AI-Generated Images.... Are We Fucked?

[Jerzy Bondov]

I honestly think you're at the stage equivalent to the coach drivers staring at the sputtering monstrosity in front of them and laughing about how superior their horses are.

I hope you're wrong. I hope in a year we'll be talking about this shit the same way we now talk about NFTs. If not, ok. It'll still be shit though.

[touchingcloth]

The thing about AI is how fast it can learn and adapt. It will become amazing efficient and effective in a relatively short time span. It doesn't have to be great, it just has to be good enough and it's really close.

"It will become amazing efficient and effective" is an article of faith. What's to say there are no limits on how far it can improve?

Edit: I reckon a couple of hundred years ago you'd find a lot of people saying "the thing about markets is how fast they can learn and adapt. They will become amazingly efficient and effective in a relatively short time span. They don't have to be great, they just have to be good enough and it's really close."

(Although in that example they probably wouldn't have said the last sentence,  but would have said "they can't NOT be great - they'll be perfectly efficient!")

[Zero Gravitas]

That is searching for an existing image, not generating a new one.

So it's your feeling of interaction with the "creation" of the image that matters?

Even if in both cases the results are predetermined by the keywords (and seed for diffusion models) you put into it?

You're still just putting text in and getting a stored result out, if google instead of showing you a wall of images, showed you a single random one each time you searched it'd be the same experience.

[bgmnts]

So it's your feeling of interaction with the "creation" of the image that matters?

Even if in both cases the results are predetermined by the keywords (and seed for diffusion models) you put into it?

You're still just putting text in and getting a stored result out, if google instead of showing you a wall of images, showed you a single random one each time you searched it'd be the same experience.

I'm confused now to be honest.

If I want an image of a smartly dressed Godzilla fighting Jet Jaguar in Yokohama Bay, I have three options:

1. Find in a search engine
2. Send description to AI thing and have it approximately interpreted and sent back to me.
3. Send description to human artist and have it approximately interpreted and sent back to me for money.

Right? I'm just talking about if one person wanted an imagined image realised, I believe those are the only options. I imagine commercially driven mass content will be partially or fully AI in a few years time because fuck it it's cheaper I presume.

There'll always be art by people if that's your thing.

[Zero Gravitas]

You've "found" in the weights and biases of the network, "interpreting" implies some kind of understanding and digestion does it not?

This is a wholly deterministic process where training puts keywords and images in on one side, and you using it is putting some keywords in and getting those images out.

The statistical nature of it obfuscates those input images, but I don't think you'd have any reason to be more pleased with a dog in a hat popping out of either.

[ZoyzaSorris]

I hope you're wrong. I hope in a year we'll be talking about this shit the same way we now talk about NFTs. If not, ok. It'll still be shit though.

He is wrong. It's hype. Yet to see a single ai generated thing that's genuinely any good rather than a fun novelty. It's not really ai, there's no understanding there, just statistics. Very clever statistics on an unimaginable scale. Not to say it won't be used to automate a fair few tasks. But it's inherently limited.

[jamiefairlie]

He is wrong. It's hype. Yet to see a single ai generated thing that's genuinely any good rather than a fun novelty. It's not really ai, there's no understanding there, just statistics. Very clever statistics on an unimaginable scale. Not to say it won't be used to automate a fair few tasks. But it's inherently limited.

That's all human intelligence is. A massive network that makes decisions based on previous patterns that are stored. This is also inherently limited.

I say again, you can't judge what it will be like in ten years time in what you see today. Technology improves really fast.

[jamiefairlie]

It would be an interesting thought experiment to have a Turing test for art. How many people could tell the human made from the AI without any prior knowledge, coming at them completely clean. It would be like those wine experts who, when stripped of their biases about labels and so on, often fail to tell the difference between wines of different reputations.

Now this Turing test might fail today, I don't know, but it will get better until it doesn't fail.

[greenman]

I'm a graphic designer by trade, and seeing the rise of this is fucking horrible. I just don't see how my job will be viable for very much longer. The worst part is, I don't have any real argument against it. It's faster than me, cheaper than me and (the occasional seven fingered hand aside) better than me. Plus, I've spent years using photoshop to do in hours what some poor painter would have done in days.

I do wonder though whether AI art isnt a bit of a Bagpuss Chocolate Biscuit Machine, the way its being sold is its taking in a vast amount of art and essentially designing something almost totally by itself from that. I'v seen examples given though were its clearly resuing the same face across multiple images of a series, makes me think what is actually happening is much more of a cut and paste job, the AI is cutting out different parts of existing images, altering them a bit and then bolting them together. That would explain it messing up hands more than anything else as they would be harder to cut and paste being an element which is more likely to be interacting with other elements of a picture.

If thats the case I'm guessing it could be facing a lot of lawsuits(which could ironically be powered by AI used to search for parts of an image in AI art) in the near future which might force it out of business? you'd end up with a situation were AI needed its own library it owed the rights to, or at least owed the rights to draw from.

As a photographer I can't say I'm that worried about AI, partly because I sell landscape images of very specific places not some thrown together fantasy location but also hopefully said images are tastefully done which I don't really see with AI art which seems to be much better at coming up with very garish bad taste.

[ZoyzaSorris]

That's all human intelligence is. A massive network that makes decisions based on previous patterns that are stored. This is also inherently limited.

I say again, you can't judge what it will be like in ten years time in what you see today. Technology improves really fast.

Nope. What goes on in an organic brain is based on a much more complex mix of incredibly intricate chemical reactions and electric signals based on billions of years of evolution and integration with a body for survival purposes, and the sense of self and understanding and context that comes from this is something that will never be replicated by digital circuitry IMO. There are things that machine learning, which is what this is and doesn't count as real AI in my book, is very good at (dealing with an enormous amount of data) and things that organic brains are very good at (actually synthesising a model of the world with a time dimension and navigating it in a highly intuitive goal-oriented way), and 'AI' can and will do some very useful things in specific ways, but it is inherently limited by its nature and a lot of what is claimed for it is pure emperors new clothes hype.

[Paul Calf]

And here are its designs for imaginary 80s glam rock records:

This is particularly irritating. This isn't a 1980s aesthetic. It's what some bell end who was born in 2008 and thinks Flock Of Seagulls is a valid pop-culture reference thinks the  80s looked like. AI will not only destroy our future, it will rewrite our past.

[El Unicornio, mang]

If thats the case I'm guessing it could be facing a lot of lawsuits(which could ironically be powered by AI used to search for parts of an image in AI art) in the near future which might force it out of business? you'd end up with a situation were AI needed its own library it owed the rights to, or at least owed the rights to draw from.

Adobe is already protecting itself from that by having its own source library for its Photoshop AI generation. I find it incredibly helpful for extending images, adding elements to images etc but wouldn't use it to create a whole thing from nothing, and I still always have to refine bits and pieces myself. Also actually really useful for film scenes (with a static camera) where you need to just add or remove something in the scene at the touch of a button because it also takes into account the lighting and perspective and such which would be a nightmare to get right without AI.

[The F Bomb]

For me, a large part of the joy of photography is in the creation - the vision, planning, execution, technique, adaptation - rather than the pure outcome.

If I look at, for example, a picture I took of a heron at eye level flying low across a lake with a crystal clear reflection, backlit by the setting sun, I will think 'nice shot' but it means far more to me than aesthetics. It's the memory of wading through a beautiful lake, waiting for the alignment of conditions and action, the fleeting moment, and capturing something that happens thousands of times every single day in many many places around the world, but on this occasion I witnessed it and recorded it really well.

That feeling, that experience is utterly irreplaceable.

[Uncle TechTip]

Sure, that why the real concern is for people who design food packaging that spells "sausages" wrong.

[touchingcloth]

Nope. What goes on in an organic brain is based on a much more complex mix of incredibly intricate chemical reactions and electric signals based on billions of years of evolution and integration with a body for survival purposes, and the sense of self and understanding and context that comes from this is something that will never be replicated by digital circuitry IMO. There are things that machine learning, which is what this is and doesn't count as real AI in my book, is very good at (dealing with an enormous amount of data) and things that organic brains are very good at (actually synthesising a model of the world with a time dimension and navigating it in a highly intuitive goal-oriented way), and 'AI' can and will do some very useful things in specific ways, but it is inherently limited by its nature and a lot of what is claimed for it is pure emperors new clothes hype.

Sam Altman made that tweet where he said "I am a stochastic parrot and so r u".

It's sad that these people are more comfortable with diminishing what it means to be human than with considering that their algorithms might not be.

[GMTV]

"AI", as its currently described is basically the creative text simulators like chatgpt, and image creation simulators like Dall E. I think creative people in the text and image generating for generatings sake (ie content creators) are (perhaps understandably) thinking this is the end of the world.

I think we're fucked, but I don't think AI will be anything to do with that. It's related to us running out of resources and earth to keep the level of prosperity we're used to in the west. But leaving the environmental issue aside, I don't think these new simulators are fundamentally altering the course of human history.

AI for me is basically the next step in the abilities of computers to complete tasks for us. And it's now able to do simple tasks such as draw a picture, or write some simple text. But in what aspect do people think that's changing humanity at present? What it appears likely to do is to move another couple of seams of human work over to computers. Where mildly creative, generic text or images are required can now be completed by a computer program. Previously it could only be completed by human. Another few tasks of the many thousands that have been computerised.

Its still mightily impressive though, and fun to play around with. I'm not sure which aspects of our life at present will be looked on as the peak of human scientific and technological endeavour. Probably lots of things will be included. But "AI", as in the LLM's like chatgpt and image generators will likely be included in the list.

[Sebastian Cobb]

^ I think the text and image generation stuff is in part a neat parlor trick to show how good these things are at understanding instructions.

Lots of white collar jobs amount to humans acting as 'glue code' between different information sources that it's too costly to automatically connect. If this can be replaced by AI it'd get rid of a lot of middling office jobs.

Of course the AI is backed by lots of cheap human labour in the south to categorise the data, and those workers could just as easily be the offshored 'human glue' (and for some companies they are), but I think the idea here is big tech thinks it can consolidate and licence out the tool.

[Famous Mortimer]

He is wrong. It's hype. Yet to see a single ai generated thing that's genuinely any good rather than a fun novelty. It's not really ai, there's no understanding there, just statistics. Very clever statistics on an unimaginable scale. Not to say it won't be used to automate a fair few tasks. But it's inherently limited.

I thought the issue that other people had brought up is that it's not replacing something you'd need to consider good or bad, it's replacing the background stuff, which used to be the bread and butter of graphic designers.

[ZoyzaSorris]

I thought the issue that other people had brought up is that it's not replacing something you'd need to consider good or bad, it's replacing the background stuff, which used to be the bread and butter of graphic designers.

Is it though? I still haven't seen anything that is even really very usable at a base level unless you want people to know immediately that it's AI.  Even so I'm not denying it will kill off some human work, and indeed is useful already for improving productivity in many fields and I use it myself in a professional capacity, my main point is that that the fundamental upper limit of its capabilities is a lot lower than a lot of people seem to think. As a brainstorming/moodboarding tool and assistive tool (as mentioned above), it has many uses. It's not in any way actually intelligent though, and never will be.

[Sebastian Cobb]

As a software developer I'm less bothered about it taking my job, and more about it making things worse. I've already had someone ask me to help them with code it turns out they'd got openai to generate, so didn't even begin to understand themselves, that they expected me to unpack.

Secondly, I reckon it's going to make for really verbose, noisy code. Even before AI I've had people complain that I've written dynamic functional code because their ide can't understand it, heaven forbid someone might have to read and understand 4 lines of code by themselves.

[Mr Vegetables]

The immediate practical things I worry about are trust-based: an AI can perfectly replicate a politician doing something awful, but that also means a politician can more easily go out and do the awful thing. If the standard of reasonable doubt rises on photographic evidence of awful things, then it becomes easier to do them. Both of these already happen a lot, but should get much worse.

I'm also worried that in search engines it looks like the alignment problem already exists: algorithms humans don't understand are gaming Google to degrade its performance for us, and the only way for Google to tackle them is to produce another algorithm we also don't understand. It becomes harder for it to actually produce things of value for humans because the gap between the model of what we want and what the algorithm does is exploited by, well, something we don't understand.

I think it's pretty plausible these things together collapse the internet, which is not a prediction I've seen anyone else make. It seems to me that the value of trusted human interaction greatly increases in those circumstances, although at the same time the power and money we have as humans is likely to decrease even more. I think the idea that civilisation is inhuman is in the air and a disturbing paradigm shift in the way that evolution was, but like evolution it's an idea we will resist because it's so uncomfortable.

None of this has much to do with the abilities of AI itself. To me the debates around it do look a lot like the ones that happened around animals, of the "that chimp definitely isn't anything like a person and definitely isn't using tools or showing love" variety. My instinct is to assume that AI is much more similar to ourselves and how we work than we are comfortable admitting, until I see some very good reason otherwise.

I haven't yet— I understand that there are things humans do that an AI would be unlikely to, but I'm not convinced that its language and art abilities aren't exactly analogous to ours. I think all the arguments otherwise look like they describe exactly the same thing in different ways.

[Famous Mortimer]

Is it though? I still haven't seen anything that is even really very usable at a base level unless you want people to know immediately that it's AI.  Even so I'm not denying it will kill off some human work, and indeed is useful already for improving productivity in many fields and I use it myself in a professional capacity, my main point is that that the fundamental upper limit of its capabilities is a lot lower than a lot of people seem to think. As a brainstorming/moodboarding tool and assistive tool (as mentioned above), it has many uses. It's not in any way actually intelligent though, and never will be.

I didn't say it was "intelligent". I was merely going by the opinions of two people in this thread who work in this area:

Again, as a working illustrator - already in abject despair about the quality of children's picture books - I still think the future is pretty dispiriting.

I'm a graphic designer by trade, and seeing the rise of this is fucking horrible. I just don't see how my job will be viable for very much longer. The worst part is, I don't have any real argument against it. It's faster than me, cheaper than me and (the occasional seven fingered hand aside) better than me. Plus, I've spent years using photoshop to do in hours what some poor painter would have done in days.

[bgmnts]

I think we're fucked, but I don't think AI will be anything to do with that. It's related to us running out of resources and earth to keep the level of prosperity we're used to in the west. But leaving the environmental issue aside, I don't think these new simulators are fundamentally altering the course of human history.

It's a technological development that will probably put loads of people out of work no? I think it has and will contribute to the absolute mound of shit building up.

[Goldentony]

the thick cunts who push this are absolutely welcome to it and I will wank onto their bloated corpses come judgment day

[mechanical blood goat]

Those record covers are beyond dogshit and look like an Amish child's GCSE art exam

[Blue Jam]

Been looking for a bit of oil painting inspiration, techniques artists have used for certain subjects, different reflective surfaces etc, and pretty much every Google image search yields a load of results from AI sites. That's not much use to me as I want to see what another human has done using paints and brushes and their hands and eyes and brain. It's also kind of depressing. That said it's making me appreciate my new hobby a bit more, looking at the current stage of a painting and knowing how I got there, and that it was me that did it rather than an algorithm.

It's also much more fun and relaxing than typing into a box.

[touchingcloth]

As a software developer I'm less bothered about it taking my job, and more about it making things worse. I've already had someone ask me to help them with code it turns out they'd got openai to generate, so didn't even begin to understand themselves, that they expected me to unpack.

We had a really pissy third party with some clout end up having his issue escalated all the way to our CEO, because he was furious that we had deprecated some feature from an API that was crucial for his business.

The first level of support he spoke to said that the product didn't support the kind of thing he wanted to do, then after he asked why we had _stopped_ supporting it he got escalated to the next few levels of specialists and whatnot.

At some point he started waving around a load of code snippets and screenshots and manual pages to prove that we had removed a critical feature and were lying to his face about having ever supported it.

You've probably guessed the punchline already, but when he finally told the CEO where he'd found these instructions about a part of the product that no one had ever heard of before (it's a small business, so the CEO was the original developer and knows the product pretty well) he said he'd got it from ChatGPT.

"Well where did ChatGPT get it from if you've 'never' supported it?"

Mate.

[Marbles]

It would be an interesting thought experiment to have a Turing test for art. How many people could tell the human made from the AI without any prior knowledge, coming at them completely clean. It would be like those wine experts who, when stripped of their biases about labels and so on, often fail to tell the difference between wines of different reputations.

Now this Turing test might fail today, I don't know, but it will get better until it doesn't fail.

Why would that be interesting? All AI is doing is copying pre-existing human art, and it can't even manage that without human intervention. I would've thought it would only be a valid experiment if the art was spontaneously generated by AI without prompting (which won't happen).

AI art kinda reminds me of this Vim Fuego classic brag : "I could play "Stairway To Heaven" when I was 12. Jimmy Page didn't actually write it until he was 22. I think that says quite a lot."

Or going to see a tribute band at your local pub and proclaiming them to be the future of music.

[Blue Jam]

How will AI affect the likes of Jim'll Paint It and Cold War Steve and others who make a living out of quickly whipping up visual art in a distinctive style? How will it affect people who sell designs as a side hustle? Is there already an influx of even more crap on Etsy?

[Sebastian Cobb]

We had a really pissy third party with some clout end up having his issue escalated all the way to our CEO, because he was furious that we had deprecated some feature from an API that was crucial for his business.

The first level of support he spoke to said that the product didn't support the kind of thing he wanted to do, then after he asked why we had _stopped_ supporting it he got escalated to the next few levels of specialists and whatnot.

At some point he started waving around a load of code snippets and screenshots and manual pages to prove that we had removed a critical feature and were lying to his face about having ever supported it.

You've probably guessed the punchline already, but when he finally told the CEO where he'd found these instructions about a part of the product that no one had ever heard of before (it's a small business, so the CEO was the original developer and knows the product pretty well) he said he'd got it from ChatGPT.

"Well where did ChatGPT get it from if you've 'never' supported it?"

Mate.

Yeah, it can write more or less working code if it's a fairly common problem, it could probably do university assignments convincingly and the like, but if you want it to work with something niche it'll spit out gibberish that looks syntactically correct.

I tried to get it to write some ansible templates for spinning up a virtual machine from pre-prepared cloud images using proxmox and it wrote commands that didn't exist.