this post was submitted on 13 Jul 2024
364 points (100.0% liked)
196
16233 readers
1700 users here now
Be sure to follow the rule before you head out.
Rule: You must post before you leave.
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
This. Art is expression. Wtf is AI art expressing?
Art is aesthetic. It can express something but it doesn't have to.
Not necessarily. Not all art is visual, not all art is made to create pleasing emosion. Not all is made to be beautiful.
It doesn't have to be a pleasant aesthetic or visual. It could be anything from a full image to a font used on a resume to the choice of words used in general to the way the email address sounds if you pronounce it out loud. It can be the sequence of smells, sounds, sights, taste, and feel of a single course or five course meal.
It can be puppets designed to last generations or an explosion that exists for a brief moment.
It can even be the cleverness of how a message is woven into an otherwise meaningless looking scene.
What is it that humans are expressing when they are arting?
Whatever the artist using the AI tool is trying to express?
If I ask a painter to paint a landscape, who's making art, me or the painter?
Is the painter just a tool?
You can't really have it both ways.
Is the things just a machine that's following instructions and synthesizing its training data into different things? Then it's a tool.
Is the things making choices and interpreting your inputs to produce a result? Then it's an artist.
The painter I buy a commission from is an artist. The ai I use to generate a scene is a tool.
Was the "training data" produced by artists or tools?
I mean, yes?
That's very pithy, but the material used as training data was probably produced by artists attempting to create art using tools (ai and otherwise), as well as more mundane data designed and produced by humans with no ai tools and some produced by humans with almost exclusively ai tools.
You probably live in a different world than I do.
Don't chicken/egg this. All of the training data was man-made at some point. Until the first LLMs started outputting based on it.
Secondly, the amount of human-produced content and LLM-produced content that's in the training data is incomparable. And will continue to be so. Otherwise the models break.
An AI doesn't understand what human emotion is.
Does a paintbrush?
This is a false equivalency. Unless the paintbrush is stolen I guess. 🙄
That's unprovable without some very strict definitions, but if we take it as a given (and for the record I don't disagree, so we should) then that's why the ai isn't the artist. It's just a tool an artist could use. MS Paint isn't an artist either, and like ai neither are many of the people using it, but it still can be used to create art.
Meh, better approach it to assume it doesn't understand emotion unless proven otherwise. Does a fork understand what human emotion is? A pillow? You wouldn't assume that either I guess.
So which of us are p-zombies? We've encountered the same problem by suggesting that human beings have consciousness or self awareness, or get what qualia are, except we can't prove that anyone has any of these things. The difference of AI consciousness within its development community is a sorites paradox. Large AI packages like GPT-4 have more awareness than previous versions, but not as much awareness as humans. But it Chat GPT4 does exceed human control subjects in the Turing test.
Mind you the Turing is only one of several tests we use to rate how advanced AI is, but we can't be sure even when an AI can make coffee given a machine and supplies, and construct flat-packed furniture given the IKEA visual instructions, that this counts as AGI, or is sentient.
Right now, there are artists who use generative AI to create art, and it is as much really art as photography was really art when illustrators were complaining they are just using a machine to replicate a real scene. As much as music production and music synthesis are art.
Now yes, I get that AI presents risks of workers losing income and their capacity to survive, but every time we toss our sabots into the gearworks to break the machines, we're kicking overthrow of the system down the line, until we're where we are today, not only looking at the dissolution of our democracy so that industrialists may continue to exist, but also the destruction of our habitat, because we can't address what makes them money.
So capitalism is going to end you either way, unless you end it first. And I expect if you actually tried to make a fortune on your art, you would eventually find yourself selling out all your rights to one of the big corporate controllers, and they would own everything you did, and pay you a pittance for it... Unless you are James Hetfield kind of skilled and lucky. Somehow I doubt you are.
What about a cat, or a person who's different from you? It's just as impossible to prove, and yet...
Welcome to radical constructivism :) The question whether other people or cats can experience emotions is in fact a problem people have been thinking about quite a lot. Answers are not very satisfactory, but one way to think about it (e.g., some constructivists would do that) is that assuming they do have a conscience simplifies your world model. In the case of "AI" though, we have good alternative explanations for their behavior and don't need to assume they can experience anything.
The other important bit is that not assuming some phenomenon exists (e.g., "AI" can experience emotions) unless proven otherwise is the basis of modern (positivistic) science.
Does it? Feels more like it merely excludes them from your model, since your model cannot explain their conscience. If that simplifies your model, then you can apply the same thinking to anything you don't understand by simply saying it is similar to something else you also can't explain.
The problem with this isn't that it's literally unprovable, it's that proving it requires defining "can experience emotions" in a way everyone can agree on. Most trivial definitions that include everything we think ought obviously be included often bring in many things we often think ought be excluded, and many complicated definitions that prune out the things we think ought be excluded, often also cut out things we think should be included
Yes, in the sense that "thing moves around and does stuff" becomes more predictable if you assume a certain degree of consciousness. This is easier than "thing is at this position now, was at a different position before, was at yet another position before that". You reduce some of the complexity and unpredictability by introducing an explanation for these changes of world state. In my world, so far I worked well with the assumption that other humans and animals have some consciousness and at least I'm not aware of any striking evidence that would raise doubt on that.
Yes, that's a problem, but it's relatively similar to the other one. It's actually quite hard to "prove" anything with real world connection. However, in the case of other humans/animal consciousness, evidence suggests so (at least for me). The evidence in the case of "AI" is different, though. For example, they seem to have no awareness of time and no awareness of the world beyond the limited context of a conversation. Besides a fancy marketing term that suggests there is something similar to living beings involved, what we currently see are admittedly impressive programs that run on statistics, but I don't need to assume any "consciousness" to explain what they do.
My concern is that "consciousness" isn't so much an explanation as it is a sort of heuristic. We feel conscious and have an internal experience, so it seems pretty reasonable to say that such a thing exists, but other than one's own self, there's no point where it is certain to exist, and no clear criteria or mechanism that we can point to.
What about the p-zombie, the human person who just doesn't have an internal experience and just had a set of rules, but acts like every other human? What about a cat, who apparently has a less complex internal experience, but seems to act like we'd expect if it has something like that? What about a tick, or a louse? What about a water bear? A tree? A paramecium? A bacteria? A computer program?
There's a continuum one could construct that includes all those things and ranks them by how similar their behaviors are to ours, and calls the things close to us conscious and the things farther away not, but the line is ever going to be fuzzy. There's no categorical difference that separates one end of the spectrum from the other, it's just about picking where to put the line.
And yes, we have perhaps a better understanding of the mechanism behind how an ai gets from input to output than we do for a human, but it's not quite a complete one. And the mechanism for how humans get from an input to an output is similarly partially understood. We can see how the arrangement and function of nerve cells in a "brain" lead to the behaviors we see, and have even fully simulated the brains of some organisms with machine code. This is not so dissimilar from how a computational neural network is operated. The categorical difference of "well one is a computer" doesn't work when we have literally simulated an organic brain also on a computer.
It does not make sense to add the qualifier "internal" unless it is being contrasted with "external." It makes no sense to say "I'm inside this house" unless you're contrasting it with "as opposed to outside the house." Speaking of "internal experience" is a bit odd in my view because it implies there is such thing as an "external experience". What would that even be?
The p-zombie argument doesn't make sense as you can only conceive of things that are remixes of what you've seen before. I have never seen a pink elephant but I've seen pink things and I've seen elephants so I can remix them in my mind and imagine it. But if you ask me to imagine an elephant a color I've never seen before? I just can't do it, I wouldn't even know what that means. Indeed, a person blind since birth cannot "see" at all, not in their imagination, not even in their dreams.
The p-zombie argument asks us to conceive of two people that are not observably different in every way yet still different because one is lacking some property that the other has. But if you're claiming you can conceive of this, I just don't believe you. You're probably playing some mental tricks on yourself to make you think you can conceive of it but you cannot. If there is nothing observably different about them then there is nothing conceivably different about them either.
This is what Thomas Nagel and David Chalmers ask and then settles on "mammals only" because they have an unjustified mammalian bias. Like I said, there is no "internal" experience, there is just experience. Nagel and Chalmers both rely on an unjustified premise that "point-of-view" is unique to mammalian brains because supposedly objective reality is point-of-view independent and since experience clearly has an aspect of point-of-view then that means experience too must be a product purely of mammalian brains, and then demands the "physicalists" prove how non-experiential reality gives rise to the experiential realm.
But the entire premise is arbitrary and wrong. Objective reality is not point-of-view independent. In general relativity, reality literally change depending on your point-of-view. Time passes a bit faster for people standing up than people sitting down, lengths of rulers can change between observers, velocity of objects can change as well. Relational quantum mechanics goes even further and shows that all variable properties of particles depend upon point-of-view.
The idea that objective reality is point-of-view independent is just entirely false. It is point-of-view dependent all the way down. Experience is just objective reality as it actually exists independent of the observer but dependent upon the point-of-view in which they occupy. It has nothing to do with mammalian brains, "consciousness," or subjectivity. If reality is point-of-view dependent all the way down, then it is not even possible to conceive of an intelligent being that would occupy a unique point-of-view, because everything occupies their own unique point-of-view, even a rock. It's not a byproduct of the "conscious mind" but just a property of objective reality: experience is objective reality independent of the observer, but dependent upon the context of that experience.
When you go down this continuum what gradually disappears is cognition, that is to say, the ability to think about, reflect upon, be self-aware of, one's point-of-view. The point-of-viewness of reality, or more simply the contextual nature of reality, does not disappear at any point. Only the ability to talk about it disappears. A rock cannot tell you anything about what it's like to be a rock from its context, it has no ability to reflect upon the point-of-view it occupies.
Although you're right there is no hard-and-fast line for cognition, but that's true of anything in nature. There's no hard-and-fast line for anything. Take a cat for example, where does the cat begin and end, both in space in time? Create a rigorous definition of its borders. You won't be able to do it. All our conceptions are human creations and therefore a bit fuzzy. Reality is infinitely complex and we cannot deal with the infinite complexity all at once so we break it up into chunks that are easier to work with: cats, dogs, trees, red, blue, hydrogen, helium, etc. But you always find when you look at these things a little more closely that their nature as discrete "things" becomes rather fuzzy and disappears.