There are infinite worlds to explore | Hacker News Reader

There are infinite worlds to explore

moultano.wordpress.com

113 pointsapsec11232 comments

I remember back in the old days we all thought that it wouldn't be long before computers and artificial intelligence would learn to do advanced science and reason about the world. But, we thought that things like art, that involve some unknown hidden 'human' or spiritual element, would be something machines might never be able to deliver.

I find it remarkable that it turned out to be the other way around. Turns out machines can do amazing art very easily. They are still nowhere near being able to replicate human intelligence though.

I guess it is because art and vision occur at such a deep level we aren't aware of it, so we used to not really understand how it worked. It's quite hard to build a machine that does something when you don't understand what it is. We built them anyway because apparently even though we didn't understand it, it wasn't very complicated compared to other tasks that we perform.

I have seen a ton of AI art, a lot of it impressive, but I’ve never seen a piece generated by a computer that had any significant meaning behind it that the human artist didn’t give to it. So AI hasn’t solved art at all really.

Are you sure that knowing it was generated did not affect your impression?

If you were told instead it was made by a human, you'd likely reach harder for a meaning without knowing it, and find it.

Meaning is subjective enough as it is.

That is probably true! I think there’s a case to be made that art is in the eye of the beholder, and that anything can be art depending on how it’s perceived.

The meaning still feels accidental though, not that far off from thinking a random sentence generator’s output is profound. It may very well be profound, but I don’t think you can attribute the profundity to the algorithm. AI (From the AI works I’ve seen) is incapable of purposeful communication with the audience. AI art isn’t solved until it can do that, let alone communicate new ideas.

i think that's the point. who cares about generated art? the point of art is that a human being is saying something to you, the human audience.

I think that is exactly not the point, because the meaning of art is determined by the beholder, not the artist. They might have had an agenda to share, but that doesn’t make another interpretation of an artwork any less valid. This means that if you’re able to draw any conclusion from a piece generated by an AI, that makes it art.

Discussing this topic further leads down an (interesting!) rabbit hole of what makes things actually worthwhile in life.

Good point. I was thinking the concept of "beauty" is comprised of both pure aesthetics (cleanness, composition, ratios, etc), and symbolism. For example, some heavy metal music might be perceived as "unaesthetic" or unpleasant, yet it carries deep meaning for someone who's best friends hear that music. Or went to a festival and met wonderful people there. All this meaning gives it "symbolic beauty".

Art museums are full of paintings worth millions (or more) that sometimes don't have any significant meaning. Tons of drawings of tables full of fruit for example, looking photorealistic but nothing else.

Nobody runs around and tries to get them removed from museum just because they are in same hall as Rembrandt.

Depends what you mean by art.

“High art”, where each piece is supposed to have meaning and advance a narrative with critics? Sure, algorithms can’t do that yet.

But for most people, art is just something they like looking at, and algorithms are great for producing that.

Also, the criticism that this isn’t good art because they all look kind of the same, is the same criticism levelled at human-produced pop music or trash fiction.

I wouldn’t be at all surprised to soon see an algorithm capable of producing a novel-length text of similar quality to that on sale at airport bookshops.

I think any decent art is trying to communicate something, That’s a core part of what makes art, art. So I think it’s way off to say that AI has “solved” art. It’s got the aesthetics part almost solved but that’s about it.

I’m not sure what you’re referencing with them all kind of looking the same, that’s not my problem with AI art.

Yeah AI can only really mass-produce a hardcoded aesthetic right now. There's much more to art than that.

> I find it remarkable that it turned out to be the other way around. Turns out machines can do amazing art very easily.

If trained on the output of human masters, James Gurney in this case.

What I feel these pieces lack above all is composition. There is nothing that is framed, no center piece, nothing to draw the eye, it's all texture, almost uniformly distributed.

This is perplexing to read. What is the last piece of art you consumed that was generated by AI? Have you read such a book, seen an imafe, listened to an album?

...this post? If you didn't realize this art was generated by computers then I guess that proves my point.

Which doesn't clarify much given this post isn't a piece of art consumed by people. You can make "art" all you want for your own purposes using computational techniques. Fact of the matter is we still don't know how to do it similar to humans make books, musics, paintings etc...

I instantly realized this "art" was generated by computers since it has all the characteristics of computer-generated "art": All surface and no substance. Which is why it's not really Art.

This will eventually reach the point where a human can direct the process and the AI will fill the gaps. It would be no more weird than artists who throw paint on a canvas and end up with a coherent piece.

Author here. Glad you all are enjoying this. It was a lot of fun to make. You might also like my previous piece that had an accompanying narrative. https://moultano.wordpress.com/2021/07/20/tour-of-the-sacred...

Hi! Thanks for making these! They're brilliant!

I enjoyed your "Tour of the Sacred Library" a lot! At first, I was just reading a nicely written, very "ambiance heavy" story, then little by little, I realized that the accompanying images were a bit "off", and finally that they were generated. Very inspiring experience.

Pretty cool stuff - I eagerly await the day when the same generative models can spit out fully explorable 3D environments that I can then explore in VR.

Yeah - that's my wish too. I think all the pieces already exist but you'd need a supercomputer to get realtime frame rates.

You don’t need a supercomputer.

You can actually hook algorithmic art from Houdini to GAN models and bake 3D art for later use.

Problem is that GAN models are still not good enough to produce human level realistic images.

It can generate trippy images but these are not very useful for commercial products.

We still need better models or a completely different approach for computer generated realistic art.

Diffusion models are better - but their inference times are orders of magnitude slower.

GAN models are really cool. Don't get me wrong. But they are popular _because_ they are fast at inference. They are still challenging to train (hopefully projected GAN helps with this) and have a very tough time representing highly diverse datasets with lots of data.

> You can actually hook algorithmic art from Houdini to GAN models and bake 3D art for later use.

I've seen some stuff like that but it's not really what I'm talking about. I want fully dynamic, spatially consistent generative worlds.

And that is pretty hard to do at 75 or so FPS in stereo.

This is some of the most compelling algorithmic art I've experienced. I hope the artist (guide?) continues.

I'm most intrigued by the use of manual modification of the images. The hybrid approach amazed me where the author entered the loop to erase and blur one of the images that the AI then expanded on.

This style of interaction actually fits very well with my current theory of consciousness. It allows "echoing" of ideas back and forth with the AI. Repeated echoes allow an idea to form using the best inputs from all involved.

I found the generation technique fascinating, but the images felt 'wrong' - I felt like I didn't know where to look to get an overview of the scene, and when I tried to focus on details it was like me eyes kept 'slipping off' because nothing would resolve itself.

Really interesting work (and the discussion it), and very nice visuals. Thank you!

Meta commentary:

The code that I started from was written by RiversHaveWings based on a method by advadnoun with further modification by Eleiber and Abulafia.

Metaverse or otherwise, we're definitely moving towards self-chosen names that only really work in the electronic-life realm. I'm aware this has been a slow shift over decades, but the sentence above is 33% online nomenclature by letters.

RiversHaveWings (Katherine Crowson): https://twitter.com/RiversHaveWings advadvnoun (Ryan Murdock) https://twitter.com/advadnoun

Don't know the other two. The ML art community is weird. They like to cite a lot due to presumably academic backgrounds but no one _really_ wants to be hyper-famous amidst the NFT crowd (recently hacked Banksy, for instance) and so they all use handles pretty frequently.

> Wish I didn’t have that shadowy figure there. He shows up a lot for some reason.

Is this leading up to "The Ring: Revenge of the AI"?

Jokes aside, great images and the technique is interesting as well.

I do really love the art. I wonder if someday, we can walk and explore inside of these artworks.

Reply on news.ycombinator.com