"AI": The New Context

[sophilos]

An interesting thing happened on the thread titled "Can we get a subforum for AI work?". Now that the new subforum is in place (thank you, MTPJeff), the discussion has broadened over the issue. I believe the topic deserves a thread of its own, maybe more than one. And so, just for starters, what is LLM imagery to art?

It seems the main impact on the TMF of the Large Language Model (LLM) has fallen on imagery. I use that word in light of the vigorous discussion as to whether LLM is truly art or a distinctly different -- and maybe lesser -- form of creativity.

Perhaps LLM is a manufacturer design studio. "I want to order a side-by-side refrigerator in polar white with an icemaker ..." "I want to order a feudal maiden barefoot in stocks, her head thrown back ..."

Perhaps LLM is what Larry Niven called a Playground of the Mind, where some of the kids crawl over the monkey bars and climb up the slide.

Perhaps LLM is something else altogether. What say you?

 

[OSUchamps]

I believe it's art and has a place, much like how a producer or Hollywood idea man may not be the one who "creates" the film but they reap the benefits of it doing well nonetheless.

 

[Soles_Scream]

I wish DeviantArt would do something similar to what the TMF has done. If you tell Gronk or some LLM to make something which steals from other sources to make it, at least separate those works from the art that takes some actual practiced skill and care to put together.

 

[TheSolemnOne]

I wish DeviantArt would do something similar to what the TMF has done. If you tell Gronk or some LLM to make something which steals from other sources to make it, at least separate those works from the art that takes some actual practiced skill and care to put together.

DA has a AI tag, but some people aren't using it.

 

[Soles_Scream]

DA has a AI tag, but some people aren't using it.

Thank you, I will try to filter it out.

 

[Zmeta02]

Thank you, I will try to filter it out.

DA also used to have a note/rider at the bottom of a work that stated if it was AI generated, especially if it used Dreamup, but they seem to have removed that, so hope the tag works.

 

[Myrmidon]

I think this is a really interesting topic.
I've recently watched a conversation with Luciano Floridi in which he briefly touched on a similar subject. I think he goes into more details in his book, which he was presenting, but I still have to read it, tbh.
I'd link the video, but it's not in English, and the automated subtitles suck ass.

Anyway, in short, he thinks we need a new conceptual category, and he talks about the concept of "ectype".
Basically, he argues that between the source used to train the AI, the AI itself, and its output, there's a relationship similar to the one between the seal and the wax that was marked by it: the mark left on the wax isn't the seal, but it has to do with it.
Similarly, there's a peculiar relationship between the source material and the output of AI: in Floridi's opinion, in such cases the stark dichotomy between "original" and "fake" falls through, and hence we need a new concept, something in the middle of both extremes (i.e., the "ectype", opposed to the "archetype").

Another point touched by Floridi Is that probably with writing (he was specifically talking about it, but I guess we can speculate it'll be the same for visual arts) it will happen what has already happened in architecture and (as someone already pointed out in this thread) movie direction: a split between the creative idea/process and its production/realization, and thus a transition from having the authorship/property of something because it was physically created by us, to authorship/property as responsibility and accountability of who had the idea and oversaw its creation (exactly how, if a house was planned badly, we don't consider the workers who built it liable, but rather the architect, whose project was followed).

I think it's a really suggestive idea, and I mostly find it convincing. The real issue, which they didn't discuss in the video, is that it's no secret that basically all the GenAI models were and are trained on massive amounts of data (papers, texts, drawings, etc.) that were scraped without the consent of the authors of said works or any form of compensation for them.

So, if you ask me, "Is something made with GenAI art?" I would answer, "It depends."
I personally don't think that the amount of mass-produced slop you see on DeviantArt (thank God for the filter) or here is art. I'd probably be more forgiving towards someone who had a clear, distinct, and original idea and spent time adjusting the prompts and the AI's outputs in a creative process that had them overseeing and correcting the results until reaching something that fully aligns with his vision (instead of lazily typing a prompt and making 10000 unoriginal images with barely any kind of difference between them).

But again, the problem is that currently most LLMs are trained on stuff without the consent of the original authors, who aren't seeing a penny for the use of their works. This means that the AI businesses (because let's be real, it's what they are) are "profiting" (bracketed because currently they aren't even breaking even) off something that was trained on stolen property. That's the big issue to me, and that's something I'm strongly hoping will change in the future. But I'm not optimistic about it.

 

[sophilos]

Thank you, Myrmidon, for your thoughtful and constructive commentary. I was afraid I'd wasted everybody's time.

A generation will have to pass before Floridi's distinction will find it's way to a formulation that the general public will embrace. There's a lot of wilderness between common knowledge and strictly owned intellectual property. Somewhere in the middle lies the fogged ground of the battle over fair use.

Nine or more of ten still speak of LLM as "AI". If a true artificial intelligence exists, it has "read" Frankenstein and the biography of Martin Luther King. Of course it remains in hiding.

On the topic of the training of LLMs ... If I read a library of a defined scope, then move on to write on the topics within that scope, what do I owe to the writers upon whom my personal library was assembled? If I derive a concept plainly inherent within the scope but never before stated -- a "new idea" -- do I incur debt to those writers and/or their estates?

I never paid for my childhood education. But my parents paid property taxes, and I have chipped in over the years to the tutelage of tens of thousands -- at least. LLMs pay property taxes, where local and state goverments have not lured them with incentives such as reductions or exclusions to taxation. Most, if not all, offer limited services to the general public free of charge.

Perhaps I have built a straw man argument. LLMs do not learn. They assemble information in the structures established by language. Then they hallucinate. This is not a function of sentience but of engineering, and faulty engineering to boot. LLMs tend to wander over the boundary of context. Or to react like a schoolkid who obviously did not do his homework.

Translation (of Si iudicas, cognosce; si regnas, iube) from sententiaeantiquae.com: "If you are called to judge, inform yourself; if you are called to lead, command." I had to look it up, so I thought I'd share.

Again, Myrmidon, I thank you for your thoughtful contribution. When the bill arrives from Penguin Random House, current publisher of Seneca, I will gladly go dutch with you.