Best Of

Tiles, you are repeatedly mixing up law (what gets written in the law books, and determines practices that are illegal), and court decisions. A new law doesn't require a previous court decision to go one way or another - it's the other way around. And cases in which a court decision then becomes the norm because the law is ambiguous is called a "precedent".

As a matter of fact, this is precisely why some defense lawyers are still interested in cases where they get to defend something immoral - because this can then outline a loophole in dire need of being patched later by a new law. If a murderer gets away freely thanks to a technicality, it becomes an incentive for new laws to be drafted, proposed and voted in.

The example of the French law about touched up models didn't require any court case either.

Something is getting lost in translation here. No one is talking about "forbidding learning". The idea is simply to not use anthropological terms like "learning" and reframing it as "using" in order to get a clearer view on what's going on.

On law : law is not something that is divinely dictated and get set it in stone forever, otherwise all humanity would be stuck in dark ages doctrines.

Current laws may not be fitting to legislate on whether or not the data collection required for the existence of these datasets is an infringement of intellectual property or not. Since all law is subject to interpretation, in some case it may actually be. But the point to understand is that current law was drafted in a context when AI images generators didn't exist. It actually doesn't matter at all if the process is similar, exactly the same, or different from what humans do, this is completely irrelevant. What is relevant is that if enough people agree that such use of image data should be outlawed, then it could eventually become so. And new law on copyright protection isn't required to be triggered by a copyright infringement case settled on way or another.

For instance, tomorrow a given country could legislate that AI image generators need to come with a human-readable version of the dataset being used, and said dataset could be, for instance, subject to government approval before being released to the market, just like it is the case for food and drugs. I am not saying that this is the way it should be done of course - just giving one very simple example/scenario that only takes a minute to think about.

This is all very similar to human cloning : "Human cloning does nothing else than humans does. Just a bit more efficient and faster". Yet once it became a reality, laws evolved accordingly to express the will of the people on the matter of human dignity. Similarly to how in some countries it is allowed to have one's child gestated by someone else, yet in some other countries it isn't.

So when people present to you their reasoning as to whether or not this kind of art theft is/should be illegal, they are expressing their conviction and the direction they want the "artistic social contract" to go towards - one in which the unauthorized use of imagery in a data set used for image generation would be illegal. The fact that it currently is or isn't isn't the end goal of the conversation, it is the starting point.

I remember those days. Polycount used to be a lot more freewheeling like that.

Sure it used to be "easier", at least for some people. But there's this thing called systemic bias.

Once you step outside the old way of doing things, and actually see all the embedded misogyny, homophobia, and racism, well you just can't unsee it anymore.

Personally I'd rather be a bit inconvenienced than to continue to reject & harm all the people who never fit into the old way, and just had to put up with all the bullshit all the time.

Fuck going back to that. DEI is a good thing, and it's been way too long time coming.

@Alex_J you kinda underestimate how small this industry at least on a very high level is. You don't have to know everyone, but knowing someone who knows someone is not unlikely. In the western market, at my age group i would say in 3d most people have gone through the "polycount school" chances are high that this alone forms a little trust between people.

This doesn't mean we know everyone in our agegroup world wide, but in the western market it's not like there are tens of thousands of people in leading positions. Its a few hundred.

If you check where someone worked, you will likely find someone you know who worked with that person before.