vvbee wrote on 2024-10-21, 18:31:

I came across a research paper that found people were more likely to enjoy AI art if they were told it was made by a human. So you might as well forget the distinction if enjoyment of art is your goal.

Indeed, preconception/bias do not do generative AI any favours unfortunately.

A thread containing two of the most subjective subject areas (AI and Art) is only going to end in pain. We just need some audiophiles to contribute now...

I still remember the days when mainstream society could not accept that anything coming out of a computer could be considered "art" (Andy Warhol excluded of course)... now we have "Digital Artists" 🤣...

I think, for many, the reason they prefer the 'art' if they believe a person did it is because they can then enter into a kind of internalised understanding or empathy regarding the artist and the very process of creating things

ai art may seem cold and distant, but its output looks vaguely familiar because of humans previous efforts, it didn't arrive from no where and start creating

in that sense there needn't be a preference, if you said "this was one artist" and "this was the algorithmically interpreted inheritance of millions of artists" it doesn't seem so bad (just long winded!)

also, it took humans to devise those algorithms

In any case, humans copy each other too, although that isn't saying its the same as AI. (some people want to appear 'challenging' by asking something like "but what's the real difference?" to any statement - AI doesn't 'know'/humans do, AI has no culture/Humans do, AI has no "instinctiveness"/Humans do. At first they may seem like interesting challenges but soon tiresome and leading no where)

for me, i kinda like AI generated "art" but i do find it generally easy to forget and pass over, very little makes an impression other than in being kind of funny, able to depict the ridiculous - like some of the funny computers in the thread

gerry wrote on 2024-10-22, 14:44:

I think, for many, the reason they prefer the 'art' if they believe a person did it is because they can then enter into a kind of internalised understanding or empathy regarding the artist and the very process of creating things

How does it differ from, say, disliking or avoiding the art of someone politically opposed to you?

Historically humans have been able to dehumanize each other to various ends. Hardline statements about AI lacking this and that human attribute are starting to sound somewhat grim as AI becomes more humanlike.

I don't think politics is the right comparison. I think it's more a case of the fact that AI art takes virtually no skill to produce. Or more aptly, a lack of perceived skill.

It's similar to how people might look down on modern art or abstract art. I've heard plenty of people when confronted with abstract art make claims like, "*I* could do that".

With AI art, anyone can spend 30 seconds typing into a prompt and generate images. This is in contrast to the months and years required to develop drawing and painting skills, and the multiple hours that go into production of any individual piece.

It's similar to if someone discovered someone had traced or copied someone else's art. It lessens the perceived value of that, since the skill required is so much less.

If AI art makes you reflect on the prompter's skill then that's the human connection and the prejudice would've been toward AI. I remember the study found that positive attitudes toward AI correlated with increased enjoyment of AI art, which to me implies exactly that. They also speculated something to the extent that increasing personal use of AI in the future will soften attitudes toward it and so increase acceptance of AI art.

vvbee wrote on 2024-10-22, 17:07:

gerry wrote on 2024-10-22, 14:44:

I think, for many, the reason they prefer the 'art' if they believe a person did it is because they can then enter into a kind of internalised understanding or empathy regarding the artist and the very process of creating things

How does it differ from, say, disliking or avoiding the art of someone politically opposed to you?

Historically humans have been able to dehumanize each other to various ends. Hardline statements about AI lacking this and that human attribute are starting to sound somewhat grim as AI becomes more humanlike.

I'm not sure recognising that there are human attributes not present in AI is 'hardline' as such so i'm not sure it is possible to dehumanise an AI, its an interesting thing - if we consider AI to have conscious self awareness and be "alive" as in some sci fi films

as i said above, i somewhat like AI art and recognise it as a blended derivation of 'human' art delivered via (human made) algorithm

vvbee wrote on 2024-10-22, 17:07:

gerry wrote on 2024-10-22, 14:44:

I think, for many, the reason they prefer the 'art' if they believe a person did it is because they can then enter into a kind of internalised understanding or empathy regarding the artist and the very process of creating things

How does it differ from, say, disliking or avoiding the art of someone politically opposed to you?

Historically humans have been able to dehumanize each other to various ends. Hardline statements about AI lacking this and that human attribute are starting to sound somewhat grim as AI becomes more humanlike.

I feel it goes deeper and is more complex than that.

Modern AI is not just another human from a different cultural or ethnic background, or with different values, viewpoints, political opinions, etc , it is fundamentally something new, different and likely nowhere near the form and abilities that it will eventually have. It may become more human like both in its interactions with humans and artistic output or it may become it's own thing completely. It may stagnate/plateau soon enough or it may eventually become infinitely more capable, possibly self aware and either extremely benevolent or malevolent towards humans.

People tend to apprehend and fear change, so a bias against AI produced art based solely on it being AI produced is somewhat understandable, in this context. Another thing that does not help is that AI is in a state of flux and is changing/improving rapidly. Many people have trouble adapting to much slower social changes that spawn years or even decades; so expecting them to adapt to, accept and embrace something much faster in pace AND with undefined potential and capabilities is asking a lot, IMHO.

Personally, I believe that we should try to keep an open mind and judge AI produced content for what it is.

EDIT: I am not implying that an AI specifically geared towards art production is ever likely to eventually go skynet on us or save the world. People tend to see things they understand superficially as a homogenous "thing". I arguably oversimplified/overgeneralized for the sake of simplicity while trying to put myself in the average non tech person's place, which is not that hard as I am no expert on AI or LLMs

Saying AI isn't a human from a different background is starting from the conclusion though if the question is to what extent is AI a human from a different background. If the subjective experience is based on electrical signals passing around in certain patterns then that checks out for AI. Some humans think in words, some in images, doesn't appear to make a categorical difference, and AI for its part may think in whatever tokens represent. AI has learnt to respond to input like a human, with its own quirks, and related to that you have the whole thing of humans raised by animals responding like animals.

vvbee wrote on 2024-10-23, 05:06:

If AI art makes you reflect on the prompter's skill then that's the human connection and the prejudice would've been toward AI.

What skill though?

Entering a text prompt to generate AI art isn't really any different than punching in numbers in a calculator to do arithmetic.

Compare those scenarios to be able to draw a realistic image by hand or solving math equations in one's head.

The latter scenarios are impressive from a skill perspective. The former scenarios are not.

You drew a line between traditional methods and algorithmic AI. Would that line move for you if you knew the person spent a month crafting the perfect prompt for an image? Or is it more that you're projecting your dislike of AI?

vvbee wrote on 2024-10-24, 17:23:

You drew a line between traditional methods and algorithmic AI. Would that line move for you if you knew the person spent a month crafting the perfect prompt for an image? Or is it more that you're projecting your dislike of AI?

The scenario you present doesn't make sense. For one, LLMs are constantly being trained and updated. Spending on a month working on prompts is working on a moving target to begin with. There is no way to really know whether better output is a result of prompts or improvements to the underlying LLM.

Second, generative AI artwork is probabilistic to begin with. Assuming a random seed, you're not going to get the same image with the same prompt. Entering prompts until you get something you like is just rolling the dice until one gets a favourable outcome.

There might be some minor learning curve in terms of specific generative AI applications. But it's a little like learning how to use the functions on a calculator. Learning how to use the sin(x) function on a calculator isn't teaching you how to calculate a sine by hand.

Neural nets, specifically forward/back propagation, is a dertministic process. Assuming the same context window for a LLM with the same models, the same prompt will produce the same result.

Training can be stochastic, but inference is not. Thats why it's called prompt engineering. If it was random, it couldn't really be engineered as each iteration would also be random.

Checkpoints are constant and the models are able to follow instructions fairly well. The point is if your reaction is to negate rather than explore the idea of skill and effort then to me it says they're not the primary factor.

spiroyster wrote on 2024-10-24, 20:12:

Neural nets, specifically forward/back propagation, is a dertministic process. Assuming the same context window for a LLM with the same models, the same prompt will produce the same result.

There is also a seed aspect to it, though. As I understand how it works the seed is randomized so you won't get the same thing each time with the same prompt. Unless one specifies the exact same seed in each instance.

This is why people will often show multiple examples of entirely different images using the same prompts.

vvbee wrote on 2024-10-24, 20:32:

Checkpoints are constant and the models are able to follow instructions fairly well. The point is if your reaction is to negate rather than explore the idea of skill and effort then to me it says they're not the primary factor.

What skill and effort are you referring to?

Btw, I'm coming at this from the perspective of someone who has learned how to draw and paint. I know firsthand both the level of effort required to develop drawing skills, and the level of effort required to produce finished artwork.

I also recognize there are scenarios where people can do initial sketches by hand and feed them into AI to do rendering, or use AI to produce reference images that they redraw or paint-over. So there are cases where traditional methods and generative AI can be blended. Though it's always going to be less effort and skill utilizing AI for these purposes than purely traditional drawing methods.

Shponglefan wrote on 2024-10-24, 20:36:

spiroyster wrote on 2024-10-24, 20:12:

Neural nets, specifically forward/back propagation, is a dertministic process. Assuming the same context window for a LLM with the same models, the same prompt will produce the same result.

There is also a seed aspect to it, though. As I understand how it works the seed is randomized so you won't get the same thing each time with the same prompt. Unless one specifies the exact same seed in each instance.

This is why people will often show multiple examples of entirely different images using the same prompts.

True, Fair point, I'll give you that 😀.. if you use a 'temperature' parameter, then yes there is a probalistic element to it. But this is used mainly for fun afaik. So this parameter (existence of) depends on which platforms you use, but Generative AI does not need this parameter to be 'Generative'. It just makes it more 'Creative' at the loss of determinism, which is where the 'skill' in prompt engineering lies...

You could say this is the difference in an artists style.... a photo-realist may want specific lines and representation, an impressionist may want to be a bit more liberal. Of course 'temperature' wrt to LLM is not do with art-style, it's a far more abstract concept than that, but it's a parameter that does not need to be used to produce 'art'.

If using it to just create some images, then yes, temperature to the max.... If you want to use it as a tool, to engineer over time and have more control over or a different style of repsonse from your prompts (the skill aspect lets say) you would not want to use it, yet can still leverage generative results.

Of course I will also say that art does not denote skill... art imo is mainly triggering the emotional reponse in the observer. It can be skillfully crafted, or easy crafted... (Piet Mondrian 🤣)

spiroyster wrote on 2024-10-24, 21:16:

Of course I will also say that art does not denote skill... art imo is mainly triggering the emotional reponse in the observer. It can be skillfully crafted, or easy crafted... (Piet Mondrian 🤣)

Hey now, don't you be slagging Mondrian. I *like* his art. 😁 (And he did a lot of realism/impressionism before going abstract.)

I'll also defend abstract art as being harder than most people think. Even in abstract art there are specific choices that effect aesthetic results. I speak from personal experience in graphics design (which I've done professionally).

In the modern world, abstract art is essentially everywhere when you start considering all the visual designs we are exposed to. This includes product design, media (print, advertising, etc.), UI design, etc.

If it was as easy as everything thinks, everyone would be a graphics designer.

😀

I'm not a hater, but not necessarily a fan.... it gives me the 'meh' vibes....

Maybe this is the crux of the discussion here? Does art have skill? As debatable as 'what is art'.... however randomness never implies skill.

Although no doubt there is a someone around who would argue there's beauty in random (is anything truly random? why the existence of an external variable called a 'seed' if it's supposed to be random?), and tbf stochastic/probalistic methods can be useful (Monte-Carlo) at producing 'art' (Ray Tracing), albeit with far more control and responsilbilty for form relagated to the author.... (constructing/crafting BSDF's is certainly an art in itself 😉)

Personally I think AI in general is a great and 'next-gen/evoultionary' tool for us and choose to embrace rather than denounce. For 'AI' generated art, imo, The skill is not in the way the observer interprets how the artist created what they created (the method)... more interprets what were they thinking when they created it (Which of course depends if the model used does probablistic inference, even then tho... had to come from something human, collective trained data or prompter?).... personally I don't get attracted to AI art (meh vibes... but a flirt of interest), rather I respect it and aspects/underpinnigns of it fascinate me. It can only get (scarily) better at it tbh, and I think we are still in the wow/shock/look-what-it-can-do phase of 'AI'. It may be artificial/calculated, but it's still based on human experiences/data.

</philisophicalBS>

Shponglefan wrote on 2024-10-24, 20:36:

Btw, I'm coming at this from the perspective of someone who has learned how to draw and paint. I know firsthand both the level of effort required to develop drawing skills, and the level of effort required to produce finished artwork.

Some people produce finished artwork by menstruating on the canvas. Goes without saying you're critical of this type of art? In any case, I remember you also said that you don't use AI generators, so I don't know if you mean to imply that what you say about them is speculation.

I missed a lot of conversation. 😜

Personally, I look at art as ... in the eye of the beholder.

Art has never been my thing, I would often look at much of it and wonder what people see that makes them call it "great", and think to myself, what am I missing? After a while I came to realize, that beauty truly is in the eye of the beholder.

When it comes to this so called "AI" art, and I use that term loosely becasue even though this is branded as AI, I don't see it that way. To me, these are just really advanced algorithms performing a task, some of them are static, some are "learning", but I don't really see it as true artificial intelligence. I think it's cool, I think it's a powerful tool, and while it is indeed artificial, I don't see it as intelligent, at least not yet.

As for the art itself, again, eye of the beholder. While I don't seem to care for most visual art styles, AI or human, there are exceptions. I often find myself liking pixel art and detailed creative photography of nature and technology, and I really enjoy it when it's based on retro/classic computing technology and electronics, probably because I'm just an old school classic geek.

That all being said, I've still been playing around with this Leonardo.AI trying to get images closer to what I see in my head, and just today, I got another output I was really happy with! Like some of the other's I've shared, this one was going for that "guy playing on his computer in the 1980's", and of course, in a pixel art style because it's what I like.

I've made several that were just ... no good ... often containing obvious "anomalies" that just make it "weird" to look at, but every so often, this particular AI gets it right. Like any algorithm, it's better at some things at others. I've learned that many of them, including Leonardo.AI, you can get 100 completely different outputs based on the same input, so it's really a crap-shoot. In a way, it's frustrating, yet fun, becasue you never know what you're going to get. In this case, I got a good one after playing with the input for a while.

I think I'm going to have to put the ones I really like on my blog. 😀