Avery
Disclaimer- most of this book makes the assumption that (super)human-level artificial intelligence is not possible within the near future. In this framing, most conversation about it doing so exists to serve solely to further the interests of those invested in the AI bubble. I don’t know that this assumption is as obvious to me as it is to Doctorow, and if you strongly disagree with it, then there’s probably not much in this book for you.
I don’t think I’d have enough patience to work at the EFF, and I’m grateful that there are people like Doctorow who can push on these issues. The legal distinctions made in the book frequently felt inconsistent to me, but I think it’s possible that this is more of a me problem than the book being inaccurate. I just can’t get over “code is speech” being a better reason than “secure crypto is secure” for the government to allow the use of cryptography.
There's a bit that I might need to reread that explains how, while the training inputs may have been stolen, and that theft is bad, the results derived from the theft are fine. This is sort of weird to me. I guess code/creative works are protected, but not “objective facts”, where all of those terms are super vague. I feel like it would follow that if I could get my hands on the model weights (complex but objective statistical facts about the inputs) for Mythos/Opus/etc then I should be able to legally publish them in the public domain? An avenue left unexplored by this book.
One of the book’s ideas for a path through the AI morass is that copyright only applies to works of *human* creativity, and so the result of a prompt should automatically be in the public domain.
This is pretty compelling for stopping the reverse-centaurization of the workforce, but some of the analogies given didn’t resonate with me. Some amateur human taking literally whatever photograph automatically gets copyright for it, but someone who decides to train a monkey to take a selfie doesn’t? The monkey selfie is clearly still an intentional piece of art made by the human who decided to get it to press the button. It's literally a creative human work.
It seems like it’s hard enough to draw a line here that it would be relatively unlikely, without an enormous effort, to draw the line that the prompted input isn’t enough to qualify the result as art. After all, the photographer didn’t place all the pixels, they just pointed their lens at something and pressed “submit” on the camera. Barring that, we might still end up with reverse-centaur employees tasked with tweaking just enough pictures on the outputs to qualify it as human.
Still, I think it’s a valuable endeavor to attempt to develop the conversation here past “pro” or “anti” AI into something a little more nuanced. We can admit that the recent AI developments can be useful for some things some times without saying that all of the ways they can be used would be good for humans. We can think about how to think about their usage such that they don’t destroy society. It’s hard to maintain enough hope that any of this will be possible, given the incredible resources on the side of the tech companies and investors here, but thank you to Doctorow for tirelessly pushing.
Disclaimer- most of this book makes the assumption that (super)human-level artificial intelligence is not possible within the near future. In this framing, most conversation about it doing so exists to serve solely to further the interests of those invested in the AI bubble. I don’t know that this assumption is as obvious to me as it is to Doctorow, and if you strongly disagree with it, then there’s probably not much in this book for you.
I don’t think I’d have enough patience to work at the EFF, and I’m grateful that there are people like Doctorow who can push on these issues. The legal distinctions made in the book frequently felt inconsistent to me, but I think it’s possible that this is more of a me problem than the book being inaccurate. I just can’t get over “code is speech” being a better reason than “secure crypto is secure” for the government to allow the use of cryptography.
There's a bit that I might need to reread that explains how, while the training inputs may have been stolen, and that theft is bad, the results derived from the theft are fine. This is sort of weird to me. I guess code/creative works are protected, but not “objective facts”, where all of those terms are super vague. I feel like it would follow that if I could get my hands on the model weights (complex but objective statistical facts about the inputs) for Mythos/Opus/etc then I should be able to legally publish them in the public domain? An avenue left unexplored by this book.
One of the book’s ideas for a path through the AI morass is that copyright only applies to works of *human* creativity, and so the result of a prompt should automatically be in the public domain.
This is pretty compelling for stopping the reverse-centaurization of the workforce, but some of the analogies given didn’t resonate with me. Some amateur human taking literally whatever photograph automatically gets copyright for it, but someone who decides to train a monkey to take a selfie doesn’t? The monkey selfie is clearly still an intentional piece of art made by the human who decided to get it to press the button. It's literally a creative human work.
It seems like it’s hard enough to draw a line here that it would be relatively unlikely, without an enormous effort, to draw the line that the prompted input isn’t enough to qualify the result as art. After all, the photographer didn’t place all the pixels, they just pointed their lens at something and pressed “submit” on the camera. Barring that, we might still end up with reverse-centaur employees tasked with tweaking just enough pictures on the outputs to qualify it as human.
Still, I think it’s a valuable endeavor to attempt to develop the conversation here past “pro” or “anti” AI into something a little more nuanced. We can admit that the recent AI developments can be useful for some things some times without saying that all of the ways they can be used would be good for humans. We can think about how to think about their usage such that they don’t destroy society. It’s hard to maintain enough hope that any of this will be possible, given the incredible resources on the side of the tech companies and investors here, but thank you to Doctorow for tirelessly pushing.