I’m worried Valve will strangle what I think is the most exciting use of AI - artist trained models generating NPC content, lore, and procedural game content to create rogue like RPG of arbitrary complexity.
I’d note they also have no way to reasonably unpack all game assets, visual, text, and audio, and verify there’s no copyright material. To my memory there are several cases of video game publishers including copyright violating assets. It seems like a copout to take on faith a no copyright violation attestation for static content but not on models.
I'm sure you're right that's all they would need to win a lawsuit, but they would probably prefer to avoid one altogether. Or at least not to be the first publisher sued.
There's no reason to suspect a generic asset is infringing on it's face, but for models trained on the open internet it's an entirely unanswered question.
Indiscriminate crawler trained models are one thing, but models trained by the game developer directly should be handled differently. In that case they actually are the same as a generic asset, as they were assembled and curated by the game developer with an understanding of copyright compliance. But there are cases where they didn’t - particularly with audio.
If you tasked a human being with sufficient knowledge, latitude, and resources to investigate a game with a significant fraction of their time (eg they had a caseload of like 1-3 games) then I'm sure they could figure this out on a case by case basis. But I doubt they're willing to make that kind of investment.
But none of that makes it right, of course. I'm ambivalent about Valve's decisions here.
Edit for reply/clarification: "it's more likely" - I'm not speculating on the internal politics at Valve. Just stating the fact that, regardless of if assets are AI-generated or not, Valve cannot be certain they don't infringe copyright. In this respect, AI-generated assets with assurance the model wasn't trained on copyrighted works, are no worse than traditional ones.
they know that because steam is such a large store, they can influence the direction of the game dev community and set a norm that they prefer to see.
So why all of a sudden, that ai generated assets would get such attention?
Given that in the past that have been very reticent about taking down games (eg see [1][2]), I don't find it likely that they have an "anti AI" stance. I personally would need to see like, a blog post or some other statement to that effect in order to buy that hypothesis.
[1] https://www.thegamer.com/valve-gabe-newells-libertarian-stea...
This is summarizing reporting from People Make Games:
> There's no reason to suspect a generic asset is infringing on it's face, but for models trained on the open internet it's an entirely unanswered question.
As in, there's no reason not to give the benefit of the doubt to a traditional asset, but it's not yet understood how copyright works with a generative asset.
Am I missing something?
Except if you have the rights to all the works used to train your AI model. In that case, there is no risk that you are infringing.
I think it's interesting to wonder what Valve would say about games using AI exclusively server side. They cannot control (or know) what technologies a developer uses on their personal game servers.
My guess would be that they can't verify what AI you used, at least without investing the kind of resources required to actually make case by case decisions.
I'm not saying this is your position, this is somewhat tangential, but I am confused why I seem to hear from people that both AI is a force of nature and companies stand in it's way at their peril, but also that it's a delicate flower and that any kind of management may strangle it in the crib. To be clear I know you aren't making both claims, I don't want to put words in your mouth, but it does feel like it's coming from the same general community of people. And I'm perplexed by that.
But yeah, I would also be interested to play and perhaps developing games like that. If someone is developing a game like that and they wants to hire a backend dev with experience integrating with models in a realtime context, shoot me an email. (I'm half kidding, I don't expect someone like that to happen to be in this particular thread, but only half.)
How do they verify copyright material isn’t in static assets? I can’t imagine it’s very thorough. I imagine most of it is based on attestation.
Regarding your question, I don’t share those views but can understand how they exist simultaneously. With an open hand, like was afforded internet and technologies, it could shape the world in hard to imagine ways as did the internet. However like the early internet it could have been strangled in the crib through knee jerk regulation and prohibitions. I think AI model builders and hosters want safe harbor similar to content hosters and providers allowing the end use of the technology to be regulated but the providers to be allowed to operate on a best effort basis. For instance the copyright question here is strangling any game that uses a generative model. A reasonable legal outcome is that it’s fair use to incorporate copyright material, but the output must not violate copyrights.