Ask A Genius 805: Artificially Intelligently Produced Art
Author(s): Scott Douglas Jacobsen and Rick Rosner
Publication (Outlet/Website): Ask A Genius
Publication Date (yyyy/mm/dd): 2022/10/18
[Recording Start]
Scott Douglas Jacobsen: I sent you four pieces of art which you have not seen. The subject line to the emails was something like AI or human made? Guess what? It turns out all four are developed by or made by AI, complete originals.
Rick Rosner: AI responding to verbal prompts.
Jacobsen: Text prompts that I put in myself.
Rosner: You type in some words and it goes to town and if you’re a skilled user of these AI art makers, you go through various iterations till you get a final product that strikes you as a place you wanted to arrive at; a nice piece of art.
Jacobsen: Yeah, and I sent you three horses of different types and one topless woman.
Rosner: And I’m sure they’re fucking great. I’ll go look at them and I’ll be very impressed because I did the same thing. I tried one of these things but I’m no good at it yet. I used one of the cheap ones and I don’t know how to iterate it.
Jacobsen: They’re already at version two of some of them. So the second generation has already come along. So they’re higher definition, they’re more accurate.
Rosner: The ones you pay for, you buy tokens and the tokens you spend them on works on and you get a piece of art and it’s like pretty good but then you spend another token to get it to take that piece of art and refine it further and these images are all over Twitter and I assume over all other visual social media people are sending them to each other and going fucking A because they’re very impressive and I was thinking of this art among the delights of tech that will be irresistible in the future.
Jacobsen: But they have this thing like, I sent you the topless woman one because if you put like naked women or breasts or a naked man or whatever or something like that, it’ll develop the image or just before it does it’ll go ‘error occurred’ and so somehow people developing these programs have the idea that they’ve developed an objective smutometer that can measure the amount of smut on something and say you’re not allowed to see that. So I took a screenshot from my phone to show you it can develop them and well and then it tells you error occurred.
Rosner: And then what? It blurs them out?
Jacobsen: No, it just cancels it out.
Rosner: So it was working on a thing and then it hit a certain level of corniness and it cut out.
Jacobsen: No, it shows you full on and then it goes ‘oh no’ and then it stops.
Rosner: Okay so people on Twitter were talking about how to dance around such prohibitions and apparently there is like that system that doesn’t want to show you what it doesn’t want to show you. People who are good at making the verbal prompts say that “Just add two words to your prompt” The prompt that got you a troublesome image, just add a couple meaningless words to it and it’ll give you the fucking piece of art almost exactly the way it was without the couple other words and somehow you’ve danced around it. It won’t give you like racist images for instance or something. I don’t know there was something that it didn’t want to give somebody and that somebody found a workaround. And then obviously what’s going to happen is somebody will develop one of these art makers specifically for porn and that porn will be some of the porn-iest porn that humanity has ever seen once people learn how to wield that tool.
Jacobsen: They have apps in some of these newer generations of them. You can upgrade and make videos.
Rosner: I’ve seen little snippets of that stuff too. All this stuff is not irresistible the way Tik-Tok is irresistible because it’s fucking art and people couldn’t give a shit sometimes, but some version of AI make like porn for instance, the porn that will be made with AI iterate tools like this will be irresistible. Among the naked picture stuff I look at are some cartoon images. I know like Adam Carolla would say I’m a fucking baby for jacking off to cartoons. Corolla likes jacking off to busty ladies in their 30s, that’s just his deal but he kind of makes fun of people who have immature tastes like he calls it the yummy phase; we’ve talked about it that people who never outgrew liking stuff that’s sweet, never learned to like the taste of beer or whiskey or cigars and I assume that Corolla’s tasted porn, he’d consider part of his mature developed tastes maybe. So yeah I’ve been known to look at muddy cartoons and there are some of them that go under the header like 3D cartoons in a space.
Its art that’s as made as realistic looking with the flesh, body shapes, and contours as it can be made. It’s one of those things that were no good, just in the uncanny valley 10 years ago. This stuff is too creepy to like and then eight years ago it was it was okay and now the images as you’re seeing are pretty freaking realistic and we’ll only get more so.
Jacobsen: Yes, I mean they work from Pixel up. So any next generation with more computing power and more sophisticated sort of neural nets simply be better more robust or more accurate especially around really nuanced things like hands and faces.
Rosner: Yeah. My kid is an expert on 17th century needlework. So I tried and my prompt was robot family in the style of a 17th century needlework sampler. Robot family was fine. Obviously a lot of people from the AI is very familiar with robots and families and maybe even robot families but the needlework was for shit. It didn’t look 17th century; it looked like very primitive like a child’s a needle point kit from the 1970s; primary colors and blocky shapes and no subtlety. So obviously the AI, it has less of a database for embroidery or for samplers.
Anyway, once the AI gets a big enough database for all these different things like for hands, like six years ago I started messing around with an app called This Is Not A Person, which is an AI that continually uses machine learning to get better and better when I first started making faces of people who weren’t real on that thing, it had trouble with teeth it had trouble with earlobes, it had trouble with earrings and it had trouble with glasses. It put glasses on a lot of people but the glasses often didn’t make sense, they’d have like parts that just kind of ended nowhere and if they were real glasses they’d fall apart because they had parts that weren’t connected. I haven’t looked at it in a few months. Last time I looked it was fine on teeth, I think fine on glasses, fine on earlobes, still might be fucking up earrings and still has a big problem with backgrounds probably because backgrounds could be fucking anything. They don’t have to be part of a body or something. Just anybody can stand in front of anything. So the AI still does shitty backgrounds.
But with hands or with anything else you could talk about in art, once its database gets large enough and its interpretation of the database gets sophisticated enough, then you’ll get some really fucking great looking hands, like once it understands the relationship and understands is not the right word. I don’t know because that implies consciousness and this shit isn’t conscious yet but once it gets how hands work, how the flesh drapes, how the tendons work under the skin, how blood pumping through flesh gives makes it pinkish except where maybe it’s veins, etc. All these different things that we kind of intuitively know after a lifetime of looking at hands, we’ve developed our own algorithms and understandings for what makes a realistic hand without even knowing what our understanding of hands really consists of. AI will do that same shit and you’ll get really good looking hands and anything else. Once the AI sees enough samplers; 3000 of them or 15 000 of them, it’ll start giving, I don’t know how it like gets those images, and it’ll be a little tougher for shit like 17th century samplers because I’m not sure that there are 15000 of them remaining in the world. It’ll be part of its net of knowledge and it’ll start doing it halfway decent job, maybe not as good a job as ‘make me a car’ because there’s just more reference material.
[Recording End]
License
In-Sight Publishing by Scott Douglas Jacobsen is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License. Based on a work at www.in-sightpublishing.com.
Copyright
© Scott Douglas Jacobsen and In-Sight Publishing 2012-Present. Unauthorized use and/or duplication of this material without express and written permission from this site’s author and/or owner is strictly prohibited. Excerpts and links may be used, provided that full and clear credit is given to Scott Douglas Jacobsen and In-Sight Publishing with appropriate and specific direction to the original content. All interviewees and authors co-copyright their material and may disseminate for their independent purposes.
In-Sight Publishing 