The Fair Use defense lies in the fact that AI doesn't make copies of anything - it's not competing on a one to one basis, which is why we have copyright in the first place - to protect IPs from counterfeits that might negatively affect the original's profitability or place in the market. This is why style and genre can't be copyrighted.

The reality is that copyright doesn't even really apply to this situation, at least not in its current form.

Also, copyright was never intended to prevent general competition, in fact, I think the intention was just the opposite.

___

Everyone thought functional AGI was so far down the road that only academic science types were even bothering with it. They were a bunch of well-paid nerdy folks, not ruthless hedge fund managers bent on world domination. So, yeah, I think their intentions are good.

The natural fallout from AI is up to the government to resolve. We're going to have a post labor society, which will turn everything upside down. We can't blame these technological visionaries for Washington's inability to get out of its own way. For example, AI isn't the one letting our infrastructure crumble away before our very eyes. Ultimately, we have to take responsibility for this mess, we elected them.

As for "Fair Use," U.S. Copyright law considers the following:

1) Purpose and Nature of the Use
Educational and non-commercial uses are more likely to be considered as "fair use" than commercial uses.  Transformative uses may also be considered fair use.  Commercial AI is likely to fail on the former.  As for that latter, is using the content of books to create more books really transformative?  Seems unlikely.

2) Nature of the Copyrighted Work
The use of creative works is less likely to be considered fair use than the use of factual works.  AI being "trained" on creative works to create simulated creative works is seems likely to fail here as well.

3) Amount of the Copyrighted Work Used
Well, they "trained" AI using whole books so, again, fail.

4) Effect of the Use on the Potential Market or Value of the Copyrighted Work
Since AI is being used to create books that would compete with the original works it was "trained" on, it would seem likely to fail here as well.  Additionally, AI has the potential to flood the market which would increase competition for the original works as well as reduce their commercial value.  That would also tend to make it fail the fair use test.

All four of those factors are weighed.  Even if you want to argue that AI is transformative, it still fails under the other three factors.  Now, sometimes courts have argued that using an entire work can be fair use, so even if AI passes that, it still fails on the remaining two.

Quote

I think the intent of these AI companies is generally good - they're trying to create a post-scarcity semi-utopian society, like Star Trek.

You seriously believe this? How many decades is it going to take before we get that awesome fake Star Trek future?

What most people forget is that Star Trek future followed a third world war that left hundred of millions dead.

And then didn't happen with a first contact situation first.

 

I'd like to believe in the semi-utopian society, but as I've said, I have a hard time seeing how that's the natural fruit of corporate greed.

I remember the anecdotes, but not specific data, including the prompts that lead to the infringing text. AI doesn't "know" the text to be copied (it doesn't store anything), it only knows the likelihood of the next word in the series or the next pixel. The Getty logo in AI images only proved that Getty owns a lot of soccer images1. These artifact abstractions show up in a lot of AI images. One weird one was rulers showing up next to moles. This is because most images of moles are medical, and the datasets often show a ruler to document the change in size. That's a learned association.

I've said this before, but I don't really care exactly how AI works. Somehow, it is trained on copyrighted material with the intent of ultimately producing competing products. Yeah, courts might find that fair use, but that's by no means a done deal. I'd argue that's a ridiculously broad definition of fair use. The fact that many companies moved of their own accord toward consent and/or compensation models suggests that the very least that the business world isn't certain about the fair use issue.

And yes, AI doesn't know the text to be copied. That's precisely part of the problem. It also doesn't know the difference between fact and fiction, which is why it makes up fake legal cases in legal briefs and fake bios (including statements that could be considered defamation). It's the intent of the developers that matters here, because AI isn't capable of forming an intent.