I recently came across a tweet on Twitter (I’m not calling it “X”) which claimed that the AI art crowd was being dishonest.

My first gut instinct was that the person tweeting this was the one being dishonest. This was, after all, a straw-man argument. I even made a reply tweet. However I’ve said in the past that you should never draw a conclusion based on just one piece of evidence (I know, I know, I should have thought it out more before responding). There’s always another explanation when you only have one piece of evidence. So what’s the alternative explanation here?

I think it’s possible that this person doesn’t quite understand some of the arguments that the AI art crowd are making. So let’s dive into exactly how programmers think, why so many of them think this isn’t copyright infringement, and what kinds of misunderstandings people might have.

First let’s start with the basics. I made a blog post on the main blog here: https://noahs-blog.net/?p=422 about the basics of computers, but to make a long story short: computers are machines that follow basic instructions which tell them how to think in any way that’s possible.

When programmers think about programming, we think in terms of how we’re shaping an artificial mind to think. That’s why so many people use the argument “It’s mimicking the human brain”. To them it sounds like a convincing argument because not only is this just a machine thinking, it’s even thinking in a way that’s kind of like how humans think.

You wouldn’t call it copyright infringement for someone to look at and remember what a piece of art looks like, would you? Of course not. That would be insane.That’s absolutely comparable to what the computer is doing. In order for it to imagine Pikachu caught in a mouse trap, it first needs to know what Pikachu, and mouse traps look like, and it needs to have some idea of what it means to be caught in a mouse trap.

The only way to do that at this level of technology is to show it a huge number of images with accurate descriptions of what’s in those images so that the machine can learn the pattern. That way it can learn what Pikachu looks like (from images with Pikachu in them), what mouse traps look like (from pictures with mouse traps in them) and what it means for something to be caught in a mouse trap (from images whose descriptions say that something was caught in said trap).

In my opinion computers should have to follow the exact same rules that humans do. We shouldn’t have a separate set of rules for bots because bots are thinking systems much like humans.

Nobody that I’m aware of is claiming that publicly available is the same thing as public domain. The idea is that it’s okay for an AI to look at images and learn the patterns behind them because that’s what humans do all the time. Why shouldn’t machines be allowed to do that as well?

While artists tend to care a lot about intellectual property laws (for some reason), programmers tend to dislike things like copyright and patents. That might be in part because of vendor lock-in (which I’ve described here: https://noahs-blog.net/?p=383 ). STEM related stuff tends to be done better when it’s done openly without worrying about things like patents and copyright.

This is possibly why artists freaked out at the thought of automated tools downloading massive amounts of drawings off the Internet and learning from them. To programmers this is far less big of a deal. Nobody on the programming side of this debate sees it as “stealing art” (whatever that’s even supposed to mean). They’re getting data samples so the computer can learn the patterns behind it.

As for those who think that AIs shouldn’t be allowed to learn the pattern behind art pieces that are publicly available: how SHOULD they learn the patterns exactly? It’s not exactly practical to pay a huge number of people to draw all that art for them. At the end of the day, this was really kind of inevitable.