It’s a simple fact that people don’t take intact images as inputs to our neural network. We don’t even see them like that – our eyes saccade around and our brain reconstructs details in the process, which is why we can miss things. At no point do we actually store a copy of the whole image, which is the concern about these models.
I doubt that any of the artists’ concerns would be diminished if ‘saccades’ and other technical artifacts unique to human vision were somehow incorporated into the training workflow. In any case, the complaint isn’t that models include exact copies of copyrighted images (they don’t), or that they produce exact copies of copyrighted images (they don’t), because either of those would be an open-and-shut case of copyright infringement.
Instead, the moral panic is that AI is learning to produce images that bear similarities to copyrighted works. We’ve been OK with human brains ‘learning’ by viewing copyrighted text and images, and imitating the styles they learned. Now that we’ve created tools that learn too, and those tools will inevitably become more integrated into our wetware, we need to decide if we’re going to treat the two differently under law, and if so, what the dividing line is. I think we’d be hard pressed to rigorously define a legal distinction that won’t be dysfunctionally archaic almost as soon as it is established.
Wow, you should really read up on this, given that it’s been demonstrated that they can. So your entire characterization about the “moral panic” is based on a mischaracterization. In any case, my point remains – the two types of “learning” are not similar and so do not need to be treated similarly. Equating them is misrepresenting things.
i"m trying to remember back to the blurry days of early crypto, and nft ponzi schemes. was there a flood of newly minted legal experts and art historians then too? i’ve kind of blocked it all out
That’s not a “moral panic”… that’s what’s happening… that people’s art is being used to train AI and sometimes replicates the art of those artist…
Because of what @chenille said above… And there are all sorts of restrictions on what people can do with the images that they see in the world, with regards to IP…
Any analysis of AI ethics that equates what a computer is doing to what a human brain is doing demonstrates a fundamental lack of understanding about how brains work, how computers work, or both.
Sure, why not. In fact, I’d go one further and say that the drm on the implants probably should self-destruct you if you think anything creative that might compete with the AI so that humans can never exceed AI in creativity!
I mean we’re inventing sci-fi dystopias based on the bullshit promises tech bros sell to the asshole kids of dudes with oil money, right? Or am I misreading the thread?
We’re not talking about self-aware systems; we’re discussing data-scraping algorithms that mimic human learning… poorly. The only ‘moral panic’ involves artists not wanting their work used without appropriate compensation. If the Mouse can protect their work, why not the guy painting custom skateboards at a kiosk in the mall?
Get back to me when you have a ‘brain’ that thinks in a format that we can tell is aware, then we’ll discuss AI rights.
Where have you seen non-trivial text to image models like DALL-E include verbatim representation of copyrighted images, or produce verbatim copies of copyrighted images? Is this really what artists see as the central problem?
For people who want to just copy images verbatim, we already have Google Image search. We don’t need novel concepts of law to regulate that. The extension of copyright-style law to viewing and learning is at issue here, with “but you’re just making a copy” thrown in as a factually incorrect distraction.
Obviously it’s not the intended use, but it does show that the algorithms manage to encode the images they are trained on – not a surprise if you understand them – which fits with it being considered theft. So no, that’s not a red herring or distortion, which is a pretty rich accusation from someone equating two completely different types of processing and then trying to make this about hypothetical cyborg rights.
Since we fundamentally don’t understand the emergent properties being demonstrated by both neurons and LLMs, I find myself in complete agreement with you.
We understand enough about both computer science and neuroscience that anyone who has spent time studying either can tell you they are fundamentally different things.
An airplane isn’t an artificial bird in any meaningful way so the legal and ethical considerations we make for birds don’t apply to airplanes. That holds true regardless of how advanced aerospace technology becomes.
Fun strawman; but if you interpret it as labor being a bit peeved that capital is attempting to replace them with their own stolen, chewed up, and regurgitated work; rather than some sort of pedagogical bigotry against machines, it makes a lot more sense.
It’s incidentally true that what the ‘AI’ guys are cooking up is vastly unlike how humans do things; but it’s not like the economics or the displeasure would change significantly(at least among the people who resent their work be appropriated; I assume that medical ethics objections would arise) if there were giant neuron culture vats being studded with electrodes and trained on what the scrapers bring in.
That’s not what we’ll end up with or what they want. They’ll be rentiers in your mind and turn off your capabilities if you don’t pay your monthly fee or for the next upgrade. And they’ll own a piece of what you produce.
Don’t get me started on how they’ll control you via what you see and effecting your emotions.
I think it’s also worth pointing out that like airplanes and birds, they’re not even trying to work the same way. Jaded found this nice diagram of GPT4:
That is a hell of a lot of neurons, but look at the arrows – they all point the same way. That’s why there’s a training phase to set the weights and then a separate application phase that uses them. So we know it is fitting a hypersurface to the input vectors, even if the actual hypersurface is going to be far too convoluted to describe otherwise.
For comparison, a nematode worm has only 300 or so neurons, but they actually feed back into each other. That adds an entirely different sort of complexity to the network. So many people keep talking about LLMs as if they were on the verge of some kind of consciousness, but they’re actually much less like it than the microscopic worm is, which can’t learn much but still learns as it goes.
