Hacker News Reader

>First, you need to prove that generative AI works fundamentally the same way as humans at the task of learning. Next you have to prove that it recalls information in the same way as humans.

No, you don't need to prove any of those things. They're irrelevant. You'd need to prove that the AI is itself morally (or, depending on the nature of the dispute, legally) equivalent to a human and therefore deserving of (or entitled to) the same rights and protections as a human. Since it is pretty indisputably the case that software is not currently legally equivalent to a human, you're stuck with the moral argument that it ought to be, but I think we're very far from a point where that position is warranted or likely to see much support.

You don't even need to do that. Art is an act of ontological framing.

Duchamp didn't need negotiate with the ceramic makers to make the Fountain into art.

> ou'd need to prove that the AI is itself morally (or, depending on the nature of the dispute, legally) equivalent to a human and therefore deserving of

No you don't.

A human using a computer to make art doesn't automatically lose their fair use rights as a human.

> indisputably the case that software is not currently legally equivalent to a human

Fortunately it is the human who uses the computer who has the legal rights to use computers in their existing process of fair use.

Human brains or giving rights to computers has absolutely nothing to do with the rights of human to use a camera, use photoshop, or even use AI, on a computer.

Few people are claiming that the AI itself has the same rights as a human. They are arguing that a human with an AI has the same rights as a human who doesn't have an AI.

> They are arguing that a human with an AI has the same rights as a human who doesn't have an AI.

This is the analogy I want people against AI use to understand and never forget, even if they reject the underlying premise - that should laws treat a human who uses AI for a certain purpose identically to a human who uses nothing or a non-AI tool for the same purpose.

> Few people are claiming that the AI itself has the same rights as a human.

I think that's the case as well. However, a lot of commenters on this post are claiming that an AI is similar in behavior to a human, and trying to use the behavior analogy as the basis for justifying AI training (on legally-obtained copies of copyrighted works), with the assumption that justifying training justifies use. My personal flow of logic is the reverse: human who uses AI should be legally the same as human who uses a non-AI tool, so AI use is justified, so training on legally-obtained copies of copyrighted works is justified.

I want people in favor of AI use particularly to understand your human-with-AI-to-human-without-AI analogy (for short, the tool analogy) and to avoid machine-learning-to-human-learning analogies (for short, behavior analogies). The tool analogy is based on a belief about how people should treat each other, and contends with opposing beliefs about how people should treat each other. An behavior analogy must contend with both 1. opposing beliefs about how people should treat each other and 2. contradictions from reality about how similar machine learning is to brain learning. (Admittedly, both the tool analogy and the behavior analogy must contend with the net harm AI use is having and will have on the cultural and economic significance of human-made creative works.)

> that it’s anthropomorphizing machines.

No, it's not. It's merely pointing out the similarity between the process of training artists (by ingesting publicly available works) and ML models (which ingest publicly available works).

> First, you need to prove that generative AI works fundamentally the same way as humans at the task of learning.

Given that there is no comprehensive model for how humans actually learn things, that would be an unfeasible requirement.

What a reductive way to describe learning art. The similarities are merely surface level.

> Given that there is no comprehensive model for how humans actually learn things, that would be an unfeasible requirement.

That is precisely why we should not be making this comparison.

I’m being told repeatedly that the similarities are surface level, but no one seems to be able to give an example of a deep difference

The mechinism backing human learning isn't well understood. Machine learning is considerably clearer. Imo, it's a mistake to assume they're close because ML seems to work.

So your example of a deep difference is the suggestion that there may be a difference owing to our own ignorance? That hardly seems convincing

> The similarities are merely surface level.

Then please, feel free to explain the deep differences.

> That is precisely why we should not be making this comparison.

Wrong. It's precisely why the claim "there is a big difference" doesn't have a leg to stand on. If you claim "this is different", I ask "how?" and the answer simply repeats the claim, I can apply Hitchens Razor[1] and dismiss the claim.

[1]: https://en.wikipedia.org/wiki/Hitchens%27s_razor

A person sitting in an art school/museum for a few hours ingests way more than just the art in question. The entire context is brought in too, including the artists own physical/emotional state. Arguably, the art is a miniscule component of all sensory inputs. Generative AI ingests a perfectly cropped image of just the art from a single angle with little context beyond labelling metadata.

It's the difference between reading about a place and actually visiting it.

Edit: This doesn't even touch how the act of creating something - often in a completely different context - interacts with the memories of the original work, altering those memories yet again.

And how is any of that a compelling argument regarding the assumption of a fundamental difference in the MO of learning?

Simply stating "it has more inputs" doesn't describe the human learning model, so nothing in this establishes a baseline for comparison.

We are machines. We just haven't evenly accepted it yet.

Our biology is mechanical, and lay people don't possess an intuition about this. Unless you've studied molecular biology and biochemistry, it's not something that you can easily grasp.

Our inventions are mechanical, too, and they're reaching increasing levels of sophistication. At some point we'll meet in the middle.

100% this. Labor and all these other concepts are outdated ways to interpret reality

Humans are themselves mechanical so at the end of the day none of these issues actually matter

This is the truth. And it's also the reason why this stuff will be in court until The Singularity itself. Most people will never be able to come to terms with this.

The way these ML models and humans operate are indeed quite different.

Humans work by abstracting concepts in what they see, even when looking at the work of others. Even individuals with photographic memories mentally abstract things like lighting, body kinetics, musculature, color theory, etc and produce new work based on those abstractions rather than directly copying original work (unless the artist is intentionally plagiarizing). As a result, all new works produced by humans will have a certain degree of originality to them, regardless of influences due to differences in perception, mental abstraction processes, and life experiences among other factors. Furthermore, humans can produce art without any external instruction or input… give a 5 year old that’s never been exposed to art and hasn’t been shown how to make art a box of crayons and it’s a matter of time before they start drawing.

ML models are closer to highly advanced collage makers that take known images and blend them together in a way that’s convincing at first glance, which is why it’s not uncommon to see elements lifted directly from training data in the images they produce. They do not abstract the same way and by definition cannot produce anything that’s not a blend of training data. Give them no data and they cannot produce anything.

It’s absolutely erroneous to compare them to humans, and I believe it will continue to be so until ML models evolve into something closer to AGI which can e.g. produce stylized work with nothing but photographic input that it’s gathered in a robot body and artistic experimentation.

Beautifully put. I wish this nuance was more widely understood in the current AI debate.

You're wrong in your concept of how AI/ML works. Even trivial 1980's neural networks generalize, it's the whole point of AI/ML or you'd just have a lookup-table (or, as you put it, something that copies and pastes images together).

I've seen "infographics" spread by anti-AI people (or just attention-seekers) on Twitter that tries to "explain" that AI image generators blend together existing images, which is simply not true..

It is however the case that different AI models (and the brain) generalize a bit differently. That is probably the case between different humans too. Not the least with for example like you say those with photographic memory, autists etc.

What you call creativity in humans is just noise in combination with a boatload of exposure to multi-modal training data. Both aspects are already in the modern diffusion models. I would however ascribe a big edge in humans to what you normally call "the creative process" which can be much richer, like a process where you figure out what you lack to produce a work, go out and learn something new and specific, talk with your peers, listen to more noise.. stuff like that seems (currently) more difficult for AIs, though I guess plugins that do more iterative stuff like chatgpt's new plugins will appear in media generators as well eventually..

ML generalization and human abstraction are very different beasts.

For example, a human artist would have an understanding of how line weight factors into stylization and why it looks the way it does and be able to accurately apply these concepts to drawings of things they’ve never seen in that style (or even seen at all, if it’s of something imaginary).

The best an ML model can do is mimic examples of line art in the given style within its training data, the product of which will contain errors due to not understanding the underlying principles, especially if you ask it to draw something it hasn’t seen in the style you’re asking for. This is why generative AI needs such vast volumes of data to work well; it’s going to falter in cases not well covered by the data. It’s not learning concepts, only statistical probabilities.

I know what you're saying, and for sure existing models can be difficult to force into the really weird corners of the distributions (or go outside the distributions). The text interfaces are partially to blame for this though, you can take the images into Gimp and do some crude naive modifications and bring them back and the model will usually happily complete the "out-of-distribution" ideas. The Stable Diffusion toolboxes have evolved far away from the original simple text2image interfaces that midjourney and dalle use.

The models will generalize (because that's the most efficient way of storing concepts) and you can make an argument that that means they understand a concept. Claiming "it's not learning concepts, only statistical probabilities" trivialises what a modern neural network with billions of parameters and dozens of layers is capable of doing. If a model learns how to put a concept like line width 5, 10 and 15 pixels into a continuous internal latent property, you can probably go outside this at inference at least partially.

I would argue that improving this is at this point more about engineering and less about some underlying unreconcilable differences. At the very least we learn a lot about what exactly generalization and learning means.

> The way these ML models and humans operate are indeed quite different.

Given that there is no comprehensive understanding of how human learning works, let alone how humans operate on and integrate on what they learned in a wider context...how do you know?

> Humans work by abstracting concepts in what they see

Newsflash: AI models do the same thing. That's the basis of generalization.

> ML models are closer to highly advanced collage makers that take known images and blend them together

Wrong. That's not even remotely how U-Net based diffusion models work. If you disagree, then please do show me where exactly the source images from where the "collage maker" takes the parts to "blend" together are stored. I think you'll find that image datasets on the scale of LAION will not quite fit into checkpoint files of about 2GB in size (pruned SD1.5 checkpoint in safetensors format).

The first perceptron was explicitly designed to be a trainable visual pattern encoder. Zero assumptions about potential feelings of the ghost in the machine need to be made to conclude the program is probably doing what humans studying art say they assume is happening in their head when you show both of them a series of previous artists' works. This argument is such a tired misdirection.

> What this means, is these systems will fall into different categories of law around copyright and free-use.

No they won't.

A human who uses a computer as a tool (under all the previous qualifications of fair use) is still a human doing something in fair use.

Adding a computer to the workflow of a human doesn't make fair use disappear.

A human can use photoshop, in fair use. They can use a camera. They can use all sorts of machines.

The fact that photoshop is not the same as a human brain is simply a completely unrelated non sequitur. Same applies to AI.

And all the legal protections that are offered to someone who uses a regular computer, to use photoshop in fair use, are also extended to someone who uses AI in fair use.

Yet the copyright office has already stated that getting an AI to create an image for you does not have sufficient human authorship to be copyrighted. There's already a legal distinction here between this "tool" and tools like photoshop and cameras.

It's also presumptive to assume that AI tools have these fair use protections when none of this has actually been decided in a court of law yet. There's still several unsettled cases here.

> Yet the copyright office has already stated that getting an AI to create an image for you does not have sufficient human authorship to be copyrighted.

Gotcha.

That has nothing to do with fair use though.

Also, the same argument absolutely applies to photoshop.

If someone didn't include sufficient human authorship while using photoshop, that wouldn't be copyrightable either.

Also, the ruling has no bearing on if someone using AI, while also inputting a significant amount of human authorship. Instead, it was only about the cases where there weren't much human authorship.

At no point did the copyright office disclude copyright protections from anything that used AI in any way what so ever. In fact, the copyright office now includes new forms and fields where you talk about the whole process that you did, and how you used AI, in conjunction with human authorship to create the work.

> It's also presumptive to assume that AI tools

I'm not talking about the computer. I never claimed that computer's have rights. Instead, I'm talking about the human. Yes, a human has fair use protections, even if they use a computer.

> There's still several unsettled cases here.

There is no reason to believe that copyright law will be interpreted in a significantly different way than it has been in the past.

There is long standing precent, regarding all sorts of copyright cases that involve using a computer.

Why do you have to prove that? There is no replication (except in very rare cases), how someone draws a line should not be copyrightable.

Therein lies the crux of the issue: AI is not “someone”. We need to approach this without anthropomorphizing the AI.

You are right, AI is nothing but a tool akin to a pen or a brush.

If you draw Mickey Mouse with a pencil and you publish (and sell) the drawing who is getting the blame? Is the pencil infringing the copyright? No, it's you.

Same with AI. There is nothijg wrong with using copyrighted works to train an algorithm, but if you generate an image and it contains copyrighted materials you are getting sued.

But there is. You are arguably making unauthorized copies to train.

Unauthorized copies? If the images are published on the internet how is it downloading them "unauthorized"?

Publicly available doesn't mean you have a license to do whatever you like with the image. If I download an image and re-upload it to my own art station or sell prints of it, that is something I can physically do because the image is public, but I'm absolutely violating copyright.

That's not an unautharized copy, it's unauthorized distribution. By the same metric me seeing the image and copying it by hand is also unauthorized copy (or reproduction is you will)

IANAL, but I’m pretty sure copying an image by hand is copyright violation.

So you cannot train your drawing skills by copying other people's artworks?

You can do it in private, but you can’t distribute the resulting image, let alone sell it.

Then I don't really understand your original reply. Simply copying a publicly available image doesn't infringe anything (unless it was supposed to be private/secret). Doing stuff with that image in private still doesn't constitute infringement. Distribution does, but that wasn't the topic at hand

You can train a neural network in private too and nobody will have a problem with that. The topic of discussion is commercial AI.

The most basic right protected by copyright is the right to make copies.

Merely making a copy can definitely be infringement. "Copies" made in the computing context even simply between disk and RAM have been held to be infringement in some cases.

Fair use is the big question mark here, as it acts to allow various kinds of harmless/acceptable/desirable copying. For AI, it's particularly relevant that there's a factor of the "transformative" nature of a use that weighs in favor of fair use.

The answer is "it depends". Distribution is not a hard requirement for copyright violation. It can significantly impact monetary judgements.

That said, there is also an inherent right to copy material that is published online in certain circumstances. Indeed, the physical act of displaying an image in a browser involves making copies of that image in cache, memory, etc.

If you are viewing the image on your browser on a website, you are making a local copy. That’s not unauthorized.

Companies aren't someone, yet in the US we seem to give them rights of someone.

They are someone in the eyes of the law. They just have a different set of rights.

What we actually need to prove is whether such technology is a net benefit to society all else is essentially hand waving. There is no natural right to poorly named intellectual property and even if there was such a matter would never be decided based on the outcome of a philosophical argument because we don't decide anything that way.

> such technology is a net benefit to society all else is essentially hand waving

Some might have said this about cars ... yet, here we are. Cars are definitely the opposite, except for longer-distance travel.

Our decisions are based on what we think is a net benefit not what is.

How do you measure "benefit", and what does "net" actually mean?

Well what normally happens when something new clearly doesn't exactly fit within existing laws and practices is a bunch of rich people consider whether there is more money to be made if its legal and if it is they give some portion of the money they expect to make in the first year to lawmakers, sometimes in the form of gold bars, and it becomes legal.

Sarcasm aside there is no moral right to ANY intellectual property. It's not a positive expression of a natural right its a negative imposition of restriction upon everyone else. It's a statement that if I take my pen and paper and write the same words that you now own my pen my paper and my labor. It adds friction to the distribution of knowledge, impoverishes the world, keeps some knowledge that might have come into being from ever being generated for lack of the knowledge that failed to travel and all the good that could have therefore been done, undone.

It is justifiable only if the minimal restrictions we are willing to impose on net supports the creation of works that enrich society more than the restrictions impoverish it.

I'm not sure you can effectively measure it and would as soon just see IP law excepting only part of trademark law to prevent fraudulent knock offs and scams go entirely down the crapper.

None of that made any sense.

>My issue with this line of argument is that it’s anthropomorphizing machines. It’s fine to compare how humans do a task with how a machine does a task, but in the end they are very different from each other, organic vs hardware and software logic.

There's an entire branch of philosophy that calls these assumptions into question:

https://en.wikipedia.org/wiki/Posthumanism

https://en.wikipedia.org/wiki/Antihumanism

>Martin Heidegger viewed humanism as a metaphysical philosophy that ascribes to humanity a universal essence and privileges it above all other forms of existence. For Heidegger, humanism takes consciousness as the paradigm of philosophy, leading it to a subjectivism and idealism that must be avoided.

>Processes of technological and non-technological posthumanization both tend to result in a partial "de-anthropocentrization" of human society, as its circle of membership is expanded to include other types of entities and the position of human beings is decentered. A common theme of posthumanist study is the way in which processes of posthumanization challenge or blur simple binaries, such as those of "human versus non-human", "natural versus artificial", "alive versus non-alive", and "biological versus mechanical".