I know a lot of people want to interpret copyright law so that allowing a machine to learn concepts from a copyrighted work is copyright infringement, but I think what people will need to consider is that all that’s going to do is keep AI out of the hands of regular people and place it specifically in the hands of people and organizations who are wealthy and powerful enough to train it for their own use.
If this isn’t actually what you want, then what’s your game plan for placing copyright restrictions on AI training that will actually work? Have you considered how it’s likely to play out? Are you going to be able to stop Elon Musk, Mark Zuckerberg, and the NSA from training an AI on whatever they want and using it to push propaganda on the public? As far as I can tell, all that copyright restrictions will accomplish to to concentrate the power of AI (which we’re only beginning to explore) in the hands of the sorts of people who are the least likely to want to do anything good with it.
I know I’m posting this in a hostile space, and I’m sure a lot of people here disagree with my opinion on how copyright should (and should not) apply to AI training, and that’s fine (the jury is literally still out on that). What I’m interested in is what your end game is. How do you expect things to actually work out if you get the laws that you want? I would personally argue that an outcome where Mark Zuckerberg gets AI and the rest of us don’t is the absolute worst possibility.
Many things in life are a privilege for these groups. AI is no different.
I’m not sure what you’re getting at with this. It will only be a privilege for these groups of we choose to artificially make it that way. And why would you want to do that?
Do you want to give AI exclusively to the rich? If so, why?
I think he was just stating a fact.
For something to be a fact, it needs to actually be true. AI is currently accessible to everyone.
I disagree. I can barely run a 13B parameter model locally. Much less a 175B parameter model like GPT3. Or GPT4, whatever that model truly is. Or whatever behemoth of a model the NSA almost certainly has and just hasn’t told anyone about. I’ll eat my sock if the NSA doesn’t have a monster LLM along with a myriad of other special purpose models by now.
And even though the research has (mostly) been public so far, the resources needed to train these massive models is out of reach for all but the most privileged. We can train a GPT2 or GPT-Neo if we’re dedicated, but you and I aren’t training an open version of GPT4.
AI is more than just ChatGPT.
When we talk about reinterpreting copyright law in a way that makes AI training essentially illegal for anything useful, it also restricts smaller and potentially more focused networks. They’re discovering that smaller networks can perform very well (not at the level of GPT-4, but well enough to be useful) if they’re trained in a specific way where reasoning steps are spelled out in the training.
Also, there are used nvidia cards currently selling on Amazon for under $300 with 24 gigs of ram and AI performance almost equal to a 3090, which puts group-of-experts models like a smaller version of GPT-4 within reach of people who aren’t ultra-wealthy.
There’s also the fact that there are plenty of companies currently working on hardware that will make AI significantly cheaper and more accessible to home users. Systems like ChatGPT aren’t always going to be restricted to giant data centers, unless (as some people really want) laws are passed to prevent that hardware from being sold to regular people.
Seriously, the average person has two FAR more immediate problems than not being able to create their own AI:
-
Losing their livelihood to an AI.
-
Losing their life because an AI has been improperly placed in a decision making position because it was sold as having more capabilities than it actually has.
1 could be solved by severe and permanent economic reforms, but those reforms are very far away. 2 is also going to need legal restrictions on what jobs an AI can do, and restrictions on the claims that an AI company can make when marketing their product. Possibly a whole freaking government agency designated for certifying AI.
Right now, it’s in our best interest that AI production is slowed down and/or prevented from being deployed to certain areas until we’ve had a chance for the law to catch up. Copyright restrictions and privacy laws are going to be the most effective way to do this, because it will force the companies to go back and retrain on public domain and prevent them from using AI to wholesale replace certain jobs.
As for the average person who has the computer hardware and time to train an AI (bear in mind Google Bard and Open AI use human contractors to correct misinformation in the answers as well as scanning), there is a ton of public domain writing out there.
The endgame, though, is to stop scenario 1 and scenario 2, and the best way to do that is any way that forces the people who are making AI to sit down and think about where they can use the AI. Because the problem is not the speed of AI development, but the speed of corporate greed. And the problem is not that the average person LACKS access to AI, but that the rich have TOO much access to AI and TOO many horrible plans about how to use it before all the bugs have been worked out.
Furthermore, if they’re using people’s creativity to make a product, it’s just WRONG not to have permission or to not credit them.
Losing their life because an AI has been improperly placed in a decision making position because it was sold as having more capabilities than it actually has.
I would tend to agree with you on this one, although we don’t need bad copyright legislation to deal with it, since laws can deal with it more directly. I would personally put in place an organization that requires rigorous proof that AI in those roles is significantly safer than a human, like the FDA does for medication.
As for the average person who has the computer hardware and time to train an AI (bear in mind Google Bard and Open AI use human contractors to correct misinformation in the answers as well as scanning), there is a ton of public domain writing out there.
Corporations would love if regular people were only allowed to train their AIs on things that are 75 years out of date. Creative interpretations of copyright law aren’t going to stop billion- and trillion-dollar companies from licensing things to train AI on, either by paying a tiny percentage of their war chests or just ignoring the law altogether the way Meta always does, and getting a customary slap on the wrist. What will end up happening is that Meta, Alphabet, Microsoft, Elon Musk and his companies, government organizations, etc. will all have access to AIs that know current, useful, and relevant things, and the rest of us will not, or we’ll have to pay monthly for the privilege of access to a limited version of that knowledge, further enriching those groups.
Furthermore, if they’re using people’s creativity to make a product, it’s just WRONG not to have permission or to not credit them.
Let’s talk about Stable Diffusion for a moment. Stable Diffusion models can be compressed down to about 2 gigabytes and still produce art. Stable Diffusion was trained on 5 billion images and finetuned on a subset of 600 million images, which means that the average image contributes 2B/600M, or a little bit over three bytes, to the final dataset. With the exception of a few mostly public domain images that appeared in the dataset hundreds of times, Stable Diffusion learned broad concepts from large numbers of images, similarly to how a human artist would learn art concepts. If people need permission to learn a teeny bit of information from each image (3 bytes of information isn’t copyrightable, btw), then artists should have to get permission for every single image they put on their mood boards or use for inspiration, because they’re taking orders of magnitude more than three bytes of information from each image they use for inspiration on a given work.
Except an AI is not taking inspiration, it’s compiling information to determine mathematical averages.
A human can be inspired because they are a human being. A Large Language Model cannot. Stable Diffusion is not near the complexity of a human brain. Just because it does it faster doesn’t mean it’s doing it the same way. Human beings have free will and a host of human rights. A human being is paid for the work they do, an AI program’s creator is paid for the work it did. And if that creator used copyrighted work, then he should be having to get permission to use it, because he’s profitting off this AI program.
I would tend to agree with you on this one, although we don’t need bad copyright legislation to deal with it, since laws can deal with it more directly. I would personally put in place an organization that requires rigorous proof that AI in those roles is significantly safer than a human, like the FDA does for medication.
I would too, but we need TIME to get that done and right now, lawsuits will buy us time. That was the point of my comment.
Except an AI is not taking inspiration, it’s compiling information to determine mathematical averages.
The AIs we’re talking about are neural networks. They don’t do statistics, they don’t have databases, and they don’t take mathematical averages. They simulate neurons, and their ability to learn concepts is emergent from that, the same way the human brain is. Nothing about an artificial neuron ever takes an average of anything, reads any database, or does any statistical calculations. If an artificial neural network can be said to be doing those things, then so is the human brain.
There is nothing magical about how human neurons work. Researchers are already growing small networks out of animal neurons and using them the same way that we use artificial neural networks.
There are a lot of “how AI works” articles in there that put things in layman’s terms (and use phrases like “statistical analysis” and “mathematical averages”, and unfortunately people (including many very smart people) extrapolate from the incorrect information in those articles and end up making bad assumptions about how AI actually works.
A human being is paid for the work they do, an AI program’s creator is paid for the work it did. And if that creator used copyrighted work, then he should be having to get permission to use it, because he’s profitting off this AI program.
If an artist uses a copyrighted work on their mood board or as inspiration, then they should pay for that, because they’re making a profit from that copyrighted work. Human beings should, as you said, be paid for the work they do. Right? If an artist goes to art school, they should pay all of the artists whose work they learned from, right? If a teacher teaches children in a class, that teacher should be paid a royalty each time those children make use of the knowledge they were taught, right? (I sense a sidetrack – yes, teachers are horribly underpaid and we desperately need to fix that, so please don’t misconstrue that previous sentence.)
There’s a reason we don’t copyright facts, styles, and concepts.
Oh, and if you want to talk about something that stores an actual database of scraped data, makes mathematical and statistical inferences, and reproduces things exactly, look no further than Google. It’s already been determined in court that what Google does is fair use.
The AIs we’re talking about are neural networks. They don’t do statistics, they don’t have databases, and they don’t take mathematical averages. They simulate neurons, and their ability to learn concepts is emergent from that, the same way the human brain is.
This is not at all accurate. Yes, there are very immature neural simulation systems that are being prototyped but that’s not what you’re seeing in the news today. What the public is witnessing is fundamentally based on vector mathematics. It’s pure math and there is nothing at all emergent about it.
If an artist uses a copyrighted work on their mood board or as inspiration, then they should pay for that, because they’re making a profit from that copyrighted work.
That’s not how copyright works, nor should it. Anyone who creates a mood board from a blank slate is using their learned experience, most of which they gathered from other works. If you were to write a book analyzing movies, for example, you shouldn’t have to pay the copyright for all those movies. You can make a YouTube video right now with a few short clips from a movie or quotes from a book and you’re not violating copyright. You’re just not allowed to make a largely derivative work.
So to clarify, are you making the claim that nothing that’s simulated with vector mathematics can have emergent properties? And that AIs like GPT and Stable Diffusion don’t contain simulated neurons?
Yes, and the math is all publicly documented.
Oh boy! Link, please!
@IncognitoErgoSum Gonna need a source on Large Language Models using neural networks based on the human brain here.
EDIT: Scratch that. I’m just going to need you to explain how this is based on the human brain functions.
I’m willing to, but if I take the time to do that, are you going to listen to my answer, or just dismiss everything I say and go back to thinking what you want to think?
Also, a couple of preliminary questions to help me explain things:
What’s your level of familiarity with the source material? How much experience do you have writing or modifying code that deals with neural networks? My own familiarity lies mostly with PyTorch. Do you use that or something else? If you don’t have any direct familiarity with programming with neural networks, do you have enough of a familiarity with them to at least know what some of those boxes mean, or do I need to explain them all?
Most importantly, when I say that neural networks like GPT-* use artificial neurons, are you objecting to that statement?
I need to know what it is I’m explaining.
@IncognitoErgoSum I don’t think you can. Because THIS? Is not a model of how humans learn language. It’s a model of how a computer learns to write sentences.
If what you’re going to give me is an oversimplified analogy that puts too much faith in what AI devs are trying to sell and not enough faith in what a human brain is doing, then don’t bother because I will dismiss it as a fairy tale.
But, if you have an answer that actually, genuinely proves that this “neural” network is operating similarly to how the human brain does… then you have invalidated your original post. Because if it really is thinking like a human, NO ONE should own it.
In either case, it’s probably not worth your time.
But, if you have an answer that actually, genuinely proves that this “neural” network is operating similarly to how the human brain does… then you have invalidated your original post. Because if it really is thinking like a human, NO ONE should own it.
I think this is a neat point.
The human brain is very complex. The neural networks trained on computers right now are more like collections of neurons grown together in a petri dish, rather than a full human brain. They serve one function, say, recognizing or generating an image or calculating some probability or deciding on what the next word should be in a sequence. While the brain is a huge internetwork of these smaller, more specialized neural networks.
No, neural networks don’t have a database and they don’t do stats. They’re trained through trial and error, not aggregation. The way they work is explicitly based on a mathematical model of a biological neuron.
And when an AI is developed that’s advanced enough to rival the actual human brain, then yeah, the AI rights question becomes a real thing. We’re not there yet, though. Still just matter in petri dishes. That’s a whole other controversial argument.
-
