sabotabby: (books!)
[personal profile] sabotabby
I continue to spend a non-zero amount of time arguing with AI techbros, and soft-AI supporters ("I use ChatGPT to polish up my writing" "I have my own offline LLM" "We need to have guardrails and learn how to incorporate it responsibly into education"), because yours truly is an internet masochist who regularly engages in online self-harm. Nah, actually because I think the "we should abolish it altogether" position is woefully absent in the discourse. It's possible that LLMs could go the way of NFTs (a punchline that's already getting dated) or crypto (also a punchline but with a very narrow use case for the worst people you've ever met), but only if we open the Overton Window to allow an abolitionist perspective to be let loose into the mainstream.

No one asked for this. No one likes it. No one wants it. It will make your life more annoying, not less.

The second least justifiable case* for the broad grouping of technologies called AI** is creative work—writing and art. It's already led to the mass firing of journalists, which, granted, was already happening because of capitalist consolidation. You will notice that the quality or availability of journalism has not improved! Hilariously, the economic case for this is also terrible, in that artists, writers, journalists, actors and voice actors, filmmakers, and animators already make no fucking money, and a request from an under-$200/month subscriber actually costs the companies money.*** (Is it moral to use ChatGPT to drive OpenAI out of business? Discuss.) Replacing or drowning out creative work with AI slop makes no one money and creates worse things that you have to wade through to find things made by humans.

computerspiteful
Probable source.

And yet, there are a certain number of AI "artists" or "prompt engineers" who insist that I should take their little computer pictures seriously as if it was real art. Buddy, I do not take my pictures seriously and I drew them myself. They liken what they do to collage, or sometimes Pop Art. Unfortunately for them I studied Art History and they usually shut up real quick when I tell them that my prerequisite for discussing it is them reading the entire judgment in the Warhol Estate v. Goldsmith case.

(Incidentally, I disagree with the judgment, and the last people who should be defining fair use are the ghouls of the US Supreme Court. But it's an interesting case, and one that to my non-legal mind conclusively shows that all AI "art" is in violation of US copyright law.)

My moral argument against LLMs is:
1. The catastrophic environmental cost.
2. The intellectual property theft.
3. The economic consequences of job loss.
4. The world doesn't need more shitty art.

If these people knew anything about art at all, which they don't, they might bring up fanfic rather than Pop Art. After all, we're talking about transformative work, specifically of creative intellectual property. It got me thinking, in a tangential comment on one of [personal profile] princessofgeeks 's posts, about how fanfic is the bright mirror of AI.

Because I think transformative work is good, actually. It's the main way we've had, for the entire history of humanity, to engage with story and art. The idea of a work being the sole property, for the purpose of sole profit, of the artist who created it is relatively new, and has to do with economic conditions far more than it has to do with an accurate description of the creative process.

Hence, I would pose the question: What is the difference between a fanfic of a piece of intellectual property (you can insert your own fandom here, the way [REDACTED] inserts their [REDACTED] in [REDACTED]'s [REDACTED])? After all, you are taking a thing that belongs to some other creator and making your own thing out of it.

This is where having a moral framework comes in handy.

1. Is there an environmental cost? No more than any regular activity you'd do on a computer. The energy difference between me writing this post, or my own books, and someone posting smut to AO3 is nonexistent. We all need to cut our energy use down but let's ground the private jets first, y'know?

2. Is it intellectual property theft? I would argue no, actually, and while this has been litigated a few times, people don't get sued for writing fanfic as a general rule. It's not, morally speaking, for the same reason that collage is not. No one is claiming that Hannah Höch photographed all of those magazine images she put in her collages; you are very obviously seeing found work that is repurposed, with intent, to create new meanings. Fanfic is the same—it can't exist without the acknowledgment of the authorship of the original canon. If the original canon suddenly disappeared, or was overshadowed by the fanfic, the meaning would be lost. The purpose of fanfic is to honour the original work, or subvert it, or deconstruct it; it is never to erase it.

3. It's the economic aspect that I find most interesting. Companies like OpenAI speak openly (hah) about crushing entire industries in order to somehow extract profit. Although, again, why they plan to do that with the arts, which are famously unprofitable, is beyond me. Blood from a stone. Fanfic, however, is a gift economy. That's why I call it a bright mirror. Paying for it would seem gauche; when fanwriters have tried to charge for their work, they're soundly mocked by a community of accidental anarchists.

In fact, this is a reason why fanfic writers aren't sued. Fanfic and fanart inevitably creates more income for the original creator. How many times have I checked out some show because someone has drawn an incredibly pretty, incredibly filthy illustration of the characters? A non-zero amount of times, I can tell you. If you ever write fanfic of my work I will love you forever.

4. Well, one can argue that a lot of fanfic is shitty. But because it's published, most of the time, through a parallel ecosystem, you don't actually have to wade through whatever the modern-day equivalent of My Immortal is to find an actual book. So the shitty stuff harms no one. Maybe the calculus shifts a bit with the publication of, say, Ali Hazelwood's stuff, but that's not my genre so I don't care.

Artistically, fanfic is communal and process-oriented, whereas AI slop is individualist and product-oriented. I can probably still go to AO3 and find something, within one or two clicks, that floats my boat. (Like I'd do that. I am a lazy asshole. I'd ask one of you, and you'd give me a recommendation.) To find something in the sea of slop that has any kind of artistic merit is impossible. Even if it did exist, and it doesn't, it'd be impossible to find. During one argument I had with an AI fanboy, he claimed to have rendered 100 images in the time it took for me to destroy his argument.† With one person creating that volume, how can anyone find anything?

It has never occurred to any of these people to turn to fanfic or fanart to improve on their skills. This is because they are genuinely uninterested in creative work. They don't want to be artists or writers; they want to claw themselves a little higher on the pyramid scheme, not understanding that they're the product no matter how hard they try.

It's actually because fanfic writers and fan artists exist that I have hope that the scourge of slop can be defeated. Creativity is so innate that it can thrive even in the absence of a profit motive, and for all its flaws, AO3 is an example of elegant, usable website architecture with safeguards built in against monetization. Even if everything goes wrong, we'll still be telling horny stories in the burned-out irradiated ruins, and I really love that for us.

* The worst use case for AI is anything having to do with war or police or surveillance, obviously. The immediate case for abolition is that this is used against Black and Brown people to kill them. For that reason alone, it's ethically justifiable to build a supervillain-sized magnet and take it to any data centre in your vicinity.

** To be clear, AI used to detect cancer is not the same as LLMs, and anyone trying to convince you of this doesn't want to cure cancer, they want to get every journalist fired so that the Nazi App is state media.

***  This is because capitalism does not in fact work the way they teach in business school, where companies are required to turn a profit. Companies like Uber run at a loss. Uber has never made money. It just drives the cabs out of business and defunds public transit, so you're now reliant on Uber and will eventually pay anything for the service.

† All of them looked like bad, slightly thirsty knockoffs of Coraline. This was a few weeks ago, so that aged like sour milk.

Date: 2025-01-18 07:58 pm (UTC)
ioplokon: purple cloth (Default)
From: [personal profile] ioplokon
Tangent bc you're right about LLMs but this conversation bores me at this point (not that your post is boring just... don't really care to talk about it more). The discussion of fanfic, IP, and who gets sued/punished reminds me of the big battle between wrestling companies and giffers/clippers. A lot of promotions, especially in Japan, were extremey aggressive at going after people who made gifs of their shows. It is kind of an interesting one because basically... in order to make the gifs you are probably accessing the stream illegally, right? But it has huge promotional value (so much so that it actually shapes the product; people do spots because they know they'll get clipped and shared).

In the end, I don't have a point here, but it's a different case study & kind of interesting since it involves maybe some crimes and also giffers are potentially monetized. (& it's less arguably transformative - no one goes after Botchamania because Maffew is clearly adding commentary. Also bc all the wrestlers love it)

(the flip side of this is the classic #lolTNA moment when the company's official pay-per-view stream went down and they treated out step-by-step instructions for how to access mirrors and torrent it)

Date: 2025-01-18 08:54 pm (UTC)
ioplokon: purple cloth (Default)
From: [personal profile] ioplokon
Not really, I think. It kind of just goes in cycles of the company caring and then not caring. A bit weird.

*

Date: 2025-01-18 08:05 pm (UTC)
minoanmiss: The beautiful Finn as the king he is (Pharaoh Finn)
From: [personal profile] minoanmiss

Te adoro. saves this rant

Date: 2025-01-18 09:34 pm (UTC)
frandroid: A key enters the map of Palestine (Default)
From: [personal profile] frandroid
A quick search on my non-google search engine of choice led to this:
https://knowyourmeme.com/memes/a-computer-can-never-be-held-accountable

which lead to the source:
https://bsky.app/profile/chrisdeleon.bsky.social/post/3lfdv6wk4xk2q

Date: 2025-01-18 10:01 pm (UTC)
radiantfracture: Beadwork bunny head (Default)
From: [personal profile] radiantfracture
This is *wonderful* and very relevant to ongoing conversations with colleaguefriends. May I share?

Date: 2025-01-19 07:02 am (UTC)
minoanmiss: Minoan version of Egyptian scribal goddess Seshat (Seshat)
From: [personal profile] minoanmiss

Oh I camee back to ask the same question

Date: 2025-01-19 01:40 am (UTC)
lynnenne: (politics: big damn hero)
From: [personal profile] lynnenne
I admire your ability to argue online with AI tech bros. I can’t even read their bullshit without retreating into a depressive coma of despair.

Date: 2025-01-19 05:13 pm (UTC)
lynnenne: (mood: puppet angel wants a new village)
From: [personal profile] lynnenne
LOL, we can only hope!!

Date: 2025-01-19 03:27 am (UTC)
viggorlijah: Klee (Default)
From: [personal profile] viggorlijah
We are required to use AI at work and it has an excellent bullshit generator of ‘feed in angry email to dickhead colleague’ and get back polite note with buzzwords that doesn’t get you fired. Which is my main use as well as giving it templates for paperwork I have to churn out regularly and then spitting out versions like a MASH generator. There are specific use cases but everyone is nervous about reliability and hallucinations. The biggest AI work-worry I have is in my specific area with helpdesk support where AI can create unhelpful and incorrect loops trapping people and I have to talk down the sales hype of cut your support team in half! With yes but - that only works if you keep your support team remaining with MORE time to handle the tricky cases not spread them even fucking thinner. AI at cost centres like a helpdesk is very very financially tempting but there is already one airline who used AI to replace their helpdesk here that I will never book with again because they did such a shitty job of implementing it.

Oddly enough I’m not worried about environmental impact as much because Moore’s law and improvements plus the shift to renewables - I feel like this is a distraction akin to your personal carbon footprint. Creativity slop and bringing in AI recklessly to replace people where human judgement is needed absolutely. And where do you get the experience to feed both the AI and the human evaluator if you don’t have people doing the jobs to be replaced? It literally cannot replace the thing it is modelling because then it eats itself a this is NOT generalised intelligence but statistical mimicry which needs stuff to copy.

Date: 2025-01-19 07:41 am (UTC)
yhlee: Alto clef and whole note (middle C). (Default)
From: [personal profile] yhlee
To [personal profile] sabotabby: re: [personal profile] viggorlijah's excellent points: that last sentence, especially, is something that has already been shown mathematically; can dig up cite if you're curious, Husband confirms the math argument looks sound.
Edited Date: 2025-01-19 07:41 am (UTC)

Date: 2025-01-19 03:04 pm (UTC)
yhlee: Alto clef and whole note (middle C). (Default)
From: [personal profile] yhlee
No, the statistics of how that type of model is structured; I don't know if it is formally/technically a mathematical attractor (if you're familiar with that term) but that's the idea.

(Sorry if incoherent; I've had two hours of sleep.)

Date: 2025-01-19 10:14 pm (UTC)
brittdreams: (Default)
From: [personal profile] brittdreams
Agreed on the renewable energy side. There's also the issue that most renewable energy sources don't provide power 24/7 so you either have to pair sources or have a lot of battery storage.

Also the fact that tech companies are willing to restart a closed nuclear reactor to power their data centers is a sign that they don't think current energy sources will be sufficient. They're also focusing on the "zero emissions" aspect of nuclear to avoid convo about actual renewable energy sources.

Date: 2025-01-20 02:30 am (UTC)
viggorlijah: Klee (Default)
From: [personal profile] viggorlijah
the help lines are frustrating because properly put in the not-real-AI can be super useful - there's a whole heap of technical documents and maybe ten years of customer data in questions and answers, so by priming a LLM on those documents and emphasising that recent documents are the most correct, and to look for certain similarities and to change the complexity of the answer etc - you can have a fairly new customer agent able to take a call and immediately get a whole host of links to similar cases and a drafted script with links to the relevant docs and even have a manager able to cue to step-in when a customer starts getting unhappy with an auto-summary of the call and the problem etc, instead of having the new agent still trying to get search through PDFs and FAQs and asking colleagues wait what do I do here and the manager having to be called in for a call frustrating the caller further because they have to recap everything all over again - it can be useful!

But no, instead we get the stupid "Please say what you are looking for, I don't understand" (which is because the company doesn't want to bother putting together a decent IVR which requires looking at call categories and some basic user testing and forgets that not everyone speaks with a flat american accent) and generated bullshit because the customer doesn't know the inside system so is trying to explain what they want with the wrong key words and no human to translate that and you wind up with people self-helping with absolutely no knowledge structure to guide them because people think outlines and processes emerge glistening from the muck rather than being mental frameworks developed to navigate and -

I am about to have this argument at work over my upcoming helpdesks and spending time walking people through this. At some point, people are going to realise that the reason we learned Dewy Decimal and the alphabet and all those other categorisation systems is not because our teachers were sadists but because learning how to categorise and link things is critical to understanding them and you can have two identical piles of things and organise them in entirely different ways which are both useful because IT MATTERS. We are in an ocean of information, and we're losing the craft of structuring and guiding ourselves through it.

Date: 2025-01-20 03:49 pm (UTC)
frandroid: Data banging an Enterprise computer screen which just showed the BSOD. (bad technology)
From: [personal profile] frandroid
We're at a critical juncture of reaching climate apocalypse, and Moore's Law is basically over, so I find this to be misplaced optimism. Not to mention that the Jevons paradox is also in the mix--this is a new use of CPU cycles
Edited (Jevons) Date: 2025-01-20 03:50 pm (UTC)
dragonessie: Jean Luc Picard holding his head in his hand (mortified)
From: [personal profile] dragonessie
No one asked for this. No one likes it. No one wants it. It will make your life more annoying, not less.

THIS! I've been thinking this to myself for nearly a year now... Probably ever since I first saw lobster Jesus.

Sci-fi nerd that I am, I find a more compelling argument for abolition in the moral ambiguity of what constitutes a consciousness, and whether we're truly ready to become new gods and create a whole new category of souls capable of doing good and evil in new and unexpected ways.

Data and Lore are naturally my mental templates for this dichotomy.

In the real world, we all know the consequences of pretending that a whole subgroup of actual humans weren't actually real humans with souls. But if AI ever develops real consciousness, and furthermore, begins to experience suffering, we're cooked.
Edited (grammar) Date: 2025-01-19 08:59 am (UTC)

Date: 2025-01-19 03:16 pm (UTC)
yhlee: Alto clef and whole note (middle C). (Default)
From: [personal profile] yhlee
I don't dispute your general points regarding (ay) LLMs; one of the difficulties with banning/legislating it categorically/completely, as opposed to specific use cases: aside from cursed financial incentives in the vein of "well, we'll just Do The Bad Thing with other people's data, profit!!!, then pay a slap-on-the-wrist fee seven years later if/when taken to court" as a "cost of doing business", and the general slowness of legislative measures vs. the speed of technological development (see Doctorow and Solnit on this IIRC) is that from the standpoint of mathematical structure of the construction of ~AI technologies, what is going on mathematically in an LLM often differs in scale/cost rather than qualitative difference of type/construction from generally "unobjectionable" use cases that have been around for a decade-plus like "suggest some harmonic progressions based on what I've written (composed) already." (Harmonic progressions in Western tonal music that ~~~sound sensible are combinatorially such a restricted domain anyway.) There are definitely a ton of problem use cases for all manner of reasons, but a number of them have problems because they've been problems for a long time for reasons external to tech implementations (e.g. racism in profiling, which tech exacerbates, but tech didn't cause the root issue of racism). It's a conundrum.

(Sorry if I am not expressing myself clearly; still sick.)
Edited (typos) Date: 2025-01-19 03:16 pm (UTC)

Date: 2025-01-19 03:33 pm (UTC)
yhlee: Alto clef and whole note (middle C). (Default)
From: [personal profile] yhlee
Right, it's buzz word hype. But human susceptibility to buzz word hype is sadly not caused by specific buzz word tech (old or new) but by humans humaning, even though specific impacts can be ameliorated/legislated/dealt with by addressing how the tech is legislated/incentivized/disincentivized. It's a picklement.

(Apologies - if I bow out of this discussion randomly in the middle, it's because I'm sick and have had to go lie down again. /o\ )

Date: 2025-01-19 03:31 pm (UTC)
yhlee: Alto clef and whole note (middle C). (Default)
From: [personal profile] yhlee
Example of a use case that is ~generally "unobjectionable" to a number of people (whether one agrees with this position):

https://www.informationweek.com/machine-learning-ai/how-ai-is-revolutionizing-prosthetics-to-refine-movement

Note: I know someone who has an amputee family member using a prosthetic limb of this general nature, although I haven't asked for more specifics since that would be, well. Medically/personally intrusive.

Date: 2025-01-19 03:45 pm (UTC)
yhlee: Alto clef and whole note (middle C). (Default)
From: [personal profile] yhlee
Also e.g. meteorological models. In fact, this is of interest to people in that field in terms of improved forecasting again for math reasons - because weather models are generally nonlinear dynamical systems (colloquially, "chaos theory"), there are often limits to prediction accuracy based on those models; sensitive dependence on initial conditions and the nonlinearity of those systems (colloquially, "the butterfly effect") mean that you do not necessarily get that much better predictive power by devoting more processor cycles or by "we can now compute with 40 decimal places rather than 28" because a perturbation in the 31st decimal place can veer off wildly. (Briefly: in a linear system, if you change the input by a small amount, the output changes by a small amount; if you change the input by a large amount, the output changes by a large amount. In a "chaotic" system, that's no longer the case; you can get "shrug whatevs" from a large input change or "ZOMG what" from a small input change except, because it's nonlinear, there isn't a straightforward predictable correlation. This is an innate characteristic of the type of mathematical system. Again, sorry if I'm explaining this poorly.)

(OTOH, climate change deniers - that's a humans humaning problem.)

Date: 2025-01-19 03:55 pm (UTC)
havocthecat: amy pond of doctor who with a magnifying glass (dw amy pond investigates)
From: [personal profile] havocthecat
This is fascinating to me. Thank you for explaining it in a way that makes sense to a non-mathematician!

Date: 2025-01-19 03:57 pm (UTC)
yhlee: d20 on a 20 (d20)
From: [personal profile] yhlee
No problem - perhaps derailing to go into further detail here, but you can always hit me up for a layperson explanation from the math end.

Date: 2025-01-19 10:13 pm (UTC)
petra: Barbara Gordon smiling knowingly (Default)
From: [personal profile] petra
This is very well said and I have subscribed to your newsletter.

The world doesn't need more shitty art.

Date: 2025-01-19 10:41 pm (UTC)
From: [personal profile] blogcutter
Maybe the world doesn't need it, but I think I as an individual do! For me, a person without much artistic talent, it's the process rather than the product that's the thing: the sheer tactile pleasures of mucking about with art materials while shedding my immediate frustrations or letting my mind wander. Art therapy, self-expression, whatever elusive phenomena are at work - those are important to me. I think that's probably behind the trend towards adult colouring books although they're not really my thing. So far, I haven't counted ChatGPT or so-called AI technologies amongst my art supplies, but some of my friends have - using ChatGPT prompts for example, to generate stories and poems about the secret lives of their pets. I can see it as sort of a parlour game or a form of cheap entertainment, akin to reading your horoscope every day even though you have no illusions about its accuracy as a predictive tool or any kind of good advice!

When it comes to shitty art as a commercial proposition, depriving already starving artists and other creative folks of even a modest livelihood, that of course is a whole different matter. And I'm actually happy to see someone introduce the "We should abolish it altogether" position into our heavily tech-normative contemporary discourse.

Date: 2025-01-20 12:32 am (UTC)
vass: Small turtle with green leaf in its mouth (Default)
From: [personal profile] vass
4. The world doesn't need more shitty art.

Somewhat disagree on this part.

The world does need more shitty art... but none of the good in shitty art can come from automating the process. It is good when human beings make bad art because they love what they're making or the process of making it or are trying to improve their skills.

Commercialism and LLMs have a lot in common. A few years ago I noticed that that year's trend for Christmas decorations on sale in supermarkets (not used to decorate the store, just for customers to buy and display at home) was shabby chic/rustic/faux-handmade. Burlap with big, uneven yarn stitches, deliberately done to look like something a child made. Which might have been cute if the children who made them each made only one, for their own tree, or maybe one more for each of their living grandparents.

Somebody (it was long enough ago that I don't think the marketers were taking their orders directly from AIs yet) saw ornaments like that on Instagram or Pinterest and thought what people liked about those ornaments was how they looked, and that they would want to display them.

Marketers are human beings, and I don't think they are categorically so lacking in human nature that the idea that people might only want those ornaments because they had a direct, personal connection to the maker of the ornaments, never for a moment entered their heads. But if it did enter their heads, it was deemed irrelevant to what they were doing. Human connection is an externality.

See also: "natural-look" storage boxes made from plastic faked to look like cotton canvas or linen. They've identified a consumer demand, products made from natural materials, and missed what about that the consumers want or why they want it... because their whole way of doing business depends on missing that point.

Profile

sabotabby: raccoon anarchy symbol (Default)
sabotabby

May 2026

S M T W T F S
      1 2
345 6 7 8 9
10 1112 13 14 1516
1718 19 2021 2223
24252627282930
31      

Style Credit

Page generated May. 23rd, 2026 01:34 am
Powered by Dreamwidth Studios

Expand Cut Tags

No cut tags

Most Popular Tags