Comment by Applejinx | Hacker News Reader

Comment by Applejinx | Hacker News Reader

Parent Full thread

They're all strangely intention-less on every imaginable level. That can be unsettling.

I've listened to, and made, a lot of music like that. When I want to relax or think, I like music that engages me in some way but has a sort of drift, an ambient (in the Eno sense) quality even if it's slamming Berlin techno. It's different from the prog-rock I grew up with, which was difficult and loaded with intentionality.

I ended up experimenting with music that was more generative, doing pretty sophisticated things with a degree of structure but without intentionality: always a random factor, a journey to an unknown goal (typical end result: wandering around for a while)

AI art is like that. We see it wandering around for a while. It's getting better at picking up the trappings of identity, but persistently lacks intentionality.

Maybe the trick is to supply the intentionality instead of the identity? Instead of 'creepy moat', a visual identity, make it do a painting of "you are not going to survive and that's good" or "thank you but I am not worthy".

If you have to feed it words, get a poet, don't describe the picture.

There's no difference between this and synthesizer music which sounded shockingly alien when it first came out. My uncle tells me about people going to Pink Floyd or Tangerine Dream concerts and it was virtually a religious experience to hear music so utterly otherworldly. Techno music had similar effects when it first hit the scene with pioneers like Kraftwerk or the Detroit and Chicago Techno artists.

It also came with the criticism/fear that "machines are replacing people as musicians" which was bullshit. We just got a new set of musical instruments based on electronics that musicians could play and that led to an explosion of new musical genres.

Photography was initially derided as "not art" or "machines replacing human artists" too. It was neither. It was just a new way to make pictures that spawned new art forms.

This is just another new set of artistic tools. It will spawn new art forms with their own rules and sense of technique, aesthetics, and style.

It's art if the artist is art-ing.

>Photography was initially derided as "not art" or "machines replacing human artists" too. It was neither. It was just a new way to make pictures that spawned new art forms.

This is one thing about progress and art, that I have found really interesting.

Like, how the invention of photography completely changed painting.

Ultra-realism became uninteresting over night, as soon as it was possible to create perfect depictions via cameras.

Painting, as an art form, didn't "die". Painters weren't replaced by machines.

Instead, artist's interpretation became more important, than their technical ability to reproduce reality.

We've since seen impressionism, expressionism, and more abstract movements like cubism.

The invention of photography ended up being a gift to art.

And now we see artists embracing digital artworks as their medium of expression.

So, I guess the fear of AI and machine learning being the end of art as we know it, might very well be unfounded.

It looks to me, more as the possible birthplace of new modes of expression. A reason for artists to rethink their art.

I think at the moment there is an important difference that the OP is talking about, namely the ability of the artist to channel their intention into the work. I think even early synthesizers offered a degree of control which even if the results were other-worldly were still channeling the artist’s intention. At the risk of wading into the murky waters of defining “what is art,” I don’t really feel like feeding an ML model a short sequence of words is “art-ing.”

However I think you’re right that in the long run these types of technology will become another tool in the artist’s toolbox and are not in danger of replacing artists. But I also think the OP’s criticism is valid. I suppose you could argue this is a bit like early criticism of photography as “not art” and that feeding the model is like picking what to photograph and how to frame it and such, but I still feel like there is a key difference in terms of the ability to have an intention and to have some ability to foresee how that intention will be realized in the work. Feeding a model inputs and guess-and-checking the results until you get something cool does feel “less artistic” to me.

Intention would be done linguistically then. What happens if you write poetry into this thing? Prose? Song lyrics? Stream of consciousness?

Here's an idea: freestyle rap MC with the output of the text fed into this projected on a screen.

Wait until the ghost hunters get hold of this shit. That'll be fun.

I think it would be one thing if these tools worked such that with experience you gained a kind of intuitive understanding of how your inputs map to the outputs and a finesse in crafting that poetry for the machine. But based on my (limited) understanding on ML models I have a hard time imagining that is the case. These models are complete black boxes, where a small perturbation of the input can create large, unpredictable variations in output. That makes me think that there is a strong “guess and check” aspect to these creations. And I think tools with that characteristic are limited and frustrating to create with, because you cannot channel your intention effectively through that unpredictable mapping of input to output. But I have no doubt these tools will continue to evolve in the direction of being able to be wielded more intentionally.

"Intention" is what the final fraction of a bit gap in predictive performance feels like from the inside of your head.

It has all the low-order correlations learned well, but there are long-range correlations still lacking. (Think of a detective novel where the clues are hidden thousands of words apart, in very slight tweaks to wording like an object being 'red' rather than 'blue'.) As models descend towards the optimal prediction, 'intention' suddenly snaps into place. You can feel the difference in music between something like a char-RNN and a GPT-2 model: it now sounds like it's "going somewhere". (When I generate Irish music with char-RNN, it definitely feels 'intention-less', but when I generate with GPT-2-1.5b, for some pieces, suddenly it feels like there's an actual coherent musical piece which builds, develops a melody and theme, and closes 'as if' it were deliberately composed. Similarly for comparing GPT-2 stories to GPT-3. GPT-2 stories or poems typically meander; GPT-3 ones often meander too but sometimes they come to an ending that feels as if planned and intended.)

Once this final gap is closed, it will just feel real. Like if you look at No Alias GAN (~StyleGAN4) faces, there's no 'lack of intention' to the faces. They just look real.

"intention-less" - agreed. Lacking in connection with anything human. It feels like a glorified /dev/urandom. I feel empty looking at these. I once wrote a program that iterated over every combination of colours of pixels to create every possible image for the specified size - similarly empty.

You feel empty because you know they were generated by an AI. If you didn't know that beforehand, you would probably ascribe intent to them as you would anything created by a human, you would feel something, but only because an algorithm was pushing buttons in your monkey brain.

That implies "connection" and "intention" aren't properties of the art or the artist but something entirely manufactured by the viewer.

no, that's not entirely the case.

there is, of course, the technical aspect: the blurred edges, the repeated elements, the relative placement of things and unrelated objects, a kind of statistical uniformity/ non uniformity. this is technical aspect which shows they were generated by machine and a particular implementation of neutral networks/ai.

but then there's also the intentlessness that is symptomatic of a lot of modern art, and symptomatic of why a lot of people don't like modern art.

these combine both aspects, and both aspects can be talked about and discussed separately.

To be sure, maybe one day we'll be able to consistently generate something artistic and convincingly filled with what our minds confuse for intent, but these aren't there and ai isn't currently at that level, and there is a particular property of these that make them feel empty, just like a lot of modern art also feels empty and intentless.

>there is, of course, the technical aspect: the blurred edges, the repeated elements, the relative placement of things and unrelated objects, a kind of statistical uniformity/ non uniformity. this is technical aspect which shows they were generated by machine and a particular implementation of neutral networks/ai.

I agree there is a specific signature to this algorithmically generated art that becomes recognizable after you've seen enough of it, but these elements have also been purposely employed by human artists for years, they don't objectively prove something was machine generated.

If you didn't know what to look for, you would probably assume “an abstract painting of a planet ruled by little castles” was done by an artist on a digital canvas. And I could absolutely see “The Wasteland” by T.S. Eliot” as an album cover.

>but then there's also the intentlessness that is symptomatic of a lot of modern art, and symptomatic of why a lot of people don't like modern art.

I think you're inadvertently proving my point. AI generated works feel empty and intentless not simply because of their repetition or arrangement but because they resemble modern art, which despite being created by humans is interpreted by many people as empty and intentless. This interpretation isn't a rejection of machine-generated art so much as a rejection of the aesthetics of human postmodernism.

This so true.

Netcraft confirms death of the author.

Reply on news.ycombinator.com