Well yes. But my point is that at some pretty near point in teh future, most of the stuff on the internet is going to be AI generated. And that will be the training data for future AI etc. etc. etc.
It’s both, which I think is one of the things that makes it especially emotive for creators; this is the customer walking into your knife shop and stabbing you with one of your own creations. (And if you want to complete the analogy, taking over the shop and replacing your handcrafted products with cheap mass produced garbage)
That will be interesting. I think a lot of people regard it as inevitable that it will totally stagnate once this happens, but all it took was replication, mutation and selection to evolve the huge diversity of life on our planet, so I don’t think it is inevitable that AI fed on the output of AI, with noise inherent to the process of manufacture, plus humans performing selection of outputs, would necessarily stagnate.
(That’s assuming humans are still viable after the economic world has been transformed by these things)
You use the word “deterministically” there like it’s se horrible flaw in the AI, but your own artistic endeavor are no less deterministic than the AI’s.
They hacked into your system? Then the people running the AI should go to jail.
But they didn’t, the AI computer asked for the photo, and your computer gave it to them because you told it to show a copy of the photo to anyone who asks. Or you gave the photo to someone to spread all over the internet, and they let the AI computer see it, just like they do to everyone who asks for it, because that’s what you told them to do.
You don’t like that something you made is being used to train your replacement. Nobody does, but it’s happened a million times before and will happen a million times again.
Knockoff products are a thing, they’ve been a thing for as long as people have made things. People look at your product, and make something similar under their own name, they don’t even have to buy it, they just have to look at it to get the idea of making a knockoff.
Nobody is wailing and gnashing their teeth over knockoff knives because knife makers aren’t artists who have been granted a legal stranglehold on their product usage until their grandchildren are retired.
This is an excellent point. Yes, there will mostly be stuff that went through an AI generator on the internet, but it won’t all be equal. Some will be true art, carefully prompted, inpainted, edited, etc by the likes of our own @Darren_Garrison; and some will be low quality memes generated by the likes of me to quickly toss into a conversation or something.
The datasets used for training now didn’t just blindly crawl the internet to include every child’s paint.png or every poorly photoshopped online meme. And remember, to be useful the training images have to be labeled.
So even if future AI training datasets contain some art generated by [humans whose tool of choice is] AI, that doesn’t mean the art will stagnate or degenerate.
The humans who use AI art extremely well - again I am gonna bring up @Darren_Garrison - can do things the rest of us lack the skill to do, just as we may have (or lack) the skill to draw with a pencil or with Photoshop. And I hope we find a way to set up the system so they can monetize that skill, which would incentivise them to keep using it, which benefits us all.
My “great skill” is time and stubborness–the willingness to generate a couple of hundred images before I reach one that works.
How does someone Xeroxing Harry Potter books benefit us all? I’m sure it’s something that could be monetized.
Have you looked at the art that @Darren_Garrison generated in other threads? It’s definitely original and creative artwork, not Xeroxing a Harry Potter book.
Yep, but this is a potential revolution in the easy mass production of knockoffs.
Citation needed.
I’ll address a few recurring issues here:
First, AI has been used to create derivative works, intentionally. Humans do that too. Whether or not a particular case infringes on the rights of an original artist has to be determined in the courts, but there is no difference to consider based on the type of artist, human or machine. That it’s easier to do with the machine is irrelevant.
Second, AI doesn’t have to produce derivative works. AI could create random images and use what it learns about images that are pleasing to human eyes to select those that would be considered art. You can quibble about the methodology involved but it is not really any different than what human artists do. Human artists sometimes create random images, sometimes have sometimes have some purpose in creating the image, but the result is the same, both human and machine can determine the perceived value of an image as art based on human tastes.
Finally, everyone should consider that humans aren’t as creative or intelligent as often assumed. AI has been considered not intelligent by many because it hasn’t thought up any new mathematical principles or worked out to produce energy from nuclear fusion. I ask simply, how often have you demonstrated that level of intelligence? By those standards only a tiny percentage of all humans qualify as intelligent.
If you want to start a new thread about it, go ahead; the two minute version: unless we introduce nonsensical fantasy or sci-fi concepts such as a “soul” or a “free will”, there is nothing but deterministic history and completely random chance from which to derive your actions; so unless you are insane, and acting totally at random, your actions are deterministic.
I can get behind that, yes.
I really should have included a smiley with that one.
But even accepting that both minds and AI are on some level deterministic (which is less of a gimme than are you presenting it as), there is a huge qualitative difference between a computable algorithm that could be run on any calculating machine, and whatever the hell goes on in the brain which is to the best of our knowledge completely inimical. When I say AI is deterministic I don’t mean simply that it’s part of a material universe, I mean that is rigidly constrained by a pre-existing algorithm. And this is in fact not a description of the human brain.
So I had a quick Google on the datasets and found that a) the LAION set has 5.8bn images and b) turned out some of them were of [trigger warning] child sexual abuse. So I don’t think it was terribly far away from blindly crawling the web - I would be sceptical that it was possible to check 5.8bn images in theory, and presence of those images kind of confirms that.
In fact:
The Large-scale Artificial Intelligence Open Network (LAION) released LAION-5B, an AI training dataset containing over five billion image-text pairs. LAION-5B contains images and captions scraped from the internet and is 14x larger than its predecessor LAION-400M, making it the largest freely available image-text dataset.
The release was announced on the LAION blog. LAION-5B was collected by parsing files in the Common Crawl dataset to find image tags with alt-text values. The corresponding images were downloaded and filtered using CLIP to keep only those images whose content resembled their alt-text description.
The new dataset, LAION-5B, was collected using a three-stage pipeline. First, a distributed cluster of worker machines parsed datafiles from Common Crawl to collect all html image tags that had alt-text attributes. Language detection was performed on the alt-text; in cases where the language detection confidence was low, the language was recorded as “unknown.” Raw images were downloaded from the tagged urls and passed, along with the alt-text, to a CLIP model to calculate embeddings for both. A similarity score was calculated for the two embeddings; pairs with low similarity were discarded. Additional filtering removed duplicates as well as samples where the text was less than five characters or the image resolution was too large.LAION engineer Romain Beaumont joined a Hacker News discussion about the release. In response to a criticism that the dataset was un-curated, Beaumont replied,
Non annotated datasets are the base of self supervised learning, which is the future of machine learning. Image/text with no human label is a feature, not a bug. We provide safety tags for safety concerns and watermark tags to improve generations…It also so happens that this dataset collection method has been proven by using LAION-400M to reproduce clip models. (And by a bunch of other models trained on it)
This is a set of images scraped off the net and labelled by AI specifically to keep humans out of it.
I stand corrected!
That doesn’t make it right, or legal. Putting a photo on the internet does not eliminate copyright. There are terms/agreements for hosting platforms, social media sites, etc. that specify their right to display the images you upload, otherwise there would be no point. However, that does not confer other rights to the platform (save for additional clauses in the terms-of-service) or to other users. You still have just as much right to pursue copyright violation against digital copying/sharing as you do physical copying/sharing. That an AI training bot can look at a photo you posted online doesn’t automatically mean it has a right to store that image (or a hash of it) in its database. That’s where the legal and ethical questions arise.
I suspect (hope?) copyright law will be changed to acknowledge AI training as a use-case that artists can control. How much control they have, and how it’s enforced is another question. Maybe it’ll be as simple as a robots.txt equivalent to block indexing, or some other flag indicating what can and can’t be vacuumed up by the training bots. In any case, training does seem to fall right on the line of acceptable fair use, so a determination needs to be made there. Yes a person can look at a work of art and internalize it to inspire a new work without violating copyright. Is an AI training bot similar, or is it more directly copying/storing the work? These are the questions that need answers.
Whenever copyright questions come up, I find it helpful to consider how things work in the fashion industry. Famously there is no copyright in fashion. Anyone can reverse engineer or copy anyone else’s designs, as long as they don’t run afoul of trademarks. Made a perfect replica of that Prada handbag all you want, but you can’t put a Prada logo on it, or try to pass it off as a genuine Prada bag. To pick up on an example from upthread, an AI generated “new” Beatles song featuring John Lennon would be a non-starter for that same reason, but a Weavils song featuring Jack Kemmom would probably be ok.
Yeah, I don’t think it will be. It’s going to be more like Mickey Mouse.
In the past, to do this sort of “copy” artist you actually needed a real life person. Here there is not going to be a real life person, just the AI making a digital copy. So the cost, the input is minimal. Plus if you can do it, a billion other people are going to make John Lennon copies as well.
So with Mickey Mouse, you can make a cartoon mouse, but you need to change it enough so it’s easily NOT recognizable as Mickey Mouse. We already have laws for that. So just apply those laws to this digital creation, which is much more similar to a cartoon character, a complete fiction, than an actual existing person. No matter how close the AI can get to making it sound real, it isn’t.
I think the additional issue with deepfakes is going to channel things towards being with artist approval only. It’s possible that current forgery and defamation law covers these issues. But I expect places like YouTube are going to have software that blocks anything sounding too much like John Lennon that doesn’t come from the John Lennon official site. So you’re going to have to change your “Jack Kemmon” digital creation enough so it is readily distinguishable from the image and the voice of the actual John Lennon.
I think it really depends on what is being trademarked/copywritten. Some things can have their form copied but not their trademarks, while other things are specifically protected. The shape of a Prada handbag may not be protected, because lots of handbags have that shape. So all they can protect is the trademark or other identifying unique characteristics. Mickeyh Mouse is completely unique and copywriteable.
In general, what I’m noticing is that AI art isn’t fooling many people into thinking it’s real, but it’s making people think real photos or art are fake. People aren’t wildly amazed by AI art, but now look at human art with a more jaundiced eye.
It’s gotta be frustrating for an artist to spend days, weeks, months on a work, only to have viewers go, “Meh. I created something like thast last night with a prompt.”