Hacker News Reader

Comment by Janicc | Hacker News Reader

It's going to be interesting to see how this is going to get enforced in the likely future in which AI art becomes in every way indistinguishable from human art.

View on HN

They can just focus on obvious cases: people that upload new images at an astounding rate, people who forget to remove the AI watermarks, and people who simply admit to using AI. No need to turn it into a witch hunt where you accuse everyone of using AI.

It will 100% turn into a constant witch hunt - I can absolutely promise it.

Want an easy reference? See First Person Shooter games.

Turns out that once the computer can mostly do the job, you will see:

1. Constant use of the computer for that job

2. Constant accusations of using the computer in places that restrict it.

See basically every online FPS community that exists. And that's with incredibly invasive software running on player machines explicitly to stop them from using the computer to play - and People both still do it (in HUGE numbers), and the communities constantly bitch about the "rampant cheating".

One way to take down someone you don't like will be to accuse them of using AI. It's an accusation that's impossible to defend against; you could require people to provide a video showing them creating the art, but that adds enough friction to be the death of a community and down the line will itself be something that could be generated.

In the future we'll hire blade runners to find and destroy AI art

Someone needs to make an AI art detector and name it Voight-Kampff. Ironically, it would be a great application for machine learning.

Wholly appropriate since Blade Runner 2049 had Replicants hunting other Replicants.

You need to go back and watch the first one again.

This would turn into an arms race. Art generators would include these detectors to reject anything that gets flagged as generated art

This is genius and if I ever get into this field I am 100% stealing it.

They're just tilting at windmills. It's already difficult enough to catch cheaters in chess, where there are, at most, a few dozen candidate moves in any position. This is a space with infinite candidate moves.

Best case scenario is that they catch a few people who have no idea how to create art otherwise. But, short of an exam-like proctoring system where art can only be created under complete supervision, I doubt there's anyone in the whole world who can reliably distinguish a creation that is 90% AI / 10% human from 100% human.

I agree it will be hard or impossible to catch cheaters. But you get it backwards: AI generated art probably is more easily detectable. A chess game can be represented in a relatively small number of bytes, while artwork must represented with a much larger number. That means more opportunities to find an unintentional signal of provenance (e.g. artefacts that aren't readily noticeable to humans).

A lot of AI art is already indistinguishable. Mid Journey is especially artsy.

until ai actually starts procedurally creating images with bespoke tools rather than blending pixels it will in fact be fairly easy to distinguish a made image from a generated one for the same reasons you can usually tell if someone painted over 3d or used a photo base. ai has a lot of dalle-isms and artifacts that i cannot see going away any time soon. on top of that you could simply require posted images to come with a process timelapse. heavypaint and a few other pieces of software already store those in the png data.

Reply on news.ycombinator.com