Sarah Silverman and other artists sue over AI tools trained on their work

Originally published at: Sarah Silverman and other artists sue over AI tools trained on their work | Boing Boing


That these models were reportedly trained using pirate torrent trackers as training material is really something, isn’t it?

Could the cost of the vast training data otherwise be even calculated? Scraping was always sharing a bottle of rum with the more overt pirates. And by far the most interesting social issues are those one can appreciate both sides of [contemplative emoji face skin color aubergine]


This issue is a massive problem for all of AI that’s trained on a large corpus of whatever. Whether it’s text models trained on millions of blog posts, or image recognition models trained on posted images, if there isn’t a clear assignment of rights, well, the AI model is encoding information from the content, so it’s a violation.

Even something as simple as “find all the squares with a bicycle”, which I assume is creating a training corpus… that’s encoding my knowledge, but I didn’t assign rights for that.

Pretty sure that Big Tech will find a legislative solution for this.


I don’t see that it’s a massive expansion of copyright not to allow AI companies to copy everyone’s pictures and writings off the internet.


I think the point is that it could be if we’re not careful in how this is done. Frankly, I’d rather see this issue be dealt with by the legislative process than by the judiciary. A court could easily issue an overly broad ruling that would expand copyright protection to the point where an author who writes in a similar style to another author could get into trouble, and we don’t want that.


We’ve already reached that point with pop music.


Yeah, fortunately, courts have pulled back from the brink in recent music copyright rulings, including in that case. See also Katy Perry and Led Zeppelin.

1 Like

There’s a difference between writing in the style of another author, and copying all their work to remix it and write a “new” work.


1 Like

One of the problems with training programs like ChatGPT (I can’t bring myself to call it Ai) is the indiscriminate gluttony of the feed stock. It’s making sausage out of everything dumped into it, without care or any recompense for source.

It’s also pretty crappy about making shit up as it goes; don’t expect anything factual to be correct.


This topic was automatically closed after 5 days. New replies are no longer allowed.