Hacker News Reader

Comment by Max-q | Hacker News Reader

If you go to art school, what you do is study and analyze what artists have done before you. We humans learn much in the same way as an AI. We study what has been made by others.

I think accusing AI generated work to theft, is a based on the wrong assumptions.

It's not just art. What do you do in computer science courses? Or math, language, physics, or any other subject? You study what has already been made. What you produce based on that knowledge is not considered theft.

Have you ever seen an AI image that, if a human made it, would be considered a theft?

View on HN

Human art can exist without anything prior. For example, I highly doubt the creators of cave paintings traveled the world to see other such work before doing theirs. But you could look at impressionism or pointillism for that matter.

AI can not.

Indeed, the problem is the utter unoriginality of the current wave of AI. Literally the only thing it can do is to produce the next token which is the most likely to be in the training set.

I doubt the creators of cave paintings stayed their entire lives in that cave. One presumes they saw animals.

Could an agentic AI generate a cave painting if you restricted it to training data of animals and also limited its ability to painting with primitive tools? Sure, why not?

What is this ‘agentic AI’ you speak of? There’s no such thing to date. There is no AI yet that can train on pictures of animals and then decide to produce cave paintings. If you limit the output to be produced only with primitive tools, then the AI is not agentic, you’re just making a fixed-function paint machine.

There is no such thing, but cave paintings are also not relevant to what people do with diffusion AI.

My argument is that art that diverges from reality is largely a function of path dependence and tooling more so than creative choice.

>If you go to art school, what you do is study and analyze what artists have done before you.

You don't have to go to art school to produce art. Even if you do go, there's nothing stopping you from producing art unlike what you studied.

>We humans learn much in the same way as an AI. We study what has been made by others. I think accusing AI generated work to theft, is a based on the wrong assumptions.

What assumptions?

>It's not just art. What do you do in computer science courses? Or math, language, physics, or any other subject? You study what has already been made.

Most of "what has already been made" is not "copyrighted work for which you have no license or exception"

>What you produce based on that knowledge is not considered theft.

It certainly can be.

>Have you ever seen an AI image that, if a human made it, would be considered a theft?

I've certainly seen AI produce things that would be considered copyright violation or misappropriation of likeness or similar offences if they were to be used commercially.

> You don't have to go to art school to produce art. Even if you do go, there's nothing stopping you from producing art unlike what you studied.

Sure, but also you do, presumably, have to see to paint. The overwhelming majority of our training corpus is just visual observation.

What are you trying to suggest is the significance of that?

The point is that nobody, not humans, not AI, develops painting skills in the absence of image data. Humans can learn to paint with less input images than AI, sure, we're a lot more sample efficient, but we still need samples. The processes are not that different.

Okay, but that's not really relevant. The whole issue is about the way copyrighted works are used.

Reply on news.ycombinator.com