Does AI Steal?

9 August, 2024. 1,061 words.

There’s a lot of good posts about AI floating around in the Blaugustosphere.

There are probably more but I’ve started to lose them all in my RSS reader. Sorry if I missed any.

I’ll try to stick to a new topic so this doesn’t get too long. I’ve been futzing with a “Everything I Think About AI” draft for about a year now and that one’s never going to be finished, and if it is, it’ll be so long it might as well be an ebook.

In broadly general terms, I see AI as a very useful tool in my everyday life. Mainly in the area of “personal assistant” types of tasks, and in software development. I’ve taken the time to test the limits of what AI can and can’t do, and it’s impossible for me to deny it has value. I personally can’t wait for the day I can get ChatGPT to navigate the impossible world of making doctor’s appointments for me.

As for Blaugust, I think it’s probably against the spirit of Blaugust to allow AI-generated blogs to participate. But I personally don’t care that much. If we disallow openly AI-generated blogs, we’re probably just opening the door for secret AI-generated blogs. It’s genuinely hard to tell which is which these days.

All that aside, the main thing I want to talk about is one particular thing that comes up a lot in AI discussions that I want to push back on a little bit.

There’s a lot of controversy about generative AI stealing the work of artists for its training data. The general line of thought is that using any AI-generated artwork is unethical because AI steals from artists.

In the absence of any compelling new evidence, I have to put that into a category of misinformation.

It’s very difficult to explain why I think that without getting into a lot of technical mumbo-jumbo, though. AI is super complicated, and I want to keep this simple. I completely understand why it’s easier to just boil it down to a simple “AI steals” or “AI is unethical.”

But broadly speaking, based on my reading of how AI works, it’s a misconception to say that AI steals, copies, or plagerizes. What AI does is consume, in roughly the same way that we do, by reading text, or looking at pictures, or listening to sounds.

It’s true that we have to trust OpenAI and the others when they say they’re not violating any licenses gathering their training data. We don’t have much choice. There’s plenty of material out there for computers to gather freely, though.

It may or may not be infringing someone’s rights to consume a piece of art, be it writing, or a picture, or a song. It depends on what it is, and how it’s presented. But, broadly speaking, if a human is allowed to view a piece of art, a computer probably is, too.

The frightening thing about AI is not that it can consume things, but how much and how fast AI consumes things.

The legal issue at stake here is what’s done with the data gleaned from a computer viewing something. In the case of AI-generated images, for example, AI is looking at a large set of pictures and “taking notes” on what it sees.

There’s no infringement taking place if, for example, I read a book and write down some notes about what I read. Every wiki would have to be taken down if there was.

It’s roughly the same with AI. It’s just that its notes are billions of lines long. The scale of it is what’s different.

Unfortunately the courts are far, far behind on this issue, so it’s probably going to remain a gray area for the forseeable future. (Certainly don’t expect anything to happen here in the U.S., where we remain fixed in time with a permanant partisan gridlock on every little thing.)

There are mountains of misinformation and misunderstandings out there in the world about copyrights and the legal rights of artists. And it varies from country to country. Adding AI into the mix makes it even more complicated.

But I feel like blanket statements like “AI steals” doesn’t come anywhere near the actual truth of what’s happening, and it just creates a lot of myths and misunderstandings that nobody needs.

I say all this as a long-winded way of saying that I’ve thought a lot about using AI-generated header images on my blog, and I know some people might not like it, but I’ve come to the conclusion that it’s well within the boundaries of my sense of ethics.

I might see if I can add some way to make it more obvious which images are AI-generated and which aren’t, maybe with some css. That’s where the ethical boundary is, in my opinion. When you intentionally or accidentally trick someone into thinking that a work is human-made when it isn’t.

Finally, I should add the caveat that I’m no expert on AI, and my view on it is constantly evolving as I learn new things. So it’s entirely possible I might reverse course someday. Who knows?

P.S. I asked ChatGPT and it also said it doesn’t steal artists’ work. Surely a massive corporation making billions of dollars wouldn’t lie about that!

P.P.S. It seemed counterproductive to use an AI-generated image on a post talking about whether AI-generated images steal from artists, so I used a screenshot from Bendy and the Dark Revival instead. It might be worth pondering that we typically don’t ask permission to use the artwork from games when we post screenshots, either.

UPDATE: Here’s some more AI-related Blaugust posts:

Archived Comments

Roger Edwards 2024-08-09T17:42:18Z I appreciate that the truth is more nuanced that “AI steal”. My concern is AI learn from what they scrape, which then leads onto the old mantra of “garbage in, garbage out”. Thus there needs to be some sort of quality control with regard to what material an AI learns from. Just like school.

Endgame Viable 2024-08-09T18:14:20Z Yeah I agree with that. I think that’s one of the things our AI-company-overlords are grappling with at the moment actually. The quality of their AI products directly relates to the quality of their input datasets.

Jussi Laasonen 2024-08-09T18:34:16Z I sometimes post AI generated images on my blog and made a specific „MJ“ (I currently use Midjourney) watermark to separate them from other pictures.

Bhag Puss 2024-08-09T19:56:15Z

That’s one of the more cogent explanations of how AI uses the data it scans that I’ve seen to date. I think the issue people who realise the AIs don’t simply search a huge database to find something to copy when you ask them for a picture are having is that they’re just straight-up not happy about non-humans imitating human behavior. It just feels as though there must be something wrong with it and if it isn’t legally wrong then it has to be morally wrong.

The note about bloggers not gaining permission to use screenshots is interesting. I actually spent quite a lot of time researching that once to check whether what I was doing was legal or not, since I do use a lot of screenshots. The answer is it depends, of course, like just about everything involving copyright. I read a bunch of EULAs for games I play to see if it was mentioned and it kind of is, sometimes, but of course EULAs have no innate legal standing until tested by the courts in many jurisdictions, including where I live, and pragmatically it’s highly unlikely any company is going to go to court over some screenshots on a personal blog. They might send you some threatening emails though.

My main issue with AI art is that I found it funny and fascinating and cute a couple of years ago, when it was all weird and unpredictable, but now it’s just too competent and predictable and not as much fun any more. I still use it occasionally but I enjoy messing around with found images, distorting them and warping them to my purpose a lot more - which, of course, is way less legal than using AI pics would be.

Endgame Viable 2024-08-09T21:02:51Z I think screenshots probably fall under “fair use,” but you’re right, I doubt if any game companies are going to bother dealing with it anymore. And depending on where your found images came from and what you’re doing to them, distorting them could fall under “transformative use” and be fine. In the U.S. at least.

Sorry, new comments are disabled on older posts. This helps reduce spam. Active commenting almost always occurs within a day or two of new posts.

Related

Archived Comments