

The law mostly disagrees with the memes = theft. A lot of it is covered through freedom of speech and fair use. If you have taken a bit of content, changed it a bit, recontextualized, and reposted it, you are most likely in the clear. Especially if the original content was publicly posted. This gets less clear if you are using the likeness of a private person but this will also depend on context. Where in the world you are, if this content was captured in a public space or from something published - the list goes on, like some stuff can be trademarked as well, and I’m no lawyer. A lot of these things run under the legal doctrine of “no plaintiff, no judge.” I feel artists in general have accepted that anything they post online is just potentially gone. And if no one steals their content to make money off it, they’re not going to hire a lawyer, whom they cannot afford.
And I’m not saying any of this is great but that’s an established status quo.
The reason why so-called AI generated art gets decried is twofold. It’s new and we don’t like new things. And in order for it to be created, the models have to suck in all the training data they can. And they don’t tend to pay for it. So that’s where some people see theft happening. But that’s not settled law yet because it’s fairly new, there are plaintiffs but not enough judges have passed judgement yet. Do they have to pay for stuff that’s publicly available? Where is the line, if any? Is imitation of a style okay if there is more to the work than just copying something from Studio Ghibli or Disney? These questions are going to keep a lot of legal professionals in bacon for a long time still.
This shit is hard. It’s more gray than black and white.
There are at least two discussions going on here simultaneously. Is the process of a beefed up spell checker sucking up all the data the same as an artist looking at what had come before, before either of them churn out new art? I’m inclined to agree with you; the process does seem similar enough. The difference remains that one is a statistical model and the other is a human being. So even if the process appears similar enough, they are two different types of player and I can also agree that we should not treat them the same. One is able to throw constant massive amounts of spaghetti at the wall as long as there are chips and power and the other is limited by their health and more limited processing power. So where the compromise lands in this discussion simply isn’t clear yet. And while you and I can discuss this, I can say for myself at least I’m not smart enough to see where this goes eventually.
The other discussion is how all of it collides with existing copyright/trademark law, which is essentially different in every country. Constitutional rights, like freedoms of expression and the arts, are given to real people, not computers. But at least one supreme court in this planet has made corporate money a form of free speech. So eff knows where LLMs end up.
This is new territory we’re in. And I fear that’s why it will take another decade until we get a legal landmark decision or a political compromise that will be similar enough all around the world.