I was reading this article about the NYT’s suit against OpenAI. OpenAI argued that NYT couldn’t sue for damages because it had been “too long” since the infringing started, and since NYT “must have known” that OpenAI was doing it, they lost the privilege of collecting damages (IANAL but I think it’s because the Doctrine of Laches). In any event, the judge sensibly threw this argument out, telling OpenAI they hadn’t demonstrated that NYT could have known the size or scale or timing of the any alleged infringement.
This made me think: now that the cat is out of the bag and everyone DOES know that everything on the Internet (and beyond) is being fed into AI factories, do we as creators have an obligation to somehow collectively sue LLM makers so that laches can’t be used as a defense in the future?
I feel like they need a test case to figure out how to define derivative work when the creator is not human.
If i make a painting and you see it and then make one in a similar style it would be considered derivative and not a violation. In your head is a distillation of my image. It doesn’t contain the image and your output would be lossy. Similarly the LLM contains statistics and not verbatim content. So the question is “how is human synthesis different than AI synthesis.”
Until that is resolved a class action would probably fall apart. Individual damages would need to be determined and even a single example of “you put your stuff out to the public and aren’t going aftet Joe who made derivative work…” would derail the case.
IMO it’s time for a reckoning of what’s systematic/automated vs what’s not.
For example, “no expectation of privacy in public” meant you should be okay with appearing in someone else’s (manual) photo while out in public. However, I don’t think that should extend to persistent systematic surveillance, e.g. suppose every Tesla’s camera captures were combined with person recognition systems and tracking.
Just because something is theoretically okay at a small scale doesn’t mean the same applies at large scales.
Another example: Society funds public roads via government taxes for personal use and for regulated commercial use. Uber systematically consumes public road space under the guise of personal use vehicles, for commercial use.
But then I’d ask how do you outlaw human systematic consumption of information. The camera on my car cant watch 24/7, then why should YOU be allowed to watch 24/7? What you’re outlawing is the literal methodology.
This has always been an issue with my thoughts on AI. If the computer became sentient does the LLM learning rule go out the window? or is it because they are made of metal?
