4 ms·

I’m dubious they need more than an affidavit from the publisher and developer attesting their model was trained without copyright material. The publisher and d

I’m dubious they need more than an affidavit from the publisher and developer attesting their model was trained without copyright material. The publisher and developer are the ones directly liable for copyright infringement. Valve does it’s own verification of game content and that combined with a legal declaration from the publisher should be enough to insulate valve. 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.

maxbond 3y ago
Steam is a publisher, no? (ETA: I guess they're not necessarily the publisher. From Wikipedia: "Initially, Valve was required to be the publisher for these games since they had sole access to Steam's database and engine, but with the introduction of the Steamworks SDK in May 2008, anyone could integrate Steam into their game without Valve's direct involvement."). 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.
- fnordpiglet 3y ago
  Except in some of the linked Reddit threads the game authors explicitly said they would train models with curated content that’s under their control and open in license, which is not the open internet. 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.
  - maxbond 3y ago
    I agree that morally and presumably legally these are completely distinct. But from a policy perspective, they aren't going to be able to verify that the output came from one model or another. So allowing any models is tantamount to allowing all models, because people who wanted to use eg Midjourney could simply lie. 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.
    - SenAnder 3y ago
      fnord already mentioned this, but to repeat: the situation with AI is the same as with other assets - Valve has no guarantee a developer didn't infringe copyright in taking some other model/texture/audio, tweaked it a bit, and put it in their game. 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.
      - chii 3y ago
        it's more likely that there's an anti-ai evangelist on the team doing these reviews, and they are responsible for causing these issues (and using the legal turmoil surrounding AI copyright to hide their bias). 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.
        maxbond 3y ago
        Why is that more likely?

maxbond 3y ago

Steam is a publisher, no? (ETA: I guess they're not necessarily the publisher. From Wikipedia: "Initially, Valve was required to be the publisher for these games since they had sole access to Steam's database and engine, but with the introduction of the Steamworks SDK in May 2008, anyone could integrate Steam into their game without Valve's direct involvement."). 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.

fnordpiglet 3y ago
Except in some of the linked Reddit threads the game authors explicitly said they would train models with curated content that’s under their control and open in license, which is not the open internet. 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.
- maxbond 3y ago
  I agree that morally and presumably legally these are completely distinct. But from a policy perspective, they aren't going to be able to verify that the output came from one model or another. So allowing any models is tantamount to allowing all models, because people who wanted to use eg Midjourney could simply lie. 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.
  - SenAnder 3y ago
    fnord already mentioned this, but to repeat: the situation with AI is the same as with other assets - Valve has no guarantee a developer didn't infringe copyright in taking some other model/texture/audio, tweaked it a bit, and put it in their game. 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.
    - chii 3y ago
      it's more likely that there's an anti-ai evangelist on the team doing these reviews, and they are responsible for causing these issues (and using the legal turmoil surrounding AI copyright to hide their bias). 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.
      - maxbond 3y ago
        Why is that more likely?