So, what's your view on how we should legally think about AI art models that use copyrighted art for training data? I'm not sure there's a compelling analogy?
Clearly this is what human artists do all day long, of course. And the question is in what way is that different form having an AI do it. Any thoughts? I'm particularly interested in people familiar with the most relevant law.
I know a lot of artists are basically just opposed to this sort of thing, and part of that is the harm done to their lifestyle. To be honest I am torn. I don't like the idea of creativity being consolidated into powerful models held by a few corporations. But then most entertainment is already held by a few corporations.
More deeply... a world of really good AI art is kind of magical?
Like, I get it - if you're like me you've spent a ton of time getting decent at a craft and then to see an AI do it is upsetting. But if you'd told me in the abstract "in the year 2030 or so, you'll be able to make a feature film in your basement" I'd be in favor of it at almost any cost!
@ZachWeinersmith Film's a pretty good one to find an example of, and dive into on this matter.
Lines in movies are, generally, only 'good' because they are built up/resonating with previous things in the film. 'No, I am your father' only works because it does a direct clashing comparison with previous moments through the movies, and the running themes. Emotional scenes (happy, sad, cathardic, epic) trigger because of the buildup and connection you have to them.
The buildup is required.
@Oggie I agree with all of that, but don't you think that several of those emotional cues are formulaic enough to be interpreted and regurgitated by AI? Think of the rom-com genre as an example. Those are just patterns and AI is great at mimicking patterns.
@LouisIngenthron I mean, unless you want to watch a dada-esque romcom where it turns into a pottery documentary for 10 minutes then has a space fight with aliens and then they go back in time to when they were lizards and then the sun explodes.
Because that's somethign that could happen, a dream-like pseudo-constructed narrative that has no overall cohesion but a moment-to-moment functional flow.
@Oggie
So, in a sense, it would be closer to something like lucid dreaming, externalized?
@LouisIngenthron
@Oggie Well, you're describing conversational chat AI. There are different forms that can write and maintain longer chains of context, even now. Especially if you train them yourself for best results.
@LouisIngenthron Certainly, but the internal inconsistencies within the AI systems I find best to think of like resonance errors- they amplify and magnify each other. Tiny flaws that disrupt attempts at unity.
People tend to think of the 'bad' movies and 'bad' video games as 'lacking flow' or 'lacking structure', but they forget how much there -is- of that even in the worst.
I think we're going to see an offshoot of the uncanny valley, where the closer it gets to human-created, the more off.
@LouisIngenthron Eventually, yes- I would say 10 years, probably.
I think people really miss exactly -how bad- AI is right now at keeping thematic pattern matching going along a consistent system right now. LLMs are a solid example, when you can (trivially) get the LLM to disagree with itself in the second half of the same sentence. AI art will start one half the picture in water color and end in photrealistic, jarringly (there are filters to cut down on this).