Why are humans and AI often treated differently in cases where they perform nearly identical processes?

Question

With respect to AI, some people appear to have an objection to the idea of

feed[ing an] AI with other people's works and then claim[ing] all the output as yours.

Let's create the following hypothetical scenario.

An AI and a human art student are both presented with the exact same set of copyrighted pictures to study. After studying them both produce a new work of art (a picture) based on what they have learned. Both new works are different from any of the originals but have some similarities in style or techniques. Both new works are based on the same source material and are theretofore very similar.

Many people would call what the AI did copyright violation or insist that every work used to train it be cited.

But very few people would have an issue with what the art student did.

Whether it be a human or a machine, in both cases, Neural networks examined objects and then created output based on what was learned. From an ethical standpoint one would think that the form of the machine should make little difference (gears, vacuum tubes, transistors or chemical machinery of living cells).

Neural networks are heavily modeled on the way human brains work. So even if the form of the machine did matter, one would think that it should matter much less for neural network based Artificial Intelligence (as opposed to a traditional computer program).

So why try and legally disincentivize one but not the other?

Why do many people choose to make that distinction?

From an ethical standpoint shouldn't the two be identical equivalent (or nearly so)?

I do acknowledge that some content creators may think they have a financial interest in limiting the use of AI. But aren't they just making an arbitrary moral distinction based on their own financial interest?

I acknowledge that an AI is not a human being. But as AI progresses towards the point of sentience, then at some point, what we are doing just becomes discrimination.

Answer 1

The training sets for generative AI systems are orders of magnitude larger than the number of images or words that a human being sees in a lifetime. If you trained a neural net on only the images that the art student studied, it most likely would produce an obvious, slavish copy of the input, if its output was coherent at all. If the art student did the same, they probably would be accused of plagiarism. As the number of images in the training set increases far beyond human capacity, it becomes harder to identify the contribution of each one, but the basic operating principles are the same.

People are able to extrapolate from fewer examples because there are other things going on in the brain that aren't modeled in the generative AI systems. Those other things are why all that art exists in the first place.

In many ways, your question assumes its conclusion. If the art student and the generative AI, trained on the same images, produced similar output, then it would be unreasonable to treat them differently. If the ethical concerns were about the "form of the machine", rather than about a large number of small-time artists losing commissions and the profits that formerly went to them going instead to a small number of wealthy corporations who flaunted copyright law in order to train their models on the work of the independent artists without their permission, then perhaps there would be no logic in the ethical concerns. If progress in AI research leads at some point to an AI that is obviously a "person", then at some point it becomes unreasonable to not treat it as a person.

Answer 2

Neural networks are not modelled on the way the brain works because no one knows how the brain works. At best a neural network is an extremely limited and highly idealized simulation of a model based on a rather controversial theory of how the brain works. At worst the term "neural network" is just a marketing buzzword with no real connection to brain theory.

Furthermore, there is no reason to assume that modeling the brain would properly simulate a mind anyway except for materialists, and not everyone is a materialist. And even for materialists, it's a bizarre sort of theory in which a simulation is supposed to somehow generate a real-world effect of what is simulating. No one thinks that if you simulate gravity in a physics engine on a computer that it will generate real-world gravity, so even if the mind is just an aspect of the brain, why would simulating a brain on a computer generate a real-world mind?

From another point of view, there is no reason to think what happens inside an ML program when it reproduces patterns from its training data is anything remotely like what happens inside a human mind when it creates new things based on examples. The criteria of success used by the programmers is not that the program mimics the mind but that it mimics behavior. They aren't even really aiming at the mind, so there is no reason they have successfully simulated the mind.

The aspects that the human notices are likely to be much different from the aspects that the AI stores in its model. The human draws associations to other things in his life, which the computer does not do. In language-based jobs the human is working based on the meanings of what he has read, not the text, whereas (at least in current technology) the computer does not even try to model meaning; it only models text patterns.

In general, there is no good reason to believe the premise of your question and some very good reasons to doubt it. If you want to prove that computer-generated content is produced by the same process as human-based content, you have a great deal of work to do.

ADDENDUM

In response to a lengthy critique below, allow me to clarify a few things. First, the brain contains neurons and those neurons can trigger each other into firing. That's basic brain anatomy. However, that's not the same as saying that the neurons in the brain work together to constitute any sort of computing device, much less the specific type of computing device called a neural network. People working in AI seem to think the notion that the neurons constitute a neural network is uncontroversial, but it's not uncontroversial among anatomists and others who study actual brains.

Materialism suggests that the mind arises as some sort of effect of the brain; it does not imply that the effect that gives rise to mind is something that can be simulated on a computer. There is no good reason (that I've ever heard) to ignore the difference between living cells and dead silicon, no reason to discount the effects of the biological systems, no reason to think that all that is needed is some sort of formal analogy between how a brain operates and how a computer program operates. No matter how well you simulate how the brain operates, there is no good reason to think that a mind will pop out of it, just like no matter how well you simulate how gravity operates, there is no reason to think that you will suddenly be pulled towards the computer. The reason this assumption is so common among AI researchers and philosophers (I claim) is because the assumption is necessary for their research. If they can't make that assumption, they have nowhere to go.

Me pointing out lack of evidence for your assumptions is not the same as making assumptions of my own. Nothing in the above discussion assumes that there is anything special about the mind. I made one small point about how not everyone is a materialist, but the rest of my argument is perfectly compatible with a materialist world view. As I said, being a materialist does not imply that you have to have any particular beliefs about exactly how the mind arises from the brain.

Answer 3

The premise is flawed because humans and AIs do not use the same processes, at least currently.

As Davie Gudeman's answer already points out, humans and AIs at least at present do not use the same processes. (I do take slight issues with his contention that "neural network" might just be a marketing buzzword. It was used in academia to describe certain types of programmed systems long before there was anything like a marketable product for there to be a buzzword around.)

For one things, while art tends to build on prior art and humans tend to be inspired by art they see, a human will bring other things into their art that an AI, at present, cannot. The AI systems that generate images do not draw inspiration from nature at least unless that nature was reduced to a carefully curated image and fed into it as art. No present AI system can draw inspiration from mood or emotion, because they lack such a thing.

AI systems also lack intentionality. Now, I have to admit that in certain deterministic forms of philosophy or if philosophical zombies were to actually exist, it might be argued that humans lack intentionality at least in the sense we want to believe they have it. But humans seem to possess intentionality while AI does not give off the same sense of intentionality.

If a truly sentient AI were ever developed, it is likely that the law, at least in a system similar to the US, would recognize them as being effectively human.

Currently, there is a clear distinction between an AI and a human. No AI has intentionality, no AI has emotions or feelings, no AI can put in effort which at least theoretically was the basis for copyright in the first place.

If an AI ever became truly sentient and displayed all of those things, then it is likely that law, at least in any system similar to what is presently in place in the US, would recognize it as a "person" as that term is defined in law. Notably, person for law already includes corporations, but a corporation is treated as acting through its officers and those officers have emotions, intentionality, and exert effort. It would be a small stretch to extend the definition to include a fully sentient and sapient AI.

Incidentally, the blog the Law and the Multiverse addresses this issue repeatedly, with Non-Human Intelligences III: Categories perhaps being a strong example. While I sometimes disagree with his conclusions, Ryan Davidson is a lawyer and provides thoughtful analysis. While entirely fictional and showing its age now, Asimov's The Bicentennial Man is essentially a meditation on this question.

Also, whether or not what ChatGPT and other AIs presently do is copyright violation is very much an open question under the law. My off the cuff suspicion is that it will be found to depend on the circumstances but will generally be found to be protected by fair use, but it remains to be seen and I have not taken the time to do a full analysis. (If I find that time, I will likely submit it to a law journal before mentioning it on stack exchange.). It is presently fairly well settled that what comes out of AI cannot be registered for copyright. I wrote that up as a blog post myself a couple of weeks ago. But that turns on the current understanding of the word "author". If an AI is ever recognized as sentient and sapient, the analysis may well change.

Answer 4

Many people, whether justified or not, whether consciously or not, believe that (human) brains are fundamentally special, that we are capable of "creativity", that we have "free will", that we might have "souls", and so forth.

Computers, on the other hand, computers are just 1s and 0s. They take some input, perform some calculations that we told it to do, and pop out some output.

Given those positions, it's not hard to see why there is a distinction between the two.

It might also be worth looking to more primitive algorithms. If we were to, for example, take a picture and simply increase one colour of every pixel by some small constant, this is still very close to the original, and would be a clear copyright violation. The things you can do in Photoshop is much more complex than that, but most of that will still classify as copyright violations. AI is still very new, so people just group it with all the other algorithms by default. For them to be convinced otherwise, one would need to make that case.

* If a person applies a sequence of algorithms to an image, that might constitute a transformative enough change to classify as a new piece of art. But in that case, the person would be deciding on the sequence, so that part would still be seen as human, not algorithmic.

---

As for whether brains are fundamentally special, materialists tend not to accept that. But materialism isn't that common among laypeople (most people are still theists, and theists tend to not be materialists).

I expect even materialists would draw a distinction between the brain and AI, given how much more complex the brain is. Whether this distinction is justified here is hard to say. Based on how different many of the images are from the originals, I'd say it's probably not justified to put that on the super-simple copyright-violation end of that (at best, it's on some blurry line in the middle).

---

What might complicate matters is that some AI-generated images were singled out for being very similar to the originals, in ways that would probably classify as copyright violations if a human produced those, from the original images.

So this certainly reinforces the idea that AI just isn't complex enough to classify as transformative (even if this is a bit of cherry-picking).

What complicates matters even further is that AI can produce a functionally-infinite amount of art. Quite a few artists are worried about their livelihood, and some consumers may dislike the reduced "specialness" of art. This doesn't say anything about whether AI is functionally transformative, but making it legal may have consequences that people don't want (on the other side of things, it could also make custom art more accessible, and allow artists to use AI to complement their work, whereas the risk of losing one's job is a common fear with many new technologies - it may be a plausible risk, but it also stops society from potentially reaching somewhere better).

(Disclaimer: not a lawyer)

Answer 5

Other answerers have already addressed the flawed premise of your question, specifically that in fact AIs and humans do not perform nearly identical processes to produce art.

I would like to address the idea that if humans and AIs performed identical processes to produce art, then the art should be treated the same way. This viewpoint is proposed in the question and conceded by @benrg's answer, which says

If the art student and the generative AI, trained on the same images, produced similar output, then it would be unreasonable to treat them differently.

I think this concession is incorrect. That is, even if humans and AIs did produce artwork of equivalent quality using equivalent processes, it would still be reasonable and right to treat the art produced by them differently in law.

The philosophical basis of copyright law, at least in the US, is described succinctly by the US Constitution, Article 1, Section 8, Clause 8 which states that the US Congress shall have the power

To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.

There is a clear mechanism by which securing copyrights for human creators could "promote the Progress of Science and the useful Arts": by giving creators a monopoly on marketing their own work, creators have a financial incentive to create works which they might not otherwise create.

In contrast, securing this right for art-generating AIs would not "promote" the creation of new work, because art-generating AIs do not respond to financial incentives. They create art whenever a button is clicked. We also don't need to incentivise humans to click it more, because whoever wants to consume more AI-created art can click the button themselves.

The point at which AI-created art should be financially incentivised by copyright protection, is the point at which incentivising humans to create it will plausibly cause more and/or better AI-created art to exist for those who want to consume it. That point is subject to debate, but in current law it is based on the human having creative input into the process.

In summary, we treat human-created and AI-created art differently in law because extending copyright protection to AIs would not have the same useful consequences as extending it to humans is supposed to.

Answer 6

You might be overthinking this as there is nothing deeply philosophical about why human beings and digital computers aren't both treated as persons in the philosophical sense. Current AI systems aren't remotely close to being people. In philosophy, this belonging to the category is known as personhood. From WP:

Personhood is the status of being a person. Defining personhood is a controversial topic in philosophy and law and is closely tied with legal and political concepts of citizenship, equality, and liberty. According to law, only a legal person (either a natural or a juridical person) has rights, protections, privileges, responsibilities, and legal liability.

Personhood is more than whether or not we can do arithmetic or write English. Personhood is a far deeper concept that revolves around how we fundamentally conceptualize human beings, animals, and machines. In fact, history has shown that often other human beings are objectified and not treated as people. Slavery in the US is a shining example of such behavior after computers were invented. Our brains are wired with in-group, out-group logic to begin with. Simply put, evolution hasn't equipped us with large portions of our brains to empathize and emulate and socialize with machines. That being said, it should be noted that at this point, the most sophisticated AI in the most sophisticated packaging at best might arouse the anxiety that comes from the uncanny valley.

Now, that being said, if AI becomes AGI in the sense that the intelligence is truly-human level, encompases affective computing, an is put in a package that is much closer to our biological bodies, then you have uncharted territory. Of course, that's the premise of movies like I Robot, AI, and Bicentennial Man.

Answer 7

By "AI" I'll presume that you mean robots: chatbots and other generative tools. Then the answer is simple: robots are mechanical replacements for slaves; where a slave is a human owned as property, a robot is an embodied algorithm owned as property. In societies which permit/endorse slavery, slaves are distinct from citizens and do not have a full palette of human rights. Similarly, in societies with robots, robots are distinct from humans and do not have rights. The word "robot" was originally introduced to English to describe forced laborers.

Answer 8

I agree with the OP that humans adapt things from each other all the time. In a research context we (usually) cite our sources, but the same concern does not apply to artistic adaptation.

Unlike areas where AI can present a safety risk (e.g. self-driving cars), in art the main harm from the unauthorized creation of a "sufficiently close" version of the artists work is the loss of benefits to the artist from their work (e.g., as one poster pointed out in copyright law).

I think the issue is largely that, at least for now, epistemological: we cannot audit the inputs that were actually used to produce a given human artistic work but we can (by design) do that with an AI.

Therefore, it is all too easy to show that the AI used exactly these particular artists' works to create its output, and hence that output would not have been made had it not been for those inputs.

Of course, as the OP points out (correctly) human artists do a very similar thing when they go to art school or study art -- the key is that nobody can prove (unless it's quite blatant) that they merely recombined the work of others.

Answer 9

If you can't judge from the work you're judging the one that did it.

All arguments against AI for how it works are flawed because nothing ensures that is how AI works. AI is growing and changing even now.

All arguments for Humans for how they work are flawed because nothing ensures that is how humans work. Humans are evolving and changing even now.

So in the end the only honest answer here is:

Tradition.