I feel like these articles never quite seem to grasp exactly what or how "AI" development works. Who is "we", exactly? There absolutely are people working on automating menial tasks, and what...
I feel like these articles never quite seem to grasp exactly what or how "AI" development works. Who is "we", exactly?
There absolutely are people working on automating menial tasks, and what hasn't been automated, hasn't because it's hard. Especially anything in the physical realm, where robotics presents an immense signals challenge. Microsoft invested in OpenAI to stick it in Excel, the very definition of boring manual work.
Why are we "automating art"? Well, for one, it's easy. Digital images are 3x(resolution) matrices. What could be easier for algorithms that revolve around linear algebra (e.g, matrices) to produce? Computer vision has always been at the forefront of neural network research because it's very well fitted to that kind of model, all the way back to when simple MLPs destroyed prior work at categorizing MNIST. Text is so much harder, and requires so much more computing power, because it's extremely nontrivial to somehow take letters and turn them into a matrix.
Secondly, it's because people like it. Yes, people. The "open source" communities for art generation and LLMs are much, much larger and filled with laypeople than any other area, because people have a lot of fun with the outputs, and they really don't with like, bounding box detection or sentiment analysis or all the other boring, but useful, areas.
There's not some secret cabal or central power which is being misguided, or misguiding, research towards "automating art". It's simply because digital art happened to be extremely suited for linear algebra based models and because of laypeople interest.
Also with visual, aesthetics-focused AI the range of “valid” outputs for a given input is usually vast. Compare that to object detection, writing code, writing an essay, etc. If you’re a visual...
Also with visual, aesthetics-focused AI the range of “valid” outputs for a given input is usually vast. Compare that to object detection, writing code, writing an essay, etc. If you’re a visual artist I’m sure you’d disagree - but that’s because you know exactly what you want to make. A random person with a goal of “make me an illustration of my dog playing the saxophone” doesn’t have particulars in mind.
I think in time tools that meet the high standards of artists will mature (Photoshop has been adding AI tools to the app for quite a while) and we’ll have a steady period where AI is seen as an enhancement to artists’ capabilities.
As a programmer that’s where I feel we are right now with LLMs. They’re very good at accelerating the boring parts - translation of English documentation into formal types, tests for code validity, etc.
"We" often seems to boil down to "someone else" when it's a negative thing. It's only "we" so that the "I" involved can claim a sense of ownership over it.
"We" often seems to boil down to "someone else" when it's a negative thing. It's only "we" so that the "I" involved can claim a sense of ownership over it.
I disagree with this premise. If you haven't read the article, the gist is that instead of using machines to automate menial and repetitive tasks with some form of complexity, we use it to...
I disagree with this premise. If you haven't read the article, the gist is that instead of using machines to automate menial and repetitive tasks with some form of complexity, we use it to automate creative processes instead. Not exactly the hottest take, it's more room temperature. And, with that point I don't have a problem. I'm a graphic designer, and while I'm not as hardline Anti-ML as some of my peers, I do see it as a bit disheartening that people think they can replace artists with unfeeling machines. Not the point here, though.
The premise here is simply wrong. "We're not creating the 'right' kind of AI" - wrong. We are. We're only marketing the wrong kind of AI.
I use LLMs at work to automate things I don't wanna do. This includes meaningless copywriting drivel nobody will give more than half a thought of brain cycles to, and it includes writing scripts to, for instance, transform an enormous XML export into separate human readable texts. I could do this with Excel, I could do this by writing Python by hand, but I let a machine do it because 1. it's orders of magnitude faster, even if i have to debug the code the machine made, and 2. it's, even with mistakes, doing a far better job than I ever could. I'm not a programmer. If you're willing to dredge through the capitalist hellscape of Product Hunt, there's some pretty useful apps in there, once you get behind all the marketing bullshit. No, I won't have ChatGPT write my essays. Only I can do that. But have it sort an unreasonably long JSON array I don't even know will be worth my time to look through? Sure, machine, go on.
I have honestly never seen AI art that is enjoyable or impressive; everything I've seen has been laughably bad, which I guess is entertaining in an underwhelming way. Look no further than Princess...
I have honestly never seen AI art that is enjoyable or impressive; everything I've seen has been laughably bad, which I guess is entertaining in an underwhelming way. Look no further than Princess Jane for stuff like that. https://knowyourmeme.com/memes/princess-jane
To each their own, but the winning picture showcases the hallmarks of most AI art: gaudy, boring, and completely derivative of existing works. You can't tell me that the amount of skill to come up...
To each their own, but the winning picture showcases the hallmarks of most AI art: gaudy, boring, and completely derivative of existing works. You can't tell me that the amount of skill to come up with a prompt is the same as painting a landscape or photographing subjects or scenery; these are two very distinct abilities that are not at all equal.
Well, you can debate the question of effort in art for ages, but the point is more that if Midjourney from a year ago could not only fool, but so impress people that judge art for living that the...
Well, you can debate the question of effort in art for ages, but the point is more that if Midjourney from a year ago could not only fool, but so impress people that judge art for living that the piece won the grand prize, over thousands of actual skilled artists, surely you would not be surprised that they both impress and are sufficient for most laypeople today.
Regardless of your own ESP allowing you to grade art based on the effort put into it, that competition proves that the rest of us are at least liable to be none the wiser.
Exemplary
Regardless of your own ESP allowing you to grade art based on the effort put into it, that competition proves that the rest of us are at least liable to be none the wiser.
I wonder if that's because we've been told to see ML-generated images as inherently inferior. No, I'm not looking forward to a future where all art will be made my machines, but I'm also not...
I wonder if that's because we've been told to see ML-generated images as inherently inferior. No, I'm not looking forward to a future where all art will be made my machines, but I'm also not saying that these systems are "bad at making pictures" or entirely useless as a product. I've seen quite fascinating AI pieces, and while I know no conscious thought was put into them, I can still look at them and be like "wow, that's fascinating, I wonder how the computer arrived at the math necessary to make this specific image". Sometimes you see patterns it clearly doesn't understand or intend. I think it's interesting on an academic level, and it's cool what the tech can do - which is why I call it Machine Learning Image Generation and not Artificial Intelligence Art. Calling it art projects some form of humanity onto it, when it's just mathematics and therefore inherently neutral.
I think the problem really boils down to AI being just the latest tool to further bury the gems in piles of garbage. By automating scriptwriting, you enable any idiot to spit out a bland, but...
I think the problem really boils down to AI being just the latest tool to further bury the gems in piles of garbage.
By automating scriptwriting, you enable any idiot to spit out a bland, but passable script. So random idiots can spit out AI-generated CGI. With AI-generated voices. And it'll still be better than a solid half the Marvel movies.
But because any barely-literate idiot can do this, we'll be further innundated with all this garbage and finding quality will be harder than ever.
Yeah, that's already true with anything for which we have the tech to simplify creation. Not for feature films yet, but for smaller types of artistic projects there's already a huge pool of...
Yeah, that's already true with anything for which we have the tech to simplify creation. Not for feature films yet, but for smaller types of artistic projects there's already a huge pool of blandly (un)acceptable output. Here's a great example: My wife is looking for a new spindle, to spin yarn in a slightly different way than the ways all of the other spindles she has will do. On Etsy there are endless shops selling things they call spindles shaped more or less like spindles, but missing attributes that would actually make them good for spinning yarn. Because they're something that can basically be made with any length of wood, a lathe, and some polish, the market is completely flooded. There was a time when few people sold them online and they were good, but it's pretty much impossible to find a good one if you're searching on a general marketplace. Artistic "AI" is just another step down the road. There's plenty of organic intelligences that'll happily flood the market as it is.
I feel like these articles never quite seem to grasp exactly what or how "AI" development works. Who is "we", exactly?
There absolutely are people working on automating menial tasks, and what hasn't been automated, hasn't because it's hard. Especially anything in the physical realm, where robotics presents an immense signals challenge. Microsoft invested in OpenAI to stick it in Excel, the very definition of boring manual work.
Why are we "automating art"? Well, for one, it's easy. Digital images are 3x(resolution) matrices. What could be easier for algorithms that revolve around linear algebra (e.g, matrices) to produce? Computer vision has always been at the forefront of neural network research because it's very well fitted to that kind of model, all the way back to when simple MLPs destroyed prior work at categorizing MNIST. Text is so much harder, and requires so much more computing power, because it's extremely nontrivial to somehow take letters and turn them into a matrix.
Secondly, it's because people like it. Yes, people. The "open source" communities for art generation and LLMs are much, much larger and filled with laypeople than any other area, because people have a lot of fun with the outputs, and they really don't with like, bounding box detection or sentiment analysis or all the other boring, but useful, areas.
There's not some secret cabal or central power which is being misguided, or misguiding, research towards "automating art". It's simply because digital art happened to be extremely suited for linear algebra based models and because of laypeople interest.
Also with visual, aesthetics-focused AI the range of “valid” outputs for a given input is usually vast. Compare that to object detection, writing code, writing an essay, etc. If you’re a visual artist I’m sure you’d disagree - but that’s because you know exactly what you want to make. A random person with a goal of “make me an illustration of my dog playing the saxophone” doesn’t have particulars in mind.
I think in time tools that meet the high standards of artists will mature (Photoshop has been adding AI tools to the app for quite a while) and we’ll have a steady period where AI is seen as an enhancement to artists’ capabilities.
As a programmer that’s where I feel we are right now with LLMs. They’re very good at accelerating the boring parts - translation of English documentation into formal types, tests for code validity, etc.
"We" often seems to boil down to "someone else" when it's a negative thing. It's only "we" so that the "I" involved can claim a sense of ownership over it.
Yes, it's a very old rhetorical usage, because it's pretty unconvincing to not assume part of the blame for a large social problem.
I disagree with this premise. If you haven't read the article, the gist is that instead of using machines to automate menial and repetitive tasks with some form of complexity, we use it to automate creative processes instead. Not exactly the hottest take, it's more room temperature. And, with that point I don't have a problem. I'm a graphic designer, and while I'm not as hardline Anti-ML as some of my peers, I do see it as a bit disheartening that people think they can replace artists with unfeeling machines. Not the point here, though.
The premise here is simply wrong. "We're not creating the 'right' kind of AI" - wrong. We are. We're only marketing the wrong kind of AI.
I use LLMs at work to automate things I don't wanna do. This includes meaningless copywriting drivel nobody will give more than half a thought of brain cycles to, and it includes writing scripts to, for instance, transform an enormous XML export into separate human readable texts. I could do this with Excel, I could do this by writing Python by hand, but I let a machine do it because 1. it's orders of magnitude faster, even if i have to debug the code the machine made, and 2. it's, even with mistakes, doing a far better job than I ever could. I'm not a programmer. If you're willing to dredge through the capitalist hellscape of Product Hunt, there's some pretty useful apps in there, once you get behind all the marketing bullshit. No, I won't have ChatGPT write my essays. Only I can do that. But have it sort an unreasonably long JSON array I don't even know will be worth my time to look through? Sure, machine, go on.
I have honestly never seen AI art that is enjoyable or impressive; everything I've seen has been laughably bad, which I guess is entertaining in an underwhelming way. Look no further than Princess Jane for stuff like that. https://knowyourmeme.com/memes/princess-jane
I mean people have won art competitions undercover with AI art so evidently many people do like the results, at least with sufficient curation.
https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html
And that was a year ago - it’s only gotten better.
To each their own, but the winning picture showcases the hallmarks of most AI art: gaudy, boring, and completely derivative of existing works. You can't tell me that the amount of skill to come up with a prompt is the same as painting a landscape or photographing subjects or scenery; these are two very distinct abilities that are not at all equal.
Well, you can debate the question of effort in art for ages, but the point is more that if Midjourney from a year ago could not only fool, but so impress people that judge art for living that the piece won the grand prize, over thousands of actual skilled artists, surely you would not be surprised that they both impress and are sufficient for most laypeople today.
Effort and skill are not the same thing, AI art just looks bad and dystopian. I'm eager to be proven wrong on this, but it hasn't happened yet.
Regardless of your own ESP allowing you to grade art based on the effort put into it, that competition proves that the rest of us are at least liable to be none the wiser.
I wonder if that's because we've been told to see ML-generated images as inherently inferior. No, I'm not looking forward to a future where all art will be made my machines, but I'm also not saying that these systems are "bad at making pictures" or entirely useless as a product. I've seen quite fascinating AI pieces, and while I know no conscious thought was put into them, I can still look at them and be like "wow, that's fascinating, I wonder how the computer arrived at the math necessary to make this specific image". Sometimes you see patterns it clearly doesn't understand or intend. I think it's interesting on an academic level, and it's cool what the tech can do - which is why I call it Machine Learning Image Generation and not Artificial Intelligence Art. Calling it art projects some form of humanity onto it, when it's just mathematics and therefore inherently neutral.
IMO, lots of the bad image from AI art comes from the low barrier to entry, and I think this kind of idea was talked about really well by kliksphillip
I think the problem really boils down to AI being just the latest tool to further bury the gems in piles of garbage.
By automating scriptwriting, you enable any idiot to spit out a bland, but passable script. So random idiots can spit out AI-generated CGI. With AI-generated voices. And it'll still be better than a solid half the Marvel movies.
But because any barely-literate idiot can do this, we'll be further innundated with all this garbage and finding quality will be harder than ever.
Yeah, that's already true with anything for which we have the tech to simplify creation. Not for feature films yet, but for smaller types of artistic projects there's already a huge pool of blandly (un)acceptable output. Here's a great example: My wife is looking for a new spindle, to spin yarn in a slightly different way than the ways all of the other spindles she has will do. On Etsy there are endless shops selling things they call spindles shaped more or less like spindles, but missing attributes that would actually make them good for spinning yarn. Because they're something that can basically be made with any length of wood, a lathe, and some polish, the market is completely flooded. There was a time when few people sold them online and they were good, but it's pretty much impossible to find a good one if you're searching on a general marketplace. Artistic "AI" is just another step down the road. There's plenty of organic intelligences that'll happily flood the market as it is.