AIs gunning for our precious freelancers

[malloc]

I know a lot actually, because I have taken the time to find people who are into crafts and art. I own several handmade items and I've kept them far longer than the closest equivalents that were factory made, and I actively try to avoid "fast fashion". I've mentioned my concerns about AI previously, as a writer myself and someone for whom most of the people I knew growing up and interact with now have been creatives of some stripe.

Then you're very much in the minority. I can hardly think of anything handmade that I own apart from a crocheted blanket my mother made and even finding handmade goods around here would be difficult. All the stores around here sell mass-produced products made half a world away. When humanity goes from making 100% of its culture to only making 10% with the rest made by inhuman machines, that strikes me as quite the loss.

And I'm not saying things aren't going to change in the future because of AI development, but I'm saying it's a matter of tool use and not the inherent evils of neural networks. They aren't even as "intelligent" as the tech sales people want to claim and people saying LLMs are anywhere near sentient are anthropomorphizing them to a degree that is dangerous for themselves.

As it currently stands, we barely understand these tools or how they work. Nobody can explain many of the decisions made by the current wave of AIs, which makes me question whether we are really using tools or supplicating alien beings. Relying on chatGPT to write novels for you really is quite different than using a printing press or even a word processor to make copies of something you composed yourself. The next wave of AI will gain even greater intelligence and independence from human control.

[Richard W]

Furthermore, the fact that art sites need explicit policies to keep AI generated images from taking over merely proves my point. If humans could easily compete with AI, no such regulation would be necessary.

I think the real reason for such rules is the intellectual properties implications of AI generated images (or code or whatnot). If an AI is trained on much of the art available on the Internet, who is to say that the art produced by said AI is not a "derived work" of all that art? There have already been lawsuits about this very matter.

Aren't there similar problems with human artwork? I think tunes are a particular problem.

[Travis B.]

Furthermore, the fact that art sites need explicit policies to keep AI generated images from taking over merely proves my point. If humans could easily compete with AI, no such regulation would be necessary.

I think the real reason for such rules is the intellectual properties implications of AI generated images (or code or whatnot). If an AI is trained on much of the art available on the Internet, who is to say that the art produced by said AI is not a "derived work" of all that art? There have already been lawsuits about this very matter.

Aren't there similar problems with human artwork? I think tunes are a particular problem.

Yes ─ there have been quite a few lawsuits related to sampling, where someone has been sued for the apparent similarity of a short segment of sound in a song that seems to be taken from someone else's song (whether it actually was or not).

[Travis B.]

Furthermore, the fact that art sites need explicit policies to keep AI generated images from taking over merely proves my point. If humans could easily compete with AI, no such regulation would be necessary.

I think the real reason for such rules is the intellectual properties implications of AI generated images (or code or whatnot). If an AI is trained on much of the art available on the Internet, who is to say that the art produced by said AI is not a "derived work" of all that art? There have already been lawsuits about this very matter.

I may have mentioned before, I would argue the AI itself is a derivative work (if the AI is trained on these pieces, it is derived from them in no uncertain terms).

Of course, the problem with this logic is that if it were followed, anyone who ever entered an art museum and later made their own painting could be regarded as making a derivative work of all the art they ever had seen. Which is ridiculous, but is the natural consequence of such logic.

[zompist]

I may have mentioned before, I would argue the AI itself is a derivative work (if the AI is trained on these pieces, it is derived from them in no uncertain terms).

Of course, the problem with this logic is that if it were followed, anyone who ever entered an art museum and later made their own painting could be regarded as making a derivative work of all the art they ever had seen. Which is ridiculous, but is the natural consequence of such logic.

And the problem with that logic is that it's no logic at all. AIs can't commit plagiarism because they're like humans, who.. commit plagiarism.

[xxx]

the problem with ia is their use in place of humans,

particularly in decision-making,
when an autonomous car has to choose how to limit damage in the event of an accident,
or how to choose a treatment or surgical procedure...

and then in work organization, given the lower cost of an algorithm...

for the rest, it's just another step towards the dematerialization of uses,
and in art in particular, it only produces simulations that it's up to the artist to fill with meaning...
we can do without an artist, but we'll need a gallery owner capable of assembling these simulations
and giving them meaning, or at worst inviting the viewer to look for one...
In the latter case, we realize that we're being deceived...

and the risk is a totally standardized society, even one controlled by the ia...
the very opposite of what a true human being is looking for:
confrontation with reality, choice of action, total autonomy...
freedom...

[zompist]

These concerns are nothing new. 200 years ago, craftspeople were worried about factories. Later, painters were worried about photography, office clerks were worried about typewriters and calculators, and musicians were worried about grammophones. And so on. And we still have craftspeople, office clerks and musicians. They have lost some jobs, but they are still around and have their place in our society.

This is the same argument as saying "War is nothing new. 200 years ago people were worried about cannons. Later people were worried about machine guns, and even later, airplanes. And so on. Nuclear weapons are no problem, nuclear war won't be anything we can't handle."

Or "Human-caused environmental problems are nothing new. We've had mass extinctions, desalinization, deforestation, devastation of entire biomes and fishing stocks. Yet there are more humans than ever! Burning off our ecosphere with ever-increasing carbon emissions is no different."

I'm not saying that AI is an existential risk. But a facile dismissal of possible risks, and historical problems with automation, doesn't help. Technological changes can get bigger and bigger, and "we always handled it before" doesn't mean we always will.

Automation, and other forms of increased productivity, are disruptive-- and not infrequently produced famine and revolt. It's hard to read about the Highland Clearances and shrug them off as a benign improvement. It was great for the landlords, not so great for the people who died, lost their livelihood, or were forced to emigrate.

Automation can be benign in the long run, if new and better jobs are created at the same time. It's not automatic that this happens, and in the US, the conditions for it happening have been systematically undermined. Jobs are replaced or improved only if there's a social safety net, good public education, unions, restrictions on how much the wealthy can keep productivity gains for themselves.

I'd also point out that AI is not really like many of these earlier innovations, not because robots are going to destroy us, but because executives are going to destroy entire industries based on misplaced hype. There are already CEOS who think they can fire their entire customer service department, or all their writers or actors, or all their reporters. This isn't even productivity improvement, it's disruption with almost no positive gain. But destroying an industry doesn't necessarily mean that it can be easily reconstituted once people realize it was a mistake. (The news industry, for instance, has been decimated (not just by AI, it's a long process), and it's not likely to be coming back in the useful form it once had.)

[Ares Land]

My own concerns are mostly along the lines zompist describes. Automation is disruptive; we'll be going through Schumpeter's creative destruction all over again... and Western society still isn't very good at handling that sort of thing.

As for the hype thing, that is true but I'm not sure it's unprecedented. In fact it has all the markings of yet another bubble. (There are useful applications beyond all the hype, but the sums invested are insanely high, in light of the possible use cases.)

[zompist]

As for the hype thing, that is true but I'm not sure it's unprecedented. In fact it has all the markings of yet another bubble. (There are useful applications beyond all the hype, but the sums invested are insanely high, in light of the possible use cases.)

Several people have noted that AI hype began just at the moment when blockchain hype crashed. I do think there's more to LLMs than there was to blockchain (still a solution in search of a problem), but the timing does seem suspicious. Not the first time one bubble led directly to another.

[WeepingElf]

I wasn't meaning to say that AI wasn't dangerous; I just wanted to put the worries about it in perspective by reminding that such worries have been there in the past. Of course, AI is a major challenge to our society, as previous innovations have been. Yet, so far we have managed to cope with those challenges, which is of course no guarantee that we can cope this time again. And of course, the people concerned could not carry on as if nothing had happened; they had to adapt. Craftspeople, for instance, lost the mass market of new products; they had to specialize in what the factories couldn't do: repairs, and custom manufacturing. For musicians, recording became a major source of income. Office clerks had to learn to work with typewriters and calculators, and later with computers. The art of painting moved to things photography could not do - and photography became an art in itself, as the composition of an expressive image was still the work of the photographer, not the camera. And so on.

And it happened that yesterday, a new music album arrived in my mail box. It came accompanied with a beautifully-made booklet that sets the lyrics against images of utopian landscapes. But on closer inspection, these images look wrong: features blending into each other that shouldn't (such as trees and buildings), roads leading nowhere, structures that would in reality collapse soon, strange shapes floating in mid-air, all signs that whoever (or whatever) made them, had no idea what they were doing. Apparently, those images were generated by AI. The algorithms have no understanding of what they do - which is of course a reason to worry what they may do in the future. There's a saying, "People are dumb, but computers are dumber" - and it is very true. And we all know that the greatest mistakes are mostly not made out of malice but out of stupidity. We should be aware of the risks.

[malloc]

And of course, the people concerned could not carry on as if nothing had happened; they had to adapt. Craftspeople, for instance, lost the mass market of new products; they had to specialize in what the factories couldn't do: repairs, and custom manufacturing. For musicians, recording became a major source of income. Office clerks had to learn to work with typewriters and calculators, and later with computers. The art of painting moved to things photography could not do - and photography became an art in itself, as the composition of an expressive image was still the work of the photographer, not the camera. And so on.

But why should artists and novelists have to adapt by ceding so much of the market to machines? Human artists and storytellers were here first, having been around for hundreds of thousands of years before AI was even a consideration. Now they must retreat to the fringes and allow AI to take the reins of our own culture. What you call a tool sounds more like an invasive species. Perhaps the oligarchs of the tech industry should adapt to the needs of artists and writers instead of demanding the reverse.

[Raphael]

I'll probably regret saying this, but...

Now they must retreat to the fringes and allow AI to take the reins of our own culture.

No they don't. That is, so far, only one possibility. As much as you hate AI, you still seem to buy into the hype of its professional salespeople.

[Raphael]

Random AI-related though: I guess from now on, when people talk about "AI" in a fiction context, they'll have to specify what they mean by "AI". Like:

-What role does AI play in your SF scenario?
-Do you mean "AI" as in "actual artificially created intelligence", or do you mean "AI" as in "the stuff that 2020s marketing people called 'AI'"?

What about the stuff that 1980s marketing people called ‘AI’?

This sounds as if you know about something very interesting that I don't know anything about yet. Could you tell me more?

[bradrn]

Random AI-related though: I guess from now on, when people talk about "AI" in a fiction context, they'll have to specify what they mean by "AI". Like:

-What role does AI play in your SF scenario?
-Do you mean "AI" as in "actual artificially created intelligence", or do you mean "AI" as in "the stuff that 2020s marketing people called 'AI'"?

What about the stuff that 1980s marketing people called ‘AI’?

This sounds as if you know about something very interesting that I don't know anything about yet. Could you tell me more?

In a nutshell: ‘expert systems’, ‘fifth-generation computing’, and all the other attempts to build an AI by explicitly teaching computers about how to reason (à la ELIZA). Stephen Wolfram has a very interesting recent article on Doug Lenat and CYC which I think captures the ideas of the time fairly well. (Not that I’d know, of course; I wasn’t alive then.)

[Raphael]

In a nutshell: ‘expert systems’, ‘fifth-generation computing’, and all the other attempts to build an AI by explicitly teaching computers about how to reason (à la ELIZA). Stephen Wolfram has a very interesting recent article on Doug Lenat and CYC which I think captures the ideas of the time fairly well. (Not that I’d know, of course; I wasn’t alive then.)

Thank you! Didn't Weizenbaum originally intent ELIZA as a parody of AI, though?

[bradrn]

In a nutshell: ‘expert systems’, ‘fifth-generation computing’, and all the other attempts to build an AI by explicitly teaching computers about how to reason (à la ELIZA). Stephen Wolfram has a very interesting recent article on Doug Lenat and CYC which I think captures the ideas of the time fairly well. (Not that I’d know, of course; I wasn’t alive then.)

Thank you! Didn't Weizenbaum originally intent ELIZA as a parody of AI, though?

I don’t really know the history in all that much detail, though Wikipedia seems to agree that it wasn’t meant as AI originally.

[malloc]

I'll probably regret saying this, but...

Now they must retreat to the fringes and allow AI to take the reins of our own culture.

No they don't. That is, so far, only one possibility. As much as you hate AI, you still seem to buy into the hype of its professional salespeople.

Only one possibility, sure, but quite a strong one. These generative AIs have already shown abilities far beyond what we could imagine from computers only several years ago. Give them another decade and they could well match even the best human artists and writers or even exceed our abilities altogether. It's less a matter of buying into hype than acknowledging the evident power and risks of this new technology. Even if current AI falls short of the hype, nothing prevents future iterations from achieving the goal of automating art and literature. The terrible truth is that humans are nothing special, just highly derived apes who lucked into incredible success. Replicating even our most vaunted abilities is simply a matter of applying enough computing power.

[KathTheDragon]

Give them another decade and they could well match even the best human artists and writers or even exceed our abilities altogether.

Extrapolating the current trajectory of the technology... no. The biggest hurdle is something that at least WeepingElf has already pointed out: the generative networks (I have decided to stop referring to them as AIs, given they are demonstrably not intelligent) have no idea what it is they're doing. They are certainly very knowledgeable and quite competent, but they utterly lack understanding, and we have no viable means of training them to acquire it. This is a manifestation of the alignment problem, namely that we can't be sure we know what exactly the system is optimising for, because what we think we told it to optimise for is typically not the actual optimum of whatever instructions we gave. There are countless examples of alignment failures, and right now the best we can hope for is that the misalignment is still nonetheless useful.

[Raphael]

the generative networks (I have decided to stop referring to them as AIs, given they are demonstrably not intelligent)

Exactly.

[linguistcat]

the generative networks (I have decided to stop referring to them as AIs, given they are demonstrably not intelligent)

Exactly.

Cosigned