Jonathan Blow on AI art and tech

[Gold-kybq]

I see his point that the very best of the best humans will still continue to create for a long time.

However, I see a mass homogenization of all popular media in the future. Similar to how 99% of Unity/Unreal games look the same, if you provide people with an out-of-the-box easy solution, they won't bother to come up with their own unique solution. This is something that Blow talks about a lot with using frameworks.

Also, every master has to go through a journeyman phase. AI replacing the journeyman could potentially prevent a lot of human potential in the future. I don't have to worry about that thankfully but worry about future generations.

[SaHaRaSquad]

The problem is that right now you'd have to review the AI generated code to make sure it's not complete nonsense, which means you did nothing other than changing your work from writing code to reviewing code, which is more tedious. Not to mention in many cases knowing what the code is supposed to do is the actual work, so those AI systems right now are just a better intellisense in the best case scenario. Which is fine but it's not what many hyped people expect.

[ArksideGames]

The problem with AI isn't the jobs, the problem is copyright theft, blatant infringement on rights.

[stumbling]

AI has an "outrunning the bear" problem. It doesn't need to be a general intelligence, it just has to convince enough humans that it's a general intelligence.

[UPMidget]

Jon will change his mind on this once the AI starts making puzzle games

[jakecreighton]

Also a human artist doesn’t necessarily train himself on other artists’ work. They can see something in the real world and then in their minds imagine a way to paint that, all without needing to train on someone else’s work

[BinaryDood]

The problem is not wether the models scale forever, but of their fast adoption rate in the socioeconomic sphere. If generative LLMs become a baseline dependency anywhere close to the internet in most markets, then I'm afraid we'll generate new sensibilities which would be "okay" with subpar output. And, because it is so much cheaper and faster, we might devalue that combo of effort+creativity to the point of it becoming not worth for anyone to pursue, as it would always hit a brick wall of endless saturation and expectations too quick for any direct human input. It becomes a race to the bottom... The middle class is fucked.

[欺软怕硬]

Generative art depends mostly on computer vision than general intelligence.

[riveteye]

we need hundreds of thousands of jobs for bad artists for the field to be strong enough to profuce a few actually good outliers. Good artists are a product of a system, if the skill ceiling gets raised so high you will only get hired if you're one of the top specialiats in your field, it's doomed.

[narnbrez]

im glad he can say it's more than he expected in his lifetime. i agree! i love to live in this surprising time

[samuelbucher]

Very good take from JB once again. I don't consider myself a good programmer, but I can't imagine myself using AI to help with anything beyond small and/or tedious tasks. If I were to order it to make a whole program, whatever it may be, I wouldn't be able to trust to not fuck something(s) up somewhere, and debugging that would not be a pleasant experience. This is by no means a diss at modern LLMs, as the fact that they are capable of doing even that just from a prompt in the first place is truly a miracles of science .

[..]

My wife is a digital artist. She still has more commission work than she can handle. AI just isn't on her level and can't do the stuff she does (AI doesn't understand layers, which are necessary for some work such as high-level vtuber models that have 100+ animated layers). It does, however, help with references. Think of how Pinterest works. That is what can be automated nicely for artists. Tons of references which are good images, but not consistent with what you asked (and full of small problems that would prevent them from being usable as-is).

[jonahbranch]

The problem isn't artists losing their jobs, the problem is why we still need jobs to survive in an increasingly automated society. We should be excited about AI replacing human labor; only in capitalism is the loss of jobs a death sentence for everyone who gets replaced. We can't keep "making new jobs" forever once AGI and advanced robotics are on earth. Humans will simply be unemployable in every industry. Under capitalism, that means 99% of the population starving to death.

Think what you will about capitalism, but it is quickly becoming outdated.

[oraz.]

AI art has an abject banal mad magazine aesthetic to it. That's the problem they have a latent quality common to everything they output and it's subtly off-putting. If people saturate media with those images it's not good. Using chat gpt to quickly look up documentation though is useful.

[Elrog]

Any idea why Jon has stopped putting out stream vods? Was it to do with people uploading clips?

[shawnmuench]

I always enjoy Jon's thinking

But why care about computer stuff while we let human life languish- bad politics, bad culture, pointless consumerism etc

[JomAnimatie]

As a artist I kind of agree about the notion of work, but at the same time its hard to not notice the stark contrast between. Oh AI wont be able to program in a long time while also saying, I did not expect to see AI art to be so good in my own lifetime.

[rodrigopetunio]

Well, here's the thing: he's a professional programmer, and Jonathan correctly states that AI cannot solve programing. He does this because he is knowledgeable about the topic.

But Jon is not an artist, yet he confidently states that we shouldn't listen to Artists, then proceeds with a weird philosopher king analogy involving horse trainers. He then callously states that he doesn't care if Artists (or whatever he refers to as "low skill" and "mediocre") don't have jobs. And as before, in a topic he knows nothing about, he again proceeds with another weird analogy, this time about coal miners.

It is fine to want to have an authoritative opinion on most things, but the problem is that he painfully doesn't know about what he is talking about in some topics, yet he proceeds with an uninformed opinion anyways. Even though all things considered, his first answer applies pretty well to both Art and programming.

[Jorbz]

People are going to have to realize that, even outside of AI, many of the job markets will be unreliable in the future. Technology changes the jobs available, and that has always been true. The problem is that we've tied people's ability to consume too tightly to their ability to produce something unique, so when a machine can produce what a person can, suddenly the person has lost the ability to consume. You COULD control how people make art, but then you're creating a convoluted system to artificially keep employment up. You're also limiting the creative output of what a writer can make with a computer, which might be violating the creative rights of the writer.
Perhaps the solution to these employment problems will eventually be the government paying for continuing education for adults that suddenly need to switch jobs. Of course people may also need a UBI if they have a family and need to go back to school. The whole thing is very complicated. But again, it's probably foolish to try to hold back technology for the purpose of keeping jobs. You are more likely to pass legislation for a stronger social safety net than you are to pass legislation limiting what a computer can observe.
And yes, people need more empathy for anyone feeling that their jobs could be replaced by a machine, but "is it good that this happens" is a separate discussion from "will this happen".

[MadsterV]

Not hypothetical, I've used AI to code. It helped me get started in a project. I ended up rewriting everything it did and taking a different approach. I would have been probably faster without it but I was stuck, so it did help.
People are way too scared. It's just a tool. A bit eerie, but the sky isn't falling.