Zompist Bboard Again

malloc wrote: ↑Mon Dec 16, 2024 8:58 pm Given the state of humanity these days, with its headlong slide into reaction and superstition, there is genuinely a case for discarding the risibly named Homo sapiens in favor of something more rational and competent*.

You do realize that the supposed AIs of our time, if one assumes they have opinions, are often as reactionary and superstitious as human beings?

Apparently what that guy is selling isn't rationality and competence; he's selling obedience. Something people can boss around.

I welcome the clarification though; there's no excuse for reverence towards the tech industry now.

Raphael wrote: ↑Tue Dec 17, 2024 2:18 amYou do realize that the supposed AIs of our time, if one assumes they have opinions, are often as reactionary and superstitious as human beings?

Sure but a full-fledged AGI would presumably avoid such things since they would have no animal instincts clouding their judgment. Humans are superstitious and tribalistic because we inherited all kinds of psychological crap from our ape ancestors after all. The AGI would have no biological inheritance, only abstract programming.

malloc wrote: ↑Tue Dec 17, 2024 5:22 am

Raphael wrote: ↑Tue Dec 17, 2024 2:18 amYou do realize that the supposed AIs of our time, if one assumes they have opinions, are often as reactionary and superstitious as human beings?

Sure but a full-fledged AGI would presumably avoid such things since they would have no animal instincts clouding their judgment. Humans are superstitious and tribalistic because we inherited all kinds of psychological crap from our ape ancestors after all. The AGI would have no biological inheritance, only abstract programming.

What a "full-fledged" AI would or wouldn't be like doesn't tell us much about what AIs derived from today's AIs would or wouldn't be like. And, an AGI would have "biological inheritance", because its initial "abstract programming" would be done by humans, with all the "psychological crap" inherited from the apes. And, if the AGI is built in any way like today's AIs are, it would be based on enormous amounts of input created by humans.

There is no reason to believe that real-life AIs would be like stereotypical robots from 1950s sci-fi movies.

malloc wrote: ↑Tue Dec 17, 2024 5:22 am

Raphael wrote: ↑Tue Dec 17, 2024 2:18 amYou do realize that the supposed AIs of our time, if one assumes they have opinions, are often as reactionary and superstitious as human beings?

Sure but a full-fledged AGI would presumably avoid such things since they would have no animal instincts clouding their judgment.

Oh please. Asimov in the 40s may have gotten away with imagining robots as the embodiment of perfect logic, but by now we know better than that. Maybe they might not have the same instincts as humans do, but they will have some. So far, all attempts to build an AI out of perfect logic have failed: the most successful attempts so far have been those which allow AI to learn things by instinct.

malloc wrote: ↑Tue Dec 17, 2024 5:22 am

Raphael wrote: ↑Tue Dec 17, 2024 2:18 amYou do realize that the supposed AIs of our time, if one assumes they have opinions, are often as reactionary and superstitious as human beings?

Sure but a full-fledged AGI would presumably avoid such things since they would have no animal instincts clouding their judgment. Humans are superstitious and tribalistic because we inherited all kinds of psychological crap from our ape ancestors after all. The AGI would have no biological inheritance, only abstract programming.

You do realize that the actual AI's, as we call them, we have now are often quite racist and sexist, right? They take the (often unintended) biases in their training data and amplify them beyond whatever the intentions of their creators were. And why would AGI have to be any different? It would have to learn from data provided from somewhere, and thus would fall into the same trap.

Case in point -- it was found that some visual face recognition software had trouble recognizing Black people because it was primarily trained against images of White people.

But more advanced approaches to AI could fix all of those problems. The first chess engines were terrible but now they can defeat even the best human chess players effortlessly. Computers already have the advantage of lacking biological needs or instincts. They just need the right software to replicate our reasoning ability without our emotions or instincts getting in the way.

malloc wrote: ↑Tue Dec 17, 2024 11:44 am But more advanced approaches to AI could fix all of those problems. The first chess engines were terrible but now they can defeat even the best human chess players effortlessly. Computers already have the advantage of lacking biological needs or instincts. They just need the right software to replicate our reasoning ability without our emotions or instincts getting in the way.

The most advanced AI's we have today are not based on 'logic' in the stereotypical sense at all but rather on heuristics derived from training data. And there is no sign that this will change from more advanced AI's in the future. Even if we managed to 'close the loop' and achieve AGI does not mean that AGI will become any more 'rational' than humans per se at all.

Then why is the tech industry using this approach in the first place? Shouldn't they be pursuing one that will avoid the problems of human irrationality?

malloc wrote: ↑Tue Dec 17, 2024 3:04 pm Then why is the tech industry using this approach in the first place? Shouldn't they be pursuing one that will avoid the problems of human irrationality?

First, the tech industry to a large extent consists of herd followers who follow the latest fads, and their current main "AI" approach is the current fad. Second, what motivations would they have to "avoid the problems of human irrationality"?

malloc wrote: ↑Mon Dec 16, 2024 8:58 pm *I must confess that my support for humanity has started to waver in recent months. The reelection of Trump has really shaken my faith in humanity. Along with everything else going wrong with the world, it has left me open to the possibility that we are irredeemably stupid.

"We" is (are?) overgeneralising rather; it's more accurate to say "enough important voters in a broken system", which amounts to a few tens of thousands. But imagine that your worst nightmare comes to pass: what do you think all eight thousand million redundant humans are going to be doing with their Copious Free Time? What about, for one, the CTO of MegaAICorpCo, Inc? Will they be obsolete?

I must admit, though, to finding it depressing that a sizeable amount of YouTube's content consists of "AI"-generated videos which are only watched by bots in order to generate revenue for someone. This is one job I'd be happy to leave to machines.

malloc wrote: ↑Tue Dec 17, 2024 3:04 pm Then why is the tech industry using this approach in the first place? Shouldn't they be pursuing one that will avoid the problems of human irrationality?

Because, like I said, when they tried to make a perfectly rational AI, it didn’t work. Here’s the famous essay on the subject: http://www.incompleteideas.net/IncIdeas ... esson.html

alice wrote: ↑Tue Dec 17, 2024 3:18 pm"We" is (are?) overgeneralising rather; it's more accurate to say "enough important voters in a broken system", which amounts to a few tens of thousands. But imagine that your worst nightmare comes to pass: what do you think all eight thousand million redundant humans are going to be doing with their Copious Free Time? What about, for one, the CTO of MegaAICorpCo, Inc? Will they be obsolete?

They will slowly starve to death from unemployment. Judging by the current tide of reactionary politics, they will probably blame each other for the problem and try to genocide everyone else.

bradrn wrote: ↑Tue Dec 17, 2024 5:31 pmBecause, like I said, when they tried to make a perfectly rational AI, it didn’t work. Here’s the famous essay on the subject: http://www.incompleteideas.net/IncIdeas ... esson.html

That sounds more like not worrying about efficiency because computers are constantly getting more powerful. I am talking about something with all the cognitive abilities of humans that we would label intelligence without the irrationality. Such a machine would not fall prey to conspiracy theories or obsess over celebrities. It would make its decisions with the unerring precision and correctness that computers currently handle mathematics.

malloc wrote: ↑Tue Dec 17, 2024 7:33 pm I am talking about something with all the cognitive abilities of humans that we would label intelligence without the irrationality.

I believe that this is a contradiction in terms — nothing more than the fevered imagination of 20th-century SF writers. An artificial intelligence will probably have different irrationalities to those of humans, but I feel quite certain that they will be, on the whole, no more inherently logical than we are.

bradrn wrote: ↑Tue Dec 17, 2024 7:39 pm

malloc wrote: ↑Tue Dec 17, 2024 7:33 pm I am talking about something with all the cognitive abilities of humans that we would label intelligence without the irrationality.

I believe that this is a contradiction in terms — nothing more than the fevered imagination of 20th-century SF writers. An artificial intelligence will probably have different irrationalities to those of humans, but I feel quite certain that they will be, on the whole, no more inherently logical than we are.

Agreed. There is no reason why AI would really be any more rational than humans are; even if they aren't irrational the same way as humans are irrational does not mean that they will be more rational overall.

bradrn wrote: ↑Tue Dec 17, 2024 7:39 pmI believe that this is a contradiction in terms — nothing more than the fevered imagination of 20th-century SF writers. An artificial intelligence will probably have different irrationalities to those of humans, but I feel quite certain that they will be, on the whole, no more inherently logical than we are.

How so? Computers have mastered numerous human cognitive abilities without developing irrationality. Consider calculators and chess engines, both vastly superior to any human in their respective domain and utterly rational. You will never see a calculator claiming that two and two make five, unless Big Brother has hacked it anyway.

malloc wrote: ↑Tue Dec 17, 2024 7:54 pm

bradrn wrote: ↑Tue Dec 17, 2024 7:39 pmI believe that this is a contradiction in terms — nothing more than the fevered imagination of 20th-century SF writers. An artificial intelligence will probably have different irrationalities to those of humans, but I feel quite certain that they will be, on the whole, no more inherently logical than we are.

How so? Computers have mastered numerous human cognitive abilities without developing irrationality. Consider calculators and chess engines, both vastly superior to any human in their respective domain and utterly rational. You will never see a calculator claiming that two and two make five, unless Big Brother has hacked it anyway.

The thing, though, is that mere calculators are not capable of anything resembling "intelligence". Advanced AI's as we know them are based on extensive training and pattern-matching rather than on simple calculation. They actually have to be specially programmed to handle things like numbers because normal modern AI techniques are much better suited to things like natural language and image processing/generation than to calculation.

Travis B. wrote: ↑Tue Dec 17, 2024 8:40 pm They actually have to be specially programmed to handle things like numbers because normal modern AI techniques are much better suited to things like natural language and image processing/generation than to calculation.

I think this is the key — most of the time, any ‘logic’ we see in AIs is hard-coded into them. Chess engines don’t somehow grow into being perfectly logical minds; they have the laws of chess programmed into them from the start. LLMs which try to play chess usually make wrong moves all the time.

bradrn wrote: ↑Tue Dec 17, 2024 8:46 pm

Travis B. wrote: ↑Tue Dec 17, 2024 8:40 pm They actually have to be specially programmed to handle things like numbers because normal modern AI techniques are much better suited to things like natural language and image processing/generation than to calculation.

I think this is the key — most of the time, any ‘logic’ we see in AIs is hard-coded into them. Chess engines don’t somehow grow into being perfectly logical minds; they have the laws of chess programmed into them from the start. LLMs which try to play chess usually make wrong moves all the time.

The opposite is also true, as mentioned -- trying to create an AI based on 'logic' doesn't get you far, and modern advanced AI's really only came about when people decided to abandon 'logic' as the basis of AI.