Speaking as a creative who also has gotten paid for creative work, I’m a bit flustered at how brazenly people just wax poetic about the need for copyright law, especially when the creator or artist them selves are never really considered in the first place.

It’s not like yee olde piracy, which can even be ethical (like videogames being unpublished and almost erased from history), but a new form whereby small companies get to join large publishers in screwing over the standalone creator - except this time it isn’t by way of predatory contracts, but by sidestepping the creator and farming data from the creator to recreate the same style and form, which could’ve taken years - even decades to develop.

There’s also this idea that “all work is derivative anyways, nothing is original”, but that sidesteps the points of having worked to form a style over nigh decades and making a living off it when someone can just come along and undo all that with a press of a button.

If you’re libertarian and anarchist, be honest about that. Seems like there are a ton of tech bros who are libertarian and subversive about it to feel smort (the GPL is important btw). But at the end of the day the hidden agenda is clear: someone wants to benifit from somebody else’s work without paying them and find the mental and emotional justification to do so. This is bad, because they then justify taking food out of somebody’s mouth, which is par for the course in the current economic system.

It’s just more proof in the pudding that the capitalist system doesn’t work and will always screw the labourer in some way. It’s quite possible that only the most famous of artists will be making money directly off their work in the future, similarly to musicians.

As an aside, Jay-Z and Taylor Swift complaining about not getting enough money from Spotify is tone-deaf, because they know they get the bulk of that money anyways, even the money of some account that only plays the same small bands all the time, because of the payout model of Spotify. So the big ones will always, always be more “legitimate” than small artists and in that case they’ve probably already paid writers and such, but maybe not… looking at you, Jay-Z.

If the copyright cases get overwritten by the letigous lot known as corporate lawyers and they manage to finger holes into legislation that benifits both IP farmers and corporate interests, by way of models that train AI to be “far enough” away from the source material, we might see a lot of people loose their livelihoods.

Make it make sense, Beehaw =(

  • A1kmm@lemmy.amxl.com
    link
    fedilink
    English
    arrow-up
    3
    ·
    1 year ago

    If AI generated art is a close derivative of another work, then copyright already applies.

    But when it comes to vague abstractions over multiple works that isn’t like any one of them, copyright is probably not the right fix for what is fundamentally a more general problem. Copyright has never covered that sort of thing, so you would be asking for an unprecedented expansion to copyright, and that would have immense negative consequences that would do more harm than good.

    There are two ways I could see in which copyright could be extended (both of which are a bad idea, as I’d explain).

    Option 1 would be to take a ‘colour of bits’ approach (borrowing the terminology from https://ansuz.sooke.bc.ca/entry/23). The analogy of ‘bits’ having a colour and not just being a 0 or 1 has been used to explain how to be conservative about ensuring something couldn’t possibly be a copyright violation - if a bit that is coloured with copyright is used to compute another bit in any way (even through combination with another untainted bit), then that bit is itself coloured with copyright. The colour of bits is not currently how copyright law works, but it is a heuristic that is overly conservative right now of how to avoid copyright violation. Theoretically the laws around copyright and computing could change to make the colour of bits approach the law. This approach, taken strictly, would mean that virtually all the commercial LLMs and Stable Diffusion models are coloured with the copyrights of all inputs that went into them, and any output from the models would be similarly coloured (and hence in practice be impossible to use legally).

    There are two major problems with this: firstly, AI models are essentially a rudimentary simulation of human thinking (neural networks are in fact inspired by animal neurons). Applying the same rule to humans would mean that if you’ve ever read a copyrighted book, everything you ever say, write, draw or otherwise create after that is copyright to the author of that book. Applying a different rule to computers than to humans would mean essentially ruling out ever automating many things that humans can do - it seems like an anti-tech agenda. Limiting technology solely for the benefit some people now seems short sighted. Remember, once people made their livelihoods in the industry of cutting ice from the arctic and distributing it on ships for people to keep their food cold. Their made their livelihoods lighting gas lamps around cities at dawn and extinguishing them at dusk. Society could have banned compressors in refrigerators and electric lighting to preserve those livelihoods, but instead, society advanced, everyone’s lives got better, and people found new livelihoods. So a colour of bits approach either applies to humans, and becomes an unworkable mess of every author you’ve ever read basically owns all your work, or it amounts to banning automation in cases where humans can legally do something.

    The second problem with the colour of bits approach is that it would undermine a lot of things that we have already been doing for decades. Classifiers, for example, are often trained on copyrighted inputs, and make decisions about what category something is in. For example, most email clients let you flag a message as spam, and use that to decide if a future message is spam. A colour of bits approach would mean the model that decides whether or not a message is spam is copyright to whoever wrote the spam - and even the Yes/No decision is also copyright to them, and you’d need their permission to rely on it. Similarly for models that detect abuse or child pornography or terrorist material on many sites that accept user-generated content. Many more models that are incredibly important to day-to-day life would likely be impacted in the same way - so it would be incredibly disruptive to tech and life as we know it.

    Another approach to extending copyright, also ill-advised, would be to extend copyright to protect more general elements like ‘style’, so that styles can be copyrighted even if another image doesn’t look the same. If this was broadened a long way, it would probably just lead to constant battles between artists (or more likely, studios trying to shut down artists), and it is quite likely that no artist could ever publish anything without a high risk of being sued.

    So copyright is probably not a viable solution here, so what is? As we move to a ‘post-scarcity’ economy, with things automated to the extent that we don’t need that many humans working to produce an adequate quality of life for everyone, the best solution is a Universal Basic Income (UBI). Everyone who is making something in the future and generating profits is almost certainly using work from me, you, and nearly every person alive today (or their ancestors) to do so. But rather than some insanely complex computation about who contributed the most that becomes unworkable, just tax all profit to cover it, and pay a basic income to everyone. Then artists (and everyone else) can focus on meaning and not profit, knowing they will still get paid the UBI no matter what, and contribute back to the commons, and copyright as a concept can be essentially retired.