I have literally no idea what point you’re trying to make here.
If I was the rights holder to the iconic Jaws movie poster and I was mad at Darren_Garrison, I would take him to court and say “He’s copying my poster!”. Then the judge would ask me HOW he’s copying my poster and I would display the two posters and say “See how the shark is doing this and looks like this? And see how it’s the same in Darren_Garrison’s poster?” and the judge would agree that Darren_Garrison’s poster is in fact derivative of mine because I was able to point out how A matches B. Then it would probably be thrown out as parody anyway but the point is that my case would rely on how one matches the other to prove that there’s a derivative element. Just like, if I was claiming someone unfairly used my song, I would point to copied harmonies or riffs or, if I wanted to say someone unfairly copied my prose, I would point to copied passages of words. Then the judge would decide how close they are and whether it counts as copying.
Note that the actual stuff being compared is the output here. If Darren_Garrison made a collage of cut images or drew it himself in crayons or took a photograph using toys or created it using digital art tools in Photoshop, my argument would still be “He copied iconic elements of my protected work” and then trying to show a judge which elements were copied and why it’s unfair.
On the other hand, if someone made an AI generated render of an ocean scene and I said “This is derivative because my poster is in the training data and my poster has an ocean and HIS image has an ocean so some of my 1s and 0s are in his image!” then the judge would ask me to show exactly HOW his image was taking elements of mine. Then I would lamely say something about how it has to be or else magical AI fairies and ask “But what about my mp3 server how is that different from an AI brain, huh?” and then I would lose my case because that’s not how proving a derivative work happens as demonstrated by the thrown-out court case trying to make that exact argument and the other case that ALSO points to the copying elements as the weakest and least sustainable parts of the case.
The problem with a “god in the gaps” is when the gaps keep closing. Computers are deterministic because a great deal of effort is put in to make them so. But non-deterministic computers have been conceived and built. They are vastly more energy-efficient than deterministic ones, too. The problem has been finding a way to use computers that can make mistakes. AI training and generation could well be the usage case for these chips, removing the need to intentionally inject noise into the output (as is done today). If poorer error-cortection is the seat of creativity, as you seem to believe, these chips should fill that gap.
Traditional digital computers depend on millions of transistors opening and closing with near perfection, making an error less than once per 1 trillion times. It is impressive that our computers are so accurate—but that accuracy is a house of cards. A single transistor accidentally flipping can crash a computer or shift a decimal point in your bank account. Engineers ensure that the millions of transistors on a chip behave reliably by slamming them with high voltages—essentially, pumping up the difference between a 1 and a 0 so that random variations in voltage are less likely to make one look like the other. That is a big reason why computers are such power hogs.
Radically improving that efficiency, Boahen says, will involve trade-offs that would horrify a chip designer. Forget about infinitesimal error rates like one in a trillion; the transistors in Neurogrid will crackle with noise, misfiring at rates as high as 1 in 10. “Nobody knows how we’re going to compute with that,” Boahen admits. “The only thing that computes with this kind of crap is the brain.”
Unless you choose to call it art.
I was going to inquire: why wouldn’t/couldn’t that be art? I don’t understand the point that anecdote is trying to make. And, for me, “art” is very much about the imperfections. But I somehow don’t think that’s the point that was trying to be made, I think.
It seems like you are really bringing 2 axes to this grindstone:
- Why should AI trainers pay for work produced by their training material?
- Why are we afraid of AI displacing actual human works on the market?
To the first, in my opinion, AI isn’t passing off the source material as its own product. It extracts some traits about many musical inputs and produces something similar. So I personally think that the critique of AI as “plagiarism machine” is inaccurate and can be dismissed.
To the second, I have no worries whatsoever about an AI actually replacing the Beatles or some other irreplaceable group. Art is a social conversation, it’s not a simple mathematical representation of an input. What value is there in AI making something exactly like Stairway to Heaven? It’s already been made, that cultural conversation is past. Moreover I’m not worried about AI displacing whatever work is destined to be the next Stairway to Heaven. AI can produce some things that are maybe interesting or attractive. But since it has no sense of its place in human culture or society, has no creative impulse of its own, it cannot take part in the social artistic conversation or reflect on the meaning of art. At the very least, because its snapshot is at least a year old, and is at once too narrow (only ingests human descriptions of work, which is flawed and context-dependent) and too broad (it takes in massive amounts of those inputs and over-fits to them). But more broadly because it doesn’t participate meaningfully in the two-sided conversations that are needed to inform one’s own perspectives and influences.
This is why AI art product will always have something like the “six-fingers” problem. For all of AI’s brilliance in figuring out how to generate rows of teeth or hands of fingers, apparently it’s an intractable problem to tell it “most humans aren’t portrayed with 6 fingers”, because it doesn’t know the meanings of “human” or “finger” or “6”, and even if it did, it’s unclear that you could explicitly direct it to behave accordingly. I’ve no doubt someone will figure a way to kludge over those obvious cases. But those cases are endless, so I’m not really worried about AI putting real artists out of work, only craft-types who create mass-producible dental-office-waiting-room art. Which is a concern, but not one that I see as incredibly important.
This is so true; there is a bit of Dunning-Kruger to it; the people who have not invested in the skills to create the ‘product’, have a tendency to think that their “no, not like that - why don’t you just do it differently” sort of input has exactly the same value as the creative process itself.
That’s simply not true. The gulf between brain works and computer works remains identically huge as it did in the days of Von Neumann. Every computer is, and has always been a simple automaton deterministically executing simple instructions based on its inputs. People have been postulating we could do something different since the inception of computers, but they have remained theories or impractical prototypes.
In fact in terms of human understanding the gap has got much bigger. In the early days of computing it was postulated that the human brain did work the same way as a computer, we now know that not be the case.
The specialness of human brains, in regards to creativity, is in fact what is being debated, not accepted a priori.
Our not understanding how fairly simple deterministic processing at synaptic levels in aggregate results in, among other things, creativity, is not relevant to it being creative. Nor is being to explain how gen AI achieves its process and results from deterministic CPU interactions. It’s irrelevant.
I might be oversimplifying here, but AI-produced work is derivative - if not in exactly the conventional sense of that word, the output of an AI model is based on training that is derived from exposure to real-world works of art, literature or other creativity.
If you want AI to continue to be able to do that in future training exercises, it sort of makes sense not to use AI as a weapon to bludgeon the source training material into nonexistence. The sort of “haha! I can take your stuff and use it to outcompete you to death!” attitude that we’ve seen from the corporations creating AI models is incredibly shortsighted and serves only to fatten the wallets of billionaires, not some betterment of humanity.
Unless we posit some future world where there will either be no need for fresh, human-created works (I don’t see it - in fact once we start seeing models that are trained on the output of models trained on the output of models, I think things will get weird and terrible), or a world in which humans would be happy to continue creating works for no reward, only to be consumed by AI training, that’s a reason to pay creative people when you use their work and not completely displace them.
Given your response I have no doubt that you didn’t read or even pause to consider the article I linked but maybe it will be interesting or useful for someone else. I’ll bow out now from attempting to debate with you that brains aren’t special magic sprinkled with fairy dust.
What is the difference between copied and inspired by?
This is critical because it would describe the difference between a computer heavily modifying an existing song and an AI creating a song via rules it set down (from analyzed training data) to define what songs sound like.
You ask AI to create a song that sounds like a Beatles song, it doesn’t cut and paste pieces of Beatles songs, it looks at rules. What properties do Beatles songs have in common? What themes do they convey, beats, melody, chords, whatever other song related analysis it’s able to do. Then it creates a work that is consistent with the rules.
If that isn’t inspired by I don’t know what is.
Clever maths and fairy magic.
About four million bucks in 1990 dollars:
I just want to re-emphasize this again because there still seems to be the common misconception in the thread that the product of AI prompts being bland and generic is a technological failing that will be improved with future technology is fundamentally mistaken, that’s how it’s intended to work.
Don’t forget, we’ve had “AAI” (aka: human subcontractors) for millennia now. If I run a graphic design firm and you come to me asking for a logo for a hardware store or accounting firm and give me no other info except for that, of course I’m going to give you the most generic, on the nose logo I can, that’s the logo most likely to please you because you haven’t made any choices, anything I do to make a choice is far, far more likely to go further away from what you want than closer.
And the worst, eye-rolliest feedback you can get from a client is stuff like “Can you make the logo pop more?” or “Can you make it funner?” (or in the programming world, “Can you make this easier to use?” or “can you make the interface cleaner?”). It’s a sign that the client fundamentally has not thought through what they want to make. The job of the designer/programmer then is to work through with the client what it is they actually want and how to make choices to get there. Again, we’ve known since the 1980s that this is the primary difficulty with programming. Even if we were to imagine the perfect, effortless tools that can manifest anything you want in the blink of an eye, it has not eliminated the fundamental problem that you don’t know what you actually want until you’re forced to think hard about it and make choices.
I think people who aren’t in a creative field don’t have an appreciation for the true combinatorial explosion of possibilities that flower from creative choices. Like, in the center of the space might be a few hundred or thousand logos for a generic hardware store (throw in a hammer, a muscly arm, big bold sans-serif text etc.) that tell you nothing beyond “this is a hardware store”. But then as you explore the space, it’s like, quadrillions upon quadrillions of possible great logos for all sorts of different hardware stores that aren’t your store, amongst a sea of googolplexes of logos that are terrible for any hardware store. Which of the quadrillion logos is right for you? That requires you to make choices. At the very least, to reduce 1 quadrillion down to 1 requires you to make at least 50 binary choices, it’s simply mathematically impossible to do it in less. Me as a human, AI as it actually exists today, any hypothetical AI in the future, simply cannot deliver you what you want if you don’t even know what you want and everyone is overconfident about what they want (including artists!) until it takes time to actually produce it.
Another thing people fundamentally misunderstand about the arts/artists is that the skill of making choices is what’s primarily honed through the practice of execution. It’s easy to look at a sketch artist who can make photorealistic pencil drawings or a programmer typing arcane runes into a terminal to make it do stuff and understand that it’s hard and you don’t want to learn that so it’s better to hire a professional to save you some time. But artistic execution is only there insofar as it aids in your making faster, better and more interesting choices. The reason why artists practise so much is only incidentally about improving their technical skill and more about using their current technical skills to explore different choices in the space to be able to more confidently make choices in the future and define “their” style. You draw a hand 1000 times and each time, you’re making different choices on how to execute it and learning from each one. Or when programming, so often, I find multiple ways of doing something (eg: using an interrupt vs polling a connection) that all accomplish the same thing but with subtly different tradeoffs that you have to take the time to understand.
There’s a mistaken belief among non-artists that, absent the technical execution barriers that stopped them from being artists, they can make choices of equivalent quality and therefore, what’s the need for a trained cadre of artists anymore? But we already went through this with the invention of photography, all of a sudden, you didn’t need to know anything about mixing oil paints or hand-eye coordination to produce an image but all it became was a new tool in which you still had to apply painstaking amount of craft and experimentation to develop an “eye” for what made for a striking photograph. It’s not appreciably any easier to become a great photographer than a great oil painter because the medium is merely there as a platform for you to develop the skill of making choices.
And so is human produced work.
There is no reason to believe that the information processes that human brains use to be inspired by past works and experiences and create new things are replicated by genAI. But creativity is always building on what’s come before, it’s always derivative. (And we are not talking law here.)
I do not believe that genAI has its own intent of what it wants to express, be it an idea or a feeling. That matters.
New art for humans by humans derives from past art and the world, and it derives from what we humans are needing in response to that world, it is a communication process between creators and receivers, both doing some of the work. I think genAI art requires the human prompt.
Not sure though about creativity for things in science and technology?
Is there? I haven’t gotten that impression at all. I don’t think there’s any technological reason for today’s AI generated content to be “bland”. I think it’s entirely possible right now to create interesting AI imagery. Music and prose might be more difficult just in how we consume them; you can cast your eyes over an image in a moment versus the time required to listen to a song or read a passage of prose so I think the experience (and tolerance) is different.
That said, I think two major obstacles in the conversation are (a) What are we calling “AI Art”? In that I personally view it as a tool to generate images based on user input and revisions to that input (via new prompts, inpainting, etc) whereas it feels like other people are just assuming some AI space brain making images of its own volition and (b) the fact that “bland” is a pretty subjective and meaningless term. I previously posted a series of images from Midjourney that captured a range of styles and content. Are all of those bland*? Someone might insist that they are and I couldn’t say they’re wrong (again, subjective) but I would likely just shrug and wander off at that point since we have such different ideas of what constitutes a “bland” image that there’s no sense in getting into it.
*
And, arguably, those would be some of the “blander” or more generic images on the site just because they were picked by community vote so would be the most attractive to the widest audience but I still wouldn’t call the array of images “bland” in any meaningful sense of the word. YMMV, etc.
Not sure if you would call it “creativity”, but AI makes designs that people wouldn’t have thought of and can’t understand, but work.
Sure, human work is derivative, but the intent of the human artist is seldom ‘outcompete my sources to destruction’ and even when it is, it is not usually successful.
Though it’s really not. The specialness of human brains is the most fundamental cornerstone of human society. The fact humans have agency, rights and responsibilities that inanimate objects do not is a fundamental basis of how society works. As far as I know no one is arguing with that (though among the more out there ai-bros who’s to say 
).
What is being debated is whether certain inanimate objects (specifically a collection of assembly language instructions, but only some collections of assembly language instructions) also have some of those human attributes. And no one has given a logical explanation of what is special about these collections of determinstic simple computer instructions that gives them these attributes.