All AI generated content is doing is ripping off peoples work, it’s just AI seems to come with a get out of jail free card.
If you'd spend any time at all looking at midjourney creations, there's no way you can conclude that there's any significant amount of "ripping off" people's work. Certainly not in all or most cases.
With the right prompt you might be able to get something that looks eerily similar to an existing work, or artist style, but hardly anyone is interested in doing that, except for memes which are by definition modified and transformative. Nobody wants to AI-generate a mimic of a copyrighted work. If they wanted to do that, they'd just rip off the original work.
None of this is a new issue. The Warhol case even resolved one remaining facet. If you take an original work and apply photoshop filters to it, or do the equivalent manually or with AI, that doesn't alleviate the copyright compliance burden. The fuzzy area is how similar something has to be to fall under copyright; that'll remain subjective until the law or courts resolve it or copyright collapses entirely under its own weight, but the vast majority of AI generated content is nowhere close.
They were trained on work of people who didn't consent to it, the authors were not compensated, and currently are losing money (demand for their work) because of it. How is that not ripping off.
Please don't compare software to people. They're not similar legally, morally or in any other meaningful way.
Copyright is not about consent to use. It's about consent to copy.
You're confusing copyright with licensing, where, for instance, I pay amazon to contractually license me an electronic copy of a book.
AI companies did not enter into and then violate contracts with anyone for the material they used to train. Some of them may have run afoul of ordinary copyright law to acquire the content in the first place.[1] Even supposing they did, that act is the violation; it wouldn't subsequently be an additional violation to train an AI using those works, instead of keeping the works around on a storage array, as long as it wasn't redistributed.
The alternative to AI models currently at issue—created by OpenAI Midjourney StabilityAI Meta Alphabet et al from a corpus of partly openly available content and partly content of dubious provenance—is not some license scheme where creators get compensated equitably. The real alternative is for companies like Disney and Adobe and book publishers and record labels to band together and have their own paywalled AI models and extract rent from that forever, which would still be sufficient to saturate the market for imagined content, and would still kill demand for everyone else's work.
[1] I'm skeptical, though open-minded, about this. I'm not aware of any successful lawsuits, or any lawsuits at all, targeting defendants who only downloaded copyrighted content from pirate sources, as long as they did not upload or re-upload at all. There might have been, and probably were sometimes, threatening letters sent to people's ISPs. Threatening letters don't mean anything at all.
And copyright applies to every copy, not just when you download something to train the model. It also applies every time you make a copy with said model.
A whole chapter could be written about turning (other peoples) images into executables and assert ownership with a new kind of DRM.