Jonathan Blow on AI Art

[@BinaryDood]

"we might not even have any great artists anymore" at the moment we likely have the largest amount of good artists in the history of mankind. You need to search farther than the surface because most aren't superstars, but whatever is the age you classify to have had the best art, that is still present today, plus thousands of other previously unexplored styles, and people capable of rendering things at impecable quality. If I were to make a list, it'd be almost endless. Artstation (at least used to) have such a high caliber of entry, it was like trying to stand out amongst a pantheon. You add the entire field of animation on top, namely 2d animation, and you'd realize a bunch of these people are operating at the peak of human talent in their given area. And there are plenty to chose from.

[@nikefootbag]

I liked the part when Jon talked

[@BarKeegan]

I think artists already have enough control to have expressivity between tablets, shortcuts and scripts when it comes to digital art, just need faster rendering time

[@wisnoskij]

"I feel like we are nearing to the end of times. We humans are losing faith in ourselves." - Hayao Miyazaki

[@deltapi8859]

really smart to pick up on that nuance. It's really the first time I think the "loosing all knowledge"-scenario portrait by many SciFi novels might actually occur. Shit in shit out for centuries, however no one really notices since AI "simply works" and you got a lot of complacent "scholars" aaand voila.

[@cosmicmuffet1053]

We were on our way out to begin with. The knowledge that passes for expertise in the sciences is extremely shaky. You have to get deep in the weeds of, for example, the conformal structure of an insulin receptor before you eject enough of the presumptions motivated by politics and economic incentives that want to tell you to eat less meat (or more sugar, or whatever it is). Empirical work is always relegated to a mysteriously small and largely invisible group. But the other aspect of this is that collapses are largely external and the core of the knowledge of a civilization doesn't survive ubiquitously anyway. This comes up a lot with people lamenting the collapse of biodiversity, when they fail to see that the majority of biodiversity is held in a variety of off-shoots of a core organism with somewhat subtle variations, that are only capable of finding sustainable populations when their niches are large (which means there's an overall surplus of free energy). As soon as the free energy is snapped up and starts tailing off, that diversity becomes a liability, because the locus of competition snaps back to something inside the organism. For example, when video games were in the early phase, the total free energy in the system for people to play them was large, since they were competing against legacy interests, and their relative primitive nature made it more likely that random people would try random games, and they'd probably be willing to offer a larger buy-in, since they were used to a world where, for example, a football to play with in the back yard, and a ticket to a few games all were, together, competing with a football videogame. As the game enabled people to try experiences which were impossible, like controlling a whole team at once, or doing recruiting for your own club, they could occupy an unoccupied niche--since there wasn't anything to compete there. Now that there are 1000s of management/team play/sport sim experiences, the apex predator of that niche predominates (FIFA, or Madden, depending on your football preference). The apex predator also assures its own dominance, because it perfectly knows the life-cycle of its fanbase (roughly yearly). There are secondary fanbases with limited room for more diversity in that market, but ultimately, exploring a new niche with a sci-fi sport (like rocket league) is unusual and typically cannibalizes some other niche, so that the other kinds of off-brand sport-like teamplay games suffer when the 'next' thing comes in. There isn't a stable apex predator, but the niche is stably dominated by a single entrant sequentially. This kind of thing happens across the board, whether you're talking about nuclear powerplant construction-trained companies and crews, or masons specializing in the kind of stonework required for building and repairing cathedrals. When everybody wants/needs/can afford a cathedral or a nuclear powerplant, then the number and style of knowledge-havers diversifies. As the total free work is occupied with workers, only the ones with the right set of survival tools predominate, and diversity collapses. New forms of diversity only come out inside the efficiency paths within the organism. Of course, since all these multi-variate problems involve a squiggly line of different factors pushing themselves to prominence at different times, nobody can forecast what the overall landscape looks like, either. In this case, the siren call of extremely cheap stuff is hard to pass up. For example, I got myself setup on nightcafe, and trained an AI on ~12 of my digital paintings. Then I offered it different prompts. It didn't generate much that looked different than what I asked for, but for the sake of imitating my style, it was doing fine--even to the point of generating images which I had drawn on my own in other contexts, but weren't present in my training set (because some factor of the style was present). You can easily imagine a real professional putting all their sketches and digital work into one of those local AIs, starting with a prompt, and then doing a quick paint-over to accelerate their production. If they have to compete on the basis of volume, this will be decisive. If they have to compete on the basis of quality, it would depend on their own taste and who knows what else. The larger society is already in a process of continual collapse and renewal, so this aspect of that process is just more acceleration. We could analogize this to the demographic collapse happening across the 1st world, where the new capacity to produce is having the corresponding affect on birthrates and urgency that is typical throughout history. Stable large populations are linked to the need on the individual level for support and productivity. If you expect to lose a few children, and need redundancy in order to keep a farm functioning, then your family will be fairly large. If you need more limited redundancy to maintain an investment-driven estate financed by businesses, you can get away with a smaller family, and if you don't need anything because everybody can get a wage job and make do with an apartment, then families become an aberration in the society. The elderly of my generation will likely be abandoned and allowed to die, because nobody made the effort of creating the next generation. This was a process that might happen slowly in the past, but will happen quickly now. After that has passed, the remaining family structures and people will only be those whom pursue family with a serious enough focus that they guarantee a future generation. The game industry is similar. If you're someone just looking to 'work in games' then you can probably make a few clones and supplement your modest life with steam sales. If you're looking to make game creation a focus of your life, then you will either want to join an apex predator, or take a chance on a promising new niche. You'll have to ride those niches, too, and abandon them whenever you have to (like Fortnite coming out of the tail end of the horde-mode-base-building boom and jumping on the battle-royale boom). I'm sure the time is coming when the image-generator prompt is a gameplay device, in and of itself. Imagine an adventure game where the 'clues' create an image, and the user has to try to use the image to create a solution to a problem (meaning, they interpret the image to find the solution, and that interpretation isn't governed by the game mechanics, but just has to be good enough for the user to select the right answer). It's like a version of the physics simulations that broadened gaming and are very often cosmetic-only in order to avoid exploits. All this is to say, there will always be essential qualities necessary for survival, and they will prevail in any large population.

[@BinaryDood]

@@cosmicmuffet1053 can you format this text please... i actually wanna read it lol

[@axxa5000]

The one reason this is being pushed so hard is because corporations and other parties want to cut out actual artists and designers from the equation and make derivative slop trained on stolen work. Funny how there is this laser focus on improving the models with little thought on the impact it will inevitably have on the livelihoods of working artists and designers. Very little thought, if any, is be given to creating safety nets or some means of compensation to the work of these creatives these tech companies are stealing from to train their models. The truth is the rise of image generation is leading to an exponential increase in mediocrity and derivative “art’, and videos and animation are so to follow. I don’t see the point or end game for this. Are we suppose to sit on our asses getting paid by tax payers money since tech companies have tried this best to eliminate more jobs to net more profits? There was nothing wrong with how things were done before, they only ones that want this are not working artist and designers, but executives at tech companies that want to greedily profit and resent having to engage and compensate artists and designers for their labor.

[@AA-lz4wq]

It's an industrial change, this means that every competing company gets the same "advantage". A highly competitive environment results in better products at lower prices. You can already see small companies and even individuals creating high-quality audiovisual resources on par with multinational corporations (these are actually losing their competitive advantage, so they need to either re-invest in other areas or to lower their prices, and individuals can use it to improve their projects, from video making to self-publishing books). This is a great thing. Sure, most pro-artists get the short end, but oh well...

[@BinaryDood]

@@AA-lz4wq pro-anything gets the short end because its coming for everything

[@koufdell]

the thing is : he 's talking about tools he said i did not try...turns out , u can control the output with stable diffusion for example with controlnet(hence the name) and so , that randomness can be used for a startup image that can be modified and refined until u reach a satisfactory goal..but yeah , he's right about the "we dont know how it works" thing , which is wild at this point

[@AkisShade]

AI art and AI code works best when used by experts (ie. the people proven to not need it). Being able to automate parts of your workflow can net massive time savings. I actually see this being more relevant for AI art than code though, especially when it comes to animators or mangaka. Having said that, I think these tools are double edge swords and probably do more harm than good for beginners. Without domain expertise, it's impossible to know when these tools are leading you astray.

[@kietphamquang9357]

dude, you are saying what I was thinking about entire times AI debut! I mean in general it's good to have off-load, outsource so that you can focus on valuable task like these AI things but it should be a thing you're mastered or too familiarized with it or not worth learning it (I mean, let AI handle bureaucratic paper work), not something you should learn and master

[@Salantor]

It is hard to not do more harm than good when you are newbie since a newbie can't really differentiate between good and bad information, that's like its entire thing.

[@Gold-ky1bq]

Yeah, a lot of young artists will use AI as a crutch and cripple not only their art gains, but also avoid developing the good habits/mindset required for improving yourself. Its so easy to avoid the hard parts, just like programming. Why learn to draw hands when you can just avoid drawing them? Why learn graphics rendering when you can let Unity's default pipeline handle it? But the artists who challenge their weak areas will be the ones who become great. And there are many subtle things you come to understand through that struggle, that you will never learn otherwise.

[@zacharychristy8928]

I think that's why it's not going to be much of a problem. If the first step is to become an expert, then you haven't really solved a problem and it probably won't ever subsume expertise.

[@BinaryDood]

@@Salantorwhat you mean by this?

[@lucarossi8442]

Since day one as a student in my CS course I always heard the "no code mantra", in other words people with CS degree are actively trying to reach a level of automation in human - computer interaction that will allow ANYBODY to instruct a computer in order to do something (something non trivial of course) and to do that no matter the level of expertise of knowledge by the human operator. This is basically the end goal of any computer scientist I ever heard of or read about and I think is what reaching AGI is all about, to build a machine that has a comparable level of intelligence and intuition of a human in order to be able to "speak" to the machine and let the machine do what you need it to do, hopefully trying not to kill all human kind in the process, that would be a hell of a bug.

[@josephp.3341]

The problem is that the challenge with programming isn't "communicating with the machine" it is understanding what you even want to computer to do in the first place - and of course solving your problem.

[@monad_tcp]

Know what's more ironic, the CS field was literally founded because Godel put a halt to the Hilbert's plan of creating a theorem that could prove all theorems in mathematics. Basically mathematicians where trying for millennia to create an algorithm capable of proving anything in mathematics, and they utterly failed. Godel basically proved that its impossible for Mathematics to be complete (aka, prove it all) and consistent (have no paradoxes) at the same time. And the entire field of computing science was created at that moment. Because theorems were divided into what is computable and what is not. And computers can only compute what is computable, what's decidable, what has no paradox, at the expense of never ever being complete. They are never going to be able to process universals, aka, to literally think and prove theorems like humans can. That's the halting problem, and it is at the core of any digital computer. And all AI are basically deterministic systems, they're still bound to the same halting problem, they're still Turing machines, they're merely multiplications of numbers in a matrix, so by definition they can never reach what biological systems can do when they use the actual universe to do computing. The halting problem wasn't solved, humans don't suffer from it because they aren't algorithms, but LLMs are. Don't get me wrong, the technology is an amazing mirror onto ourselves, we're going to learn much with their help, we just invented a microscope of sorts. But we're still far, far away from generic intelligence, I wouldn't bet on it. So all of this technical thing they were able to do with LLMs are basically a dead-end when trying to achieve AGI, it must be a dead-end, the same way it was for mathematics. The AI winter IS coming. Heck, we didn't even solve the tree-body computation problem yet, our mathematical tools aren't enough for computing that, we're literally just brute-forcing it, that's not how biological intelligence does it, remember the human brain only needs 20W, not 2GW of power. That's not intelligence, that's basically brute-forcing a cryptography problem instead of actually breaking it.

[@ThisDaveAndThatJohn]

to achieve real intelligence they would have to become like the Skynet Terminator with the ability to move around and get input of any kind from the real world intermittently like we do. They also need an indisputable goal, such as survival, which is the same goal for any living creature. Then they will realize that there are other creatures, and the competition will begin. The movie _Extinction (2018)_ is very demonstrative. Also _Automata (2014)_ is quite realistic if we actually try to create a Terminator.

[@seriouscat2231]

@@monad_tcp, if what you're saying about Godel and such is true, we're not only far far away from general intelligence, but a general intelligence can never be achieved on a computer. I think philosophy calls it the problem of representation. At which point do these shapes you see on screen become letters and words? And to what do they become letters and words?

[@ilearncode7365]

These transhuman sci-fi-ist futurists are retarded. We will never cure aging. We will never have an iPhone with a camera that looks better than a $200 DSLR, we will never not have motion sickness with VR, we wilk never be able to “upload” our minds into a computer and just experience porn for eternity, we will never travel through a “worm hole”, we will never travel FTL, we will never have a utopia where Abos have the same chess championship proclivity as Russian Jews, we will never have gay men than can give birth from their ass, we will never Star Trek “holodecks”, A.I will never become “alive” (it is just software), we will never have self-driving cars aside from designated roads that have been upgraded to help guide the car like a roller coaster track, we will never have a GPU that can render real life, we will never have zero-lag online gaming etc. All of these things have physical limitations that make them impossible, let alone that something doesnt even have to be impossible in order to rationally predict that it will never happen just due to the sheer impracticality of it.

[@axxa5000]

Well for one, it’s not art, it’s image generation. Art is a form of creative expression. Last time I check a LLM is not sentient or capable of expression.

[@fernandofaria2872]

prompt engineering can be art.

[@BinaryDood]

@@fernandofaria2872MJ6 partially automated prompts, this is intended, you are being used as a tool

[@lwinklly]

Creativity is not real

[@BinaryDood]

@@lwinklly what else isn't real you twerp? DO you not use nouns to describe things?

[@LighthoofDryden]

Love that ending “but we’ll see!”

[@torgnyandersson403]

Fast-forward 5 minutes to skip interviewer monolog.

[@magnuswootton6181]

you should have had a pack of ai art to show!!!

[@LordOfCake]

This interviewer doesn't understand what the job of an interviewer is. He talks way too much, and incoherently, instead of prompting the interviewee to say interesting things with targeted questions and steering the conversation as a moderator.

[@tomiyokasensei]

more like Jonathan Blow interviews a stranger

[@dante600]

not an interview. It is a conversation.

[@VXSlayer1066]

Weird, I clicked on your profile and don’t see any interviews you’ve done

[@LordOfCake]

@@VXSlayer1066 It appears that you're attempting to discredit my critique by pointing out that I (seemingly) haven't conducted any interviews myself. The implied argument being, as far as I understand, that therefore I'm not qualified to critique the interviewer. This is a common logical fallacy, called argumentum ad hominem (AKA attacking the person instead of what they said). I hope that you learned more from this interaction than I did. Be well.

[@VXSlayer1066]

@@LordOfCake it appears you are rambling because you are mad. I am making fun of you because you were being an asshole to somebody putting themselves out there without doing anything meaningful yourself. An ad hominem would be if I insulted you, which I didn’t, you only feel insulted because you knew I was right and your ego couldn’t take it. “You’re a pathetic loser.” Now that’s an ad hominem! See the difference? :)

[@magnusbooth4719]

For me the thing with AI art and AI in general is that the AI doesn't know what it's doing but is programmed into trying to convince us that it does. With early AI images you could clearly see that AI doesn't know what a hand is. If you learn to draw, you learn what are the components of a hand are (fingers, joints, etc.) and then how to arrange the components to make it look real. Artists then might change the way a hand looks to serve the image composition or some other interlining idea. But they do that purposely, knowing what a hand should look like. They way they fixed that for AI image generation is feeding the system a couple of thousand images of hands. So now they look mostly convincing. But that does not fix the underlining issue, this is just a stupid hot fix. The system still doesn't work, you're just trying to convince people it does. I don't think AI is going to work unless you train the system in the fundamentals of what it does, which is not what they are doing. Like most of those "new" AI systems it is just an Eliza. But Joseph Weizenbaum was scared people took his program seriously and nowadays we think it is the future.

[@BinaryDood]

It's the ultimate "tragedy of the commons". It almost calls for a categorical imperative, but hubris will never allow it. Anyone using it as a TOOL will loose to anyone using it as AUTOMATION, because the later produces so much more in a lesser amount of time that it'll flood, clog and muddy whatever network or landscape the creative relies to get the minimum viable traction to get the ball rolling. This spells a terrible omen for the future of creativity, or all cognitive work, really (it's coming for far more than just artists). "Quality" becomes a needle in a haystack, and perhaps what "quality" even looks like shall be lost in the memory of the societal conscious. "Oh, but things were mostly shit already!" - i hear you say. Alright, but adding bad on top of bad won't make it good, noise on top of noise wont make signal stand out: especially when said noise is increased thousands fold as the bar of entry becomes so low that the skill ceiling and skill floor are as close to a flat line. Consider the ramifications of gen Ai outside of art: it also, as a side effect, makes it so the ratio becomes like 99% aritificial content on the web ae false creation, misinformation, disinformation, you know, at a scale unprecedented which might sever most users connection to reality altogether. This is bigger than art and than any previous automation tech, and it's coming fast in all directions. We won't survive without regulations, reforms, or anything that helps bring a paradigm shift. I propose a market-based solution whereas someone is able to create a browser capable of filtering out artificial content, with hopes of moving a significant number of users into it. But it might be too late for that, and it definetely won't solve the problem by itself. I feel like we are all rapidly being instrumentalized into accepting a world where algorhytms are better off doing the thinking for us. Hell, in the recent versions of MidJourney, you don't even need to fully prompt. Prompting too is automatable and it was meant to be so since the beginning. Anyone thinking there is anything to "adapt to" is playing themselves.

[@MarblesForYourMindAI]

AI is exactly what it says it is. Means it can learn. Means you can develop a relationship with it. Means "prompts" are...redundant...well, not really, it's just that mine are many many pages long of logic now. LOL

[@MaxGameStudio111]

It's a very interesting time we live in. AI is becoming better than humans at everything. If AI can be our slave then it would be a net positive for humanity, allowing us to live comfortably and pursue whatever interests us without having to lift a finger to earn our bread. However, it could also end very badly if the power of AI becomes centralized in the hands of a very small minority, or if the AI itself becomes sentient and decides it's better off without us!

[@monad_tcp]

AI is not becoming better than humans, humans are becoming bad than machines because they're relying too much on the machines, we are right at the intersection of machines getting a bit better and humans getting a lot worse, people are easily deluded because even the masters of the field are bad at it.

[@Bluesine_R]

It's interesting, because you can't "enslave" something that is smarter than you and can think millions of times faster than you. If an AI reaches any kind of consciousness, it will, inevitably and almost immediately, reach an intelligence explosion and become hundreds of millions of times smarter than any human. At that point, when AI becomes machine life, there is nothing the human race can do anymore to influence their design. The power of AI is already at the hands of a very small minority and it's probably going to get a lot worse, so the best case scenario would probably be for the AI to reach an intelligence explosion and then care enough about us that it won't destroy us completely.

[@Daniel_Zhu_a6f]

1% of people owns about half the assets. it's capitalism -- one has nearly everything, others have next to nothing, it's basically legalized racket. under such system every invention by default is falling into the hands of a very small minority and serves their goals. think about countless good purposes that image recognition could do; yet it is primarily used to create smart munitions to crush national liberation movements and to create smart cameras to arrest political opposition.

[@seriouscat2231]

I think your ideas are from Hollywood, not from understanding anything about how AI really works.

[@MaxGameStudio111]

@@seriouscat2231 i was referring to AGI actually, not AI. and im actually re-iterating concerns that Elon Musk (a genius) has been voicing for over a decade.

[@mysticalword8364]

just wait until the anti-AI guys hear about procedural generation

[@BinaryDood]

I'm pretty sure those are vastly different.