5 ms·

> Do you ask for permission when you train your mind on copyrighted books? […] Humans are constantly ingesting gobs of “copyrighted” insights This comment fund

by dahart 4y ago

> Do you ask for permission when you train your mind on copyrighted books? […] Humans are constantly ingesting gobs of “copyrighted” insights This comment fundamentally and dangerously misunderstands Copyright Law. Insights are not copyrighted, nor are they copyrightable. Copyright law controls who gets to distribute a specific “fixation” or performance of work. It is not, and never was about preventing the spread of ideas. Authors and artists have always intended for you to read/observe/listen to their work when you legally acquire a copy. They just want you to not copy it verbatim, but go do your own original work if you want to distribute or sell something. The whole problem is that today’s NNs are specifically designed to remember and remix only the fixed performative parts of the work, and they, unlike humans, don’t understand the insights at all. They are just deterministic machines that copy and remix at a large scale. As such, it’s pretty clear the people training AI today should expect to have to ask permission before “training” (copying) other people’s work.

jonkho 4y ago
So if I visually look at a piece of work and “run” the NN training algorithm in my meat-space brain or even on pen and paper, do I need to ask for permission for doing so? Or is permission required only if silicon chips “run” the algorithm? Asking for a friend.
- dahart 4y ago
  This isn’t what the comment I replied to was suggesting, which is important because the NN training algorithm isn’t how humans observe creative work, nor how we make insights. But, yes, the same standards apply whether your deterministic machine is silicon based, or meat based. Copyright law applies to both. If you reproduce significant parts of a fixed performative work, then you are in violation of the law. If your algorithm mixes enough snippets from enough sources, then it’s hard to tell, and you probably won’t get caught, but it doesn’t really change the fact that you’re mechanically copying. FWIW, copyright precedent in music seems to allow human-made remixes that involve multiple sources, as long as the work as a whole is original and doesn’t reproduce significant parts of the sources.
  - jonkho 4y ago
    You are loading up my question with your own assumptions. My point specifically is if I’m just observing a piece of work and running the NN algorithm in my brain, does this constitute as illegal thought then? Do my thoughts violate rights? Note that I am not “reproducing” anything (whatever that term means). I am just observing the work and running the algorithm in my brain while silently sitting.
    - dahart 4y ago
      What assumptions are you referring to? It doesn’t seem like you understand Copyright Law, so that’s why I keep trying to explain it. Under Copyright Law, you have to acquire material legally, and it’s illegal to distribute copies you made to other people. If you’re executing a NN algorithm in your mind, or via pen & paper, then you are copying from the training samples, because that’s what the algorithm does. During training you compute errors against the samples, and update your weights to reduce error. During inference or generation, you use the weights (the results you remembered across all your training data) to produce an output. When your training samples are clustered in the latent space, the network will only remember an average of the samples, but samples that are sparse and don’t have close neighbors are sometimes remembered verbatim because there’s nothing nearby to average from. You can legally run the algorithm all you want on your own. Once you run it and then distribute the output, it might be in violation of Copyright Law if you accidentally reproduced one of the samples. Same is true for traditional human learning, you can free copy ideas legally, but reproducing too closely something that someone else made may be against the law, even if it was accidental.
      - jonkho 4y ago
        So we are in agreement that it is not violating copyright laws to run the algorithm on copyrighted works to produce the model, because if it is, my thoughts could be illegal too. In the end only actions such as reproducing the work and distributing it can be a violation. In other words, the end user of the model is the one to be held responsible if they reproduce and distribute the copyrighted material.
feliixh 4y ago
I don't think that's so clear. When you train a deep learning model you are making it extract the gist or insight of many works and then use that pattern to produce new works. While the NN does not experience the work like a human it is definitely not memorizing. A silly example. Making GPT write a rap battle between Keynes and Mises goes beyond a performative remix, it is transformational work, nothing is copied explicitly. If a human were to write it that would not violate copyright. I think that to tackle this we need a new lens other than copyright in the long term.
- dahart 4y ago
  You’re right that it’s not so clear, perhaps I overstated for brevity. I don’t actually think requesting permission is absolutely necessary, what I really think is that there aren’t good reasons AI people shouldn’t at least first try to establish training sets that are unambiguously legal, either through use of public domain work, or through an actual attempt to curate licensing models that allow re-use. We have plenty of precedent for doing this, so people claiming they should have access to everything without permission strikes me as lazy. There’s also the problem that the AI winners already are, and will continue to be, the monopoly tech and media companies who stand to make handsome profits off of the results of their trained networks. Even if you believe the results of their tech is “transformational”, there is no question that it wouldn’t work at all without access to the source material. The argument that NNs aren’t memorizing is definitely debatable and not necessarily true. They are designed to memorize deltas and averages from examples. They are, at the most fundamental level, building high dimensional splines to approximate their training data, and intentionally trying to minimize the error between the output and the examples. It’s fair to say that “usually” they don’t remember any single training sample, but it’s very easy for NNs to accidentally remember outliers verbatim. The whole reason the lawsuits mentioned in the article are happening is because we keep finding more and more examples where the network has reproduced someone’s specific work in large part. If we’re going to claim that today’s AI is producing original work, then we have to guarantee it, not just assert that it doesn’t usually happen. > a rap battle between Keynes and Mises goes beyond a performative remix, it is a transformational work, nothing is copied explicitly. I don’t buy that the work can be called transformational just because the remix doesn’t have any recognizable snippets. GPT is in fact copying individual words explicitly, and it’s putting words together by studying the statistical occurrence of words in context of other words. > I think that to tackle this we need a new lens other than copyright I totally agree with that. This question is legitimately hard. We do need a new lens, but we might have to keep and respect the old one too at the same time. I feel like AI work should acknowledge that difficulty and step up to lead the curation of training sets that are legal wrt copyright by design, rather than ignoring the concerns of the very people who made the work they are leveraging.

by dahart 4y ago

jonkho 4y ago

So if I visually look at a piece of work and “run” the NN training algorithm in my meat-space brain or even on pen and paper, do I need to ask for permission for doing so? Or is permission required only if silicon chips “run” the algorithm? Asking for a friend.

dahart 4y ago
This isn’t what the comment I replied to was suggesting, which is important because the NN training algorithm isn’t how humans observe creative work, nor how we make insights. But, yes, the same standards apply whether your deterministic machine is silicon based, or meat based. Copyright law applies to both. If you reproduce significant parts of a fixed performative work, then you are in violation of the law. If your algorithm mixes enough snippets from enough sources, then it’s hard to tell, and you probably won’t get caught, but it doesn’t really change the fact that you’re mechanically copying. FWIW, copyright precedent in music seems to allow human-made remixes that involve multiple sources, as long as the work as a whole is original and doesn’t reproduce significant parts of the sources.
- jonkho 4y ago
  You are loading up my question with your own assumptions. My point specifically is if I’m just observing a piece of work and running the NN algorithm in my brain, does this constitute as illegal thought then? Do my thoughts violate rights? Note that I am not “reproducing” anything (whatever that term means). I am just observing the work and running the algorithm in my brain while silently sitting.
  - dahart 4y ago
    What assumptions are you referring to? It doesn’t seem like you understand Copyright Law, so that’s why I keep trying to explain it. Under Copyright Law, you have to acquire material legally, and it’s illegal to distribute copies you made to other people. If you’re executing a NN algorithm in your mind, or via pen & paper, then you are copying from the training samples, because that’s what the algorithm does. During training you compute errors against the samples, and update your weights to reduce error. During inference or generation, you use the weights (the results you remembered across all your training data) to produce an output. When your training samples are clustered in the latent space, the network will only remember an average of the samples, but samples that are sparse and don’t have close neighbors are sometimes remembered verbatim because there’s nothing nearby to average from. You can legally run the algorithm all you want on your own. Once you run it and then distribute the output, it might be in violation of Copyright Law if you accidentally reproduced one of the samples. Same is true for traditional human learning, you can free copy ideas legally, but reproducing too closely something that someone else made may be against the law, even if it was accidental.
    - jonkho 4y ago
      So we are in agreement that it is not violating copyright laws to run the algorithm on copyrighted works to produce the model, because if it is, my thoughts could be illegal too. In the end only actions such as reproducing the work and distributing it can be a violation. In other words, the end user of the model is the one to be held responsible if they reproduce and distribute the copyrighted material.

feliixh 4y ago

I don't think that's so clear. When you train a deep learning model you are making it extract the gist or insight of many works and then use that pattern to produce new works. While the NN does not experience the work like a human it is definitely not memorizing. A silly example. Making GPT write a rap battle between Keynes and Mises goes beyond a performative remix, it is transformational work, nothing is copied explicitly. If a human were to write it that would not violate copyright. I think that to tackle this we need a new lens other than copyright in the long term.

dahart 4y ago
You’re right that it’s not so clear, perhaps I overstated for brevity. I don’t actually think requesting permission is absolutely necessary, what I really think is that there aren’t good reasons AI people shouldn’t at least first try to establish training sets that are unambiguously legal, either through use of public domain work, or through an actual attempt to curate licensing models that allow re-use. We have plenty of precedent for doing this, so people claiming they should have access to everything without permission strikes me as lazy. There’s also the problem that the AI winners already are, and will continue to be, the monopoly tech and media companies who stand to make handsome profits off of the results of their trained networks. Even if you believe the results of their tech is “transformational”, there is no question that it wouldn’t work at all without access to the source material. The argument that NNs aren’t memorizing is definitely debatable and not necessarily true. They are designed to memorize deltas and averages from examples. They are, at the most fundamental level, building high dimensional splines to approximate their training data, and intentionally trying to minimize the error between the output and the examples. It’s fair to say that “usually” they don’t remember any single training sample, but it’s very easy for NNs to accidentally remember outliers verbatim. The whole reason the lawsuits mentioned in the article are happening is because we keep finding more and more examples where the network has reproduced someone’s specific work in large part. If we’re going to claim that today’s AI is producing original work, then we have to guarantee it, not just assert that it doesn’t usually happen. > a rap battle between Keynes and Mises goes beyond a performative remix, it is a transformational work, nothing is copied explicitly. I don’t buy that the work can be called transformational just because the remix doesn’t have any recognizable snippets. GPT is in fact copying individual words explicitly, and it’s putting words together by studying the statistical occurrence of words in context of other words. > I think that to tackle this we need a new lens other than copyright I totally agree with that. This question is legitimately hard. We do need a new lens, but we might have to keep and respect the old one too at the same time. I feel like AI work should acknowledge that difficulty and step up to lead the curation of training sets that are legal wrt copyright by design, rather than ignoring the concerns of the very people who made the work they are leveraging.