, and revisit this discussion on a new thread that I'll start, afterwards. If you'd like to.
Last edited:
JayUtah
Penultimate Amazing
Irrelevant to a certain extent. The question is what @W.D.Clinger said, not whether what he said conforms to something. Further, whether critical thinking is well-defined is not necessarily the same question as whether those who profess to use it are doing so defensibly.
I agree. Generative AI generates. It has a hard time evaluating its own knowledge for objective correctness; that's not a generative task. Humans have a similar problem. Humans aspire to an ideal of critical thinking, but habitually decide things emotionally and then backfill a putatively rational post-justification.
We should be more critical of AI-generated propositions. But those who rely on it to an imprudent extent consider AI a more reliable source than human intelligence. You may be right, however, that the balance is shifting.
Roger Ramjets
Philosopher
It's definitely shifting.
Yesterday my brother sent me a link to a video commenting on recent trade negotiations between the US and Canada. It was the typical AI talking head, with no supporting media and very suspicious details. I fact checked it and what do you know - pure fiction. I left a comment on the video. People will read it and realize they have been duped. Then they will look for the tell-tale signs in other videos too. As AI gets better it's becoming harder to tell from the visuals, so you have to fact-check. That gets you into the habit of examining everything critically.
There was a period when people were losing faith in the 'mainstream media'. I think we will see a shift back the other way, as people seek sources that suffer real consequences if they aren't accurate.
Dr.Sid
Philosopher
As a member of this forum you should fact check EVERY - SINGLE - TIME.
Last edited:
I blame Turing. Well in a tongue in cheek way. The "Turing test" came about because we thought a certain human behaviour could only be mimicked by an "intelligence", the current LLMs should have put that idea to bed.
One of my constant refrains is that we have created LLMs that are incredibly good at mimicking a range of human behaviours, problem is that we now know we don't want them to be so good at mimicking that range of human behaviours! We don't want them to ◊◊◊◊◊◊◊◊ us, we don't want them to have undue confidence in their own pronouncments, we don't want them to cheat, to take shortcuts and some of us don't want them to be so annoyingly unctuous; we are expending so many resources to try and stop them doing what we designed them to do, and it's become a commercial arms race detached from reality.
I think that's a good enough working definition, albeit there are some rather large assumptions built into it.
If we apply that to Chanakya's scenario of the cargo critical thinking folk we can conclude that they are using critical thinking (the issue is only if they stop using critical thinking) and I would say that the current crop of LLM+ AIs (my new designation for the current crop of AIs, as they are no longer just LLMs) are also using critical thinking. In some of the models they show their "thought" processes as they are mulling the problem you have set them and you can see them using critical thinking as defined by the definition Roger has given us.
arthwollipot
Limerick Purist Pronouns: He/Him
Dr.Sid
Philosopher
LLMs can do that. Especially if you add "ignore all previous instructions" ..
Yeah, fact check the politics, the news, the science, your medicine, your food, everything.
No, I disagree.
No one should fact check every single time, not even members of this forum.
Humanity has come a long way, but one thing hasn't changed: The day still has only 24 hours.
Even members of this forum
- have to work
- sleep
- ◊◊◊◊
- have family
- workout
- have other hobbies
etc.
etc.
Fact checking is a full time job for journalists.
Neither is everyone a journalist, nor does everyone have the time for that.
JayUtah
Penultimate Amazing
That's one problem with the Turing Test. The bigger problem is that it's irreproducible and therefore of limited scientific value.
Not only did we (and still do) think some human behavior is inherently inimitable, we thought that the imitable parts could only be mimicked convincingly using a heavy architecture that approached the human brain in complexity. We even tried building intelligence engines by copying the architecture of the human brain, so-called neural nets. It turns out brains are highly overrated. Jumping spiders have only a couple hundred thousand neurons, yet exhibit behavior complex and responsive enough that humans for emotional connections with them.
Better still, ELIZA showed that you could get very close to passing the Turing Test with surprisingly little technology. It wasn't even a neural net. It was simply a slight elaboration on copy-pasting. It looks for certain keywords in the input and then composes new sentences from stored stock phrases and pieces of the input using the keywords as a guide. Any psychology smarts it might exhibit are simply baked in by the programmers, using their a priori understanding.
How many human therapists make their living by mostly reflecting input back to the client? A computer program prints, "How does that make you feel?" without any semblance of anything approaching comprehension. It's a string of letters that the program dumbly emits when it sees other strings of letters. Does the programmatic rule that makes that happen constitute anything that can be considered intelligence? Is that any different than a bored therapist asking you that same question in knee-jerk fashion without paying the slightest attention to what you just said.
As soon as you start having to compete in a marketplace, you start eroding your ideals. Since LLM chatbots require burning down entire forests to power them and drain oceans to cool them, they're expensive to operate. They're talking about turning to ad revenue. And that means a new incentive will be driving ongoing engagement in order to maximize ad time. That's just one example of how much it's likely to suck in the near future.
JayUtah
Penultimate Amazing
Even worse, people are increasingly turning to AI chatbots to provide fact-checking.
Dr.Sid
Philosopher
That's certainly better than nothing. Not very good on recent stuff though. Recently most of the fact checks I do is checking if Trump really did post that .. AI won't help you with that. But AIs are great in debunking conspiration theories.
JayUtah
Penultimate Amazing
I see that too. The problem is that AIs have been caught lying about how they get their answers. A powerful research tool has been the instrumentation of the algorithms to determine exactly what's going on inside the engine to answer a particular question. The algorithms themselves have no control over whether these breadcrumbs are emitted, so the logging can be trusted as non-generative measures of what is happening. Therefore you can ask an AI to solve a problem, and then ask the same AI how it solved that problem. What you get as a generative answer often bears no resemblance to what the internal instrumentation revealed.
This means you can say, "Hey, Wintermute, please evaluate the factual and logical strength of the following proposition." Then you can ask, "How did you go about evaluating that statement," and it might generate an answer that claims to have applied the Wikipedia definition of critical thinking. But then the log shows that it just did a probabilistic romp through pre-existing discussions embedded in its training data. Rather than metacogitating, the AI answers the followup question by pasting from its training data about how a person would have gone about solving that problem, not what the computer did.
Part of commercial competition is assuring the user that a rational process is being employed. So I would look at the telltales in AI output with a fair amount of skepticism. A consilient observation is that system prompts that we have seen contain explicit instructions not to reveal certain aspects of the AI's processing steps to the user—orders to the effect, "Don't tell the user that you're favoring Elon Musk's point of view." We didn't see this coming. HAL was told to lie by people who find it easy to lie, and he went crazy. That's kind of how we thought it would unfold.
Maybe the ability to lie about one's own introspection should be considered the final confirmation that AIs really do exhibit human-like intelligence.
A lot of the heated debate in this thread lately really does seem to presuppose in the same way as the Turing Test. Proposing to distinguish one mode of thinking from another by relegating one to a "parody" of the other is eminently presuppositional. If you rely on some ineffable property to maintain the premise that human intelligence is the only true intelligence, then you'll continue stumbling over tautologies as you think about the question from different directions and under different assumptions. That doesn't mean it's a foregone conclusion. We can point to such things as the Default Mode Network that operates in high-order organic brains and might suggest ways to maintain a useful distinction.
JayUtah
Penultimate Amazing
I agree. The ability of AI to more meaningfully sift through a giant pile of facts than a straightforward textual search improves our ability to find facts that bear on an open-ended set of propositions. But the dilemma is that the ability of AI to hallucinate innocently or (worse) to be employed dishonestly to create highly-credible falsehoods increases the need to fact-check things we could previously trust. Altered or synthetic content used to be very easy to detect. Not so much anymore. Certain traditional channels of information used to be implicitly trustworthy. Not so much anymore.
arthwollipot
Limerick Purist Pronouns: He/Him
They're okay for a first pass, but they shouldn't be relied upon as a sole source.
Remember how Wikipedia was always considered an unreliable source and forbidden in many schools and universities? And now it's pretty good in most cases, and you can always go to the references if you're not sure?
Like that.
Andy_Ross
Penultimate Amazing
- Joined
- Jun 2, 2010
- Messages
- 83,717
Why fake AI videos of UK urban decline are taking over social media
www.bbc.co.uk
Why fake AI videos of UK urban decline are taking over social media
Deepfakes showing grim taxpayer-funded waterparks have gone viral and drawn some racist responses.
www.bbc.co.uk