Why the Training of Generative AI Models Is a Violation of Copyright Law

This conversation has always been muddled by the fact that the history and law around "copyright" has always been very muddled, and the begging question of if "copyright" is even a good thing or services its presumed purpose.
Independent of AI, we should all be considering if we even agree with the modern notion of copyright.
A reminder of an older time, but very much worth the watch..
https://www.youtube.com/watch?v=mhBpI13dxkI

Twelve paragraphs in, and you still haven’t made your argument?
Are the AI companies at odds with copyright laws when they train new models? I think, yes.
Ah, because it’s a bullshit opinion piece being presented as fact, that makes sense.

It doesn’t matter.
The ultimate utility of the ai and the threat that it poses if the west falls behind in ai research, is too great to allow copyright laws to get in the way.
So either they’ll say that the law doesn’t apply or they’ll make new laws.
Pandora’s box has been opened and we can’t go back.
- It's the new nuclear arms race. But even if we win, we may still lose.
  
  Yep, but we can’t afford to “not compete”

The article never actually answers it's own question. I think the courts will eventually find gen AI transformative, like it has done with all forms of AI trained on public facing data before it.
- Exactly. Trash article.

any copyrighted work must be given (copied) to the AI company to be injested
if that transaction is not sanctioned (licensed) by the copyright owner it is an infringement
as for social media content, i think we sign our rights away when we sign up for them, but they are subject to copyright by Meta or whoever

Yes, but the real issue is the egregious violation of personal privacy, especially for cases like Meta’s glasses.
- If you think that hasn't been happening for decades, you're younger than you think.

Thought experiment: What if it an extremely intelligent person memorised every Metallica song on a free Spotify account, down to every small detail. Later that person writes a new song, it's heavily inspired by Metallica, it sounds like them, you might even mistake it for them except for the vocals, but the lyrics are new, the chords are new, etc. Did that person then violate copyright, even though it's a completely new song?
I know the AI techbros are just scraping every datapoint they can get their grubby little hands on, but it makes me think.
Imagine for a sec that all that AI buzz and hype leads to something that is indistinguishable from that extremely intelligent person (however unlikely).
We're not anywhere close to a scenario like that, but at what point is a regular artist that spent their youth listening to Britney Spears violating copyright law, philosophically speaking, when they decide they want to make music like that?
Penny for your thoughts.
- I feel like the important difference between an intelligent human and an AI is that the AI can be replicated easily whereas we are yet to be able to clone a human with their memories. Imagine if we added a restriction to AI training in that only 1 copy of the model can ever be run at a time, like a human. In that situation, all the trouble with copyright would go away.
- On a free Spotify account, there are qds, so you pay for the service and listening experience with your time. Obviously advertisers have no interest in advertising services to an AI with no money. So there is no legal way to access the content, like a person would.
  
  That's a whole other can of worms obviously, but I get your point. What about a company subscription to Spotify or some such?

How is this a violation of copyright law but not outright theft?
- Find the door.