7 ms·

Don't humans operate similarly? We gain knowledge through experiences. These AI models effectively condense a vast amount of experience data into weights. Consi

Don't humans operate similarly? We gain knowledge through experiences. These AI models effectively condense a vast amount of experience data into weights. Considering the global race in AI advancements, I'm skeptical about the success of these copyright claims. I do find it hypocritical that OpenAI says that other LLMs can't be trained on data generated by their LLMs.

villgax 3y ago
True
LatteLazy 3y ago
Not just humans in general, the NYT specifically relies on the fact news cannot be copyrighted. Then claims it's articles are sacred...
- dspillett 3y ago
  IIRC the distinction is that facts can't be copyrighted, a particular arrangement of facts can, particularly if something more subjective (analysis or opinion) is included. So I can write my own article about the sky being shown to appear blue much of the time, but I can't copy someone else's article about the same subject.
  - paulmd 3y ago
    How would one even go about a phonebook-style mechanical listing of facts and occurrences? You’re listing an impossibility and then saying that garden-variety connective sentences somehow make it not a factual listing. Like yes if you copy a NYT article verbatim it’s like copying a phone book ads and all, and that’s infringement. But that’s not what a LLM does, NYT doesn’t like their content being used and summarized at all, even in a rearranged form that merely relies on the factual information included in the article. That’s what they want to get paid for, and unfortunately that’s not copyrightable and OpenAI is correct they don’t have to pay for that. NYT disagrees but again, they are kinda attempting to claim copyright on the factual information because they wrote some connective sentences between.
    - dspillett 3y ago
      > How would one even go about a phonebook-style mechanical listing of facts and occurrences? The traditional trick there is to include some small amount of fake data in the directory. You know someone has copied your collection of facts instead of compiling their own because it includes your fake facts. Mapmakers have used the method for at least as long as cartography has been part of our recorded history, see https://en.wikipedia.org/wiki/Trap_street https://en.wikipedia.org/wiki/Trap_street for details. As noted in that page, the legal status of this, like many IP related issues, depends upon jurisdiction. > But that’s not what a LLM does, What does it do that means it is only summarizing factual information? While NYT effectively trying to claim copyright on facts is wrong, OpenAI claiming it can't reproduce copyrightable information while it can reproduce/summarize facts found within the same training set seems at best disingenuous.
      - paulmd 3y ago
        How are trap streets “the trick” if they are not copyrightable or enforceable under US law? Like the Supreme Court has specifically considered and dismissed this “one clever trick to keep people from copying lists of facts” because to allow it would completely undermine the idea of facts and figures not being copyrightable. > Trap streets are not copyrightable under the federal law of the United States. In Nester's Map & Guide Corp. v. Hagstrom Map Co. (1992),[3][4] a United States federal court found that copyright traps are not themselves protectable by copyright. There, the court stated: "[t]o treat 'false' facts interspersed among actual facts and represented as actual facts as fiction would mean that no one could ever reproduce or copy actual facts without risk of reproducing a false fact and thereby violating a copyright ... If such were the law, information could never be reproduced or widely disseminated." (Id. at 733) And yes the EU has the concept of “database rights” but notionally there is still supposed to be a creative step required in the selection or arrangement of records. So just a raw copy of the numbers in a telephone directory is theoretically not copyrightable, but a telephone book might be because of the creative/transformational step. It’s possible this might be such a low bar that it’s impossible to fail to clear, but, at least on paper you can’t copyright mere facts and figures either. But either way it’s generally true that simple facts and figures are not protected and trap streets are a discredited and clumsy attempt to work around this.
bmitc 3y ago
It's quite different, is it not? I don't get the analogy. These models are scanning, storing, and ingesting more material than any one human could. Not only is the method completely different, the end goal and applications are as well. The analogy basically isn't one, at all. I'm pretty upset at companies using our personal data to make gobs of money off of. I'm also upset that they're now using our knowledge work to make even more money off of. We don't exist as computational nodes for them, a free resource to exhaust. It is a completely one way street with no consent. So I am in favor of all of these companies getting a reality check.
- machdiamonds 3y ago
  Consider a mega consulting firm with millions of von Neumann-like analysts. Together, they've processed the same vast data that LLMs have, but individually, none could. It's not an LLM, but its purpose is like ChatGPT: assisting clients with their tasks. If LLMs concern you due to their data processing, would a firm like this do the same?
  - bmitc 3y ago
    Bringing up poor fitting analogies won't change my opinion.
    - sebzim4500 3y ago
      I normally consider these discussion to be more about the people reading the comments than the people writing them. You've clearly made up your mind, but others presumably haven't so I think it's good he makes these arguments, even if it looks like tilting at windmills to you.
      - bmitc 3y ago
        That's a good point, but I guess I was implicitly getting at that continually permutating poor analogies and hypotheticals isn't an interesting discussion. The fact of the matter here is that parties, such as OpenAI, are benefiting from others' knowledge work, protected or not, in a completely one-sided way and all for free. And I don't feel sorry for companies that need to build Trojan horse products, such as OpenAI and Google, in order to survive off of other people's data that they never compensate for.
mattmanser 3y ago
Humans don't have perfect recall, don't have virtually infinite storage and can't process requests in milliseconds. I wouldn't be surprised if the avenue of attack is that fair use laws are for humans, not robots, and if an AI has been trained on copyrighted data, that's not fair use. Also, don't forget that in reality what's happened is that a bunch of copyrighted text is encoded in the LLM in a way a human can't understand, but that the LLM essentially CAN understand.
- fjfuvucucuc 3y ago
  Yup, and that reproducing it is a copyright violation. Still a copyright abolitionist though. Maybe now more people will join the fight?
  - gmerc 3y ago
    Well if use the tool to reproduce copyrighted content you are violating copyright. But that’s not the primary usecase and nobody in their right mind is arguing that. The weights are not a reproduction of the content. They are capable of it but so is a photocopier a lot more and we didn’t ban those either despite them technically being a lot more useful for violation. Nah, this is expansionist doctrine and agenda for copyright - these companies are trying to copyright style and locations in latent space now.
    - mattmanser 3y ago
      Photocopiers are for personal use, training an AI is not. If you photocopy 10,000 copies of copyrighted text and starting distributing it you will get sued. It would be different if I trained my own AI, for my personal use.
      - gmerc 3y ago
        businesses use photocopiers, there’s Xerox shops, etc. Again, LLMs don’t copy so it’s not a good metaphor.
RandomLensman 3y ago
I don't think any unaided human or collective of humans could rent seek on something close to the sum total of human knowledge and expression. Irrespective of copyright issues, the question is how to avoid creating a new class of large rent seekers in the LLM space.
jeroenhd 3y ago
Computers aren't humans and LLMs aren't human brains. We have no way to reconstruct memories from a preserved brain (yet). The exact ways in which humans form memories and store information isn't even known yet; we're still drilling into the specifics from higher-level concepts. Modeling the human brain like nodes with weights ignores a lot of biological processes. Blood/oxygen flow, hormones, neurotransmitter decay, physical locality, chemical delays and interference from things like myelin sheaths, and other physical processes affect the synapses that are partially mirrored by computer simulations of neural networks. Unlike neural networks, human brains also don't work based on a single clock signal triggering input and output from all notes in instant steps. Human memories are also not just "data in, weights out". They are heavily modified by things like mood, concentration, language(s) spoken, context, and emotional triggers. There's no way to feed a dictionary into a brain. Memory preservation consists of multiple stages, with differing memory types, involving various brain segments with dedicated functionality that can actually grow back due to neuroplasticity in some cases. Efforts are being made to emulate living cells on computers, but LLMs aren't that. Inversely, efforts are also made to feed brain cells artificial signals and train them to play video games, which results in different behaviour compared to the systems we use for LLMs or other AI systems.
- virgildotcodes 3y ago
  So if we add a bunch of complex processes to an LLM, in order to produce a better analog of a human in terms of degree of complexity if not actual function, does that have some bearing on this copyright question? It doesn’t seem clear to me that it does. Is the argument that sufficient complexity in how an “intelligence” processes this copyrighted data leads to the output being transformative vs not transformative in a more simple mind/model? What if the output is exactly the same, or comparable enough, regardless of the degree of complexity of the mind/model?
  - lewhoo 3y ago
    IMHO no because LLM's are not legal persons and most likely will never be, so they can't acquire or operate under laws, privileges and agreements simply because they exists.
genocidicbunny 3y ago
> Don't humans operate similarly? I'm going to bypass this question a bit and say, who cares? Why do we need to treat these things the same way we treat humans? Why can we not say that it's okay if a human does it, and not okay if it's a computer? There's nothing that requires us to establish 'fair' as treating them the same as people.
- idle_zealot 3y ago
  > Why do we need to treat these things the same way we treat humans? Because it would be absurd for it to be legal to do X, but illegal to do so with an efficient tool. Especially when the activity X in question is "learning".
  - genocidicbunny 3y ago
    Why would it be absurd? Did we somehow stealthily develop a neural interface that lets us feed the 'learning' that 'AI' is doing into a human brain? Have we actually figured out how to do that? No, we haven't. So humans are still learning the same way, but with a new tool to condense and summarize some information. Kinda like a textbook in school. But we don't treat those as human beings with human rights do we?
  - johnnyworker 3y ago
    > Especially when the activity X in question is "learning". No such thing, "learning" in a vacuum describes nothing here. Might as well ask why I am allowed to make noise, e.g. speak, but when I install 5000 watt speakers on every square meter of the planet suddenly it's a problem, and roll my eyes at the inconsistency of not being allowed to "do X more efficiently with a tool".
  - allturtles 3y ago
    I don't find it at all absurd. Order-of-magnitude differences in quantity often become differences in quality. It's legal in most places in the U.S. to own a semi-automatic weapon that fires x bullets/minute, but not a fully-automatic one that fires 10*x bullets/minute. It's legal for me to send water into my sewer by flushing the toilet, but not for me to pump the larger volume of water in my sump pump into that same sewer. It's legal for a child to throw a handful of sand from the beach into the ocean. Do you think it should be legal for anyone to build and operate a sand throwing machine that methodically throws all of the sand on the beach into the ocean?
dspillett 3y ago
Yes, and if we reproduce something significant that way we have (depending on licences etc.) committed copyright infringement. The definition of significant is a very grey hairy one though – there are many obvious examples where you would say it is definitely copying from memory (so plagiarism rather than research or accidental similarity) or copying directly, and many examples where we'd all agree that the result is so likely to be independently produced that it isn't an issue, but even more examples where there is room for interpretation and disagreement. This isn't as well-defined for humans as you might think, so can't be well-defined for LLM techniques by comparing them to human agents. My argument against the current hoovering up of data under various licences for AI training, which they claim can#'t reproduce anything verbatim, is CoPilot. If there is no risk, then why did they only use public repositories and none of their own private ones? Surely they think their code contains good training material, unless they think their own code is gobbledygook. Or back in terms of licences: if it can't breach the GPL family, then it can't breach their own commercial licensing arrangements. Would OpenAI have a problem with humans for using some of their code/documents/other in this way?
akolbe 3y ago
Well, there is also the fact that if you train LLMs on LLM output the quality degrades very quickly. It is not a good thing to do.
batmenace 3y ago
Yes, humans operate similarly. We gain knowledge through learning and assimilation of experiences. But there is a cost associated with each new input, in one form or another. You can read and reproduce the NYT articles or their content and build based on what you have learned. But aside from piracy, you have to somehow pay a cost associated with access (ads, subscriptions etc). The LLM does not do so in an equal way.
layer8 3y ago
Humans can be sued for plagiarism, so AI should be as well.
CatWChainsaw 3y ago
You don't think perfect recall and sheer scale make this apples to pumpkins? What a disingenuous argument.

villgax 3y ago

True

LatteLazy 3y ago

Not just humans in general, the NYT specifically relies on the fact news cannot be copyrighted. Then claims it's articles are sacred...

dspillett 3y ago
IIRC the distinction is that facts can't be copyrighted, a particular arrangement of facts can, particularly if something more subjective (analysis or opinion) is included. So I can write my own article about the sky being shown to appear blue much of the time, but I can't copy someone else's article about the same subject.
- paulmd 3y ago
  How would one even go about a phonebook-style mechanical listing of facts and occurrences? You’re listing an impossibility and then saying that garden-variety connective sentences somehow make it not a factual listing. Like yes if you copy a NYT article verbatim it’s like copying a phone book ads and all, and that’s infringement. But that’s not what a LLM does, NYT doesn’t like their content being used and summarized at all, even in a rearranged form that merely relies on the factual information included in the article. That’s what they want to get paid for, and unfortunately that’s not copyrightable and OpenAI is correct they don’t have to pay for that. NYT disagrees but again, they are kinda attempting to claim copyright on the factual information because they wrote some connective sentences between.
  - dspillett 3y ago
    > How would one even go about a phonebook-style mechanical listing of facts and occurrences? The traditional trick there is to include some small amount of fake data in the directory. You know someone has copied your collection of facts instead of compiling their own because it includes your fake facts. Mapmakers have used the method for at least as long as cartography has been part of our recorded history, see https://en.wikipedia.org/wiki/Trap_street https://en.wikipedia.org/wiki/Trap_street for details. As noted in that page, the legal status of this, like many IP related issues, depends upon jurisdiction. > But that’s not what a LLM does, What does it do that means it is only summarizing factual information? While NYT effectively trying to claim copyright on facts is wrong, OpenAI claiming it can't reproduce copyrightable information while it can reproduce/summarize facts found within the same training set seems at best disingenuous.
    - paulmd 3y ago
      How are trap streets “the trick” if they are not copyrightable or enforceable under US law? Like the Supreme Court has specifically considered and dismissed this “one clever trick to keep people from copying lists of facts” because to allow it would completely undermine the idea of facts and figures not being copyrightable. > Trap streets are not copyrightable under the federal law of the United States. In Nester's Map & Guide Corp. v. Hagstrom Map Co. (1992),[3][4] a United States federal court found that copyright traps are not themselves protectable by copyright. There, the court stated: "[t]o treat 'false' facts interspersed among actual facts and represented as actual facts as fiction would mean that no one could ever reproduce or copy actual facts without risk of reproducing a false fact and thereby violating a copyright ... If such were the law, information could never be reproduced or widely disseminated." (Id. at 733) And yes the EU has the concept of “database rights” but notionally there is still supposed to be a creative step required in the selection or arrangement of records. So just a raw copy of the numbers in a telephone directory is theoretically not copyrightable, but a telephone book might be because of the creative/transformational step. It’s possible this might be such a low bar that it’s impossible to fail to clear, but, at least on paper you can’t copyright mere facts and figures either. But either way it’s generally true that simple facts and figures are not protected and trap streets are a discredited and clumsy attempt to work around this.

bmitc 3y ago

It's quite different, is it not? I don't get the analogy. These models are scanning, storing, and ingesting more material than any one human could. Not only is the method completely different, the end goal and applications are as well. The analogy basically isn't one, at all. I'm pretty upset at companies using our personal data to make gobs of money off of. I'm also upset that they're now using our knowledge work to make even more money off of. We don't exist as computational nodes for them, a free resource to exhaust. It is a completely one way street with no consent. So I am in favor of all of these companies getting a reality check.

machdiamonds 3y ago
Consider a mega consulting firm with millions of von Neumann-like analysts. Together, they've processed the same vast data that LLMs have, but individually, none could. It's not an LLM, but its purpose is like ChatGPT: assisting clients with their tasks. If LLMs concern you due to their data processing, would a firm like this do the same?
- bmitc 3y ago
  Bringing up poor fitting analogies won't change my opinion.
  - sebzim4500 3y ago
    I normally consider these discussion to be more about the people reading the comments than the people writing them. You've clearly made up your mind, but others presumably haven't so I think it's good he makes these arguments, even if it looks like tilting at windmills to you.
    - bmitc 3y ago
      That's a good point, but I guess I was implicitly getting at that continually permutating poor analogies and hypotheticals isn't an interesting discussion. The fact of the matter here is that parties, such as OpenAI, are benefiting from others' knowledge work, protected or not, in a completely one-sided way and all for free. And I don't feel sorry for companies that need to build Trojan horse products, such as OpenAI and Google, in order to survive off of other people's data that they never compensate for.

mattmanser 3y ago

Humans don't have perfect recall, don't have virtually infinite storage and can't process requests in milliseconds. I wouldn't be surprised if the avenue of attack is that fair use laws are for humans, not robots, and if an AI has been trained on copyrighted data, that's not fair use. Also, don't forget that in reality what's happened is that a bunch of copyrighted text is encoded in the LLM in a way a human can't understand, but that the LLM essentially CAN understand.

fjfuvucucuc 3y ago
Yup, and that reproducing it is a copyright violation. Still a copyright abolitionist though. Maybe now more people will join the fight?
- gmerc 3y ago
  Well if use the tool to reproduce copyrighted content you are violating copyright. But that’s not the primary usecase and nobody in their right mind is arguing that. The weights are not a reproduction of the content. They are capable of it but so is a photocopier a lot more and we didn’t ban those either despite them technically being a lot more useful for violation. Nah, this is expansionist doctrine and agenda for copyright - these companies are trying to copyright style and locations in latent space now.
  - mattmanser 3y ago
    Photocopiers are for personal use, training an AI is not. If you photocopy 10,000 copies of copyrighted text and starting distributing it you will get sued. It would be different if I trained my own AI, for my personal use.
    - gmerc 3y ago
      businesses use photocopiers, there’s Xerox shops, etc. Again, LLMs don’t copy so it’s not a good metaphor.

RandomLensman 3y ago

I don't think any unaided human or collective of humans could rent seek on something close to the sum total of human knowledge and expression. Irrespective of copyright issues, the question is how to avoid creating a new class of large rent seekers in the LLM space.

jeroenhd 3y ago

Computers aren't humans and LLMs aren't human brains. We have no way to reconstruct memories from a preserved brain (yet). The exact ways in which humans form memories and store information isn't even known yet; we're still drilling into the specifics from higher-level concepts. Modeling the human brain like nodes with weights ignores a lot of biological processes. Blood/oxygen flow, hormones, neurotransmitter decay, physical locality, chemical delays and interference from things like myelin sheaths, and other physical processes affect the synapses that are partially mirrored by computer simulations of neural networks. Unlike neural networks, human brains also don't work based on a single clock signal triggering input and output from all notes in instant steps. Human memories are also not just "data in, weights out". They are heavily modified by things like mood, concentration, language(s) spoken, context, and emotional triggers. There's no way to feed a dictionary into a brain. Memory preservation consists of multiple stages, with differing memory types, involving various brain segments with dedicated functionality that can actually grow back due to neuroplasticity in some cases. Efforts are being made to emulate living cells on computers, but LLMs aren't that. Inversely, efforts are also made to feed brain cells artificial signals and train them to play video games, which results in different behaviour compared to the systems we use for LLMs or other AI systems.

virgildotcodes 3y ago
So if we add a bunch of complex processes to an LLM, in order to produce a better analog of a human in terms of degree of complexity if not actual function, does that have some bearing on this copyright question? It doesn’t seem clear to me that it does. Is the argument that sufficient complexity in how an “intelligence” processes this copyrighted data leads to the output being transformative vs not transformative in a more simple mind/model? What if the output is exactly the same, or comparable enough, regardless of the degree of complexity of the mind/model?
- lewhoo 3y ago
  IMHO no because LLM's are not legal persons and most likely will never be, so they can't acquire or operate under laws, privileges and agreements simply because they exists.

genocidicbunny 3y ago

> Don't humans operate similarly? I'm going to bypass this question a bit and say, who cares? Why do we need to treat these things the same way we treat humans? Why can we not say that it's okay if a human does it, and not okay if it's a computer? There's nothing that requires us to establish 'fair' as treating them the same as people.

idle_zealot 3y ago
> Why do we need to treat these things the same way we treat humans? Because it would be absurd for it to be legal to do X, but illegal to do so with an efficient tool. Especially when the activity X in question is "learning".
- genocidicbunny 3y ago
  Why would it be absurd? Did we somehow stealthily develop a neural interface that lets us feed the 'learning' that 'AI' is doing into a human brain? Have we actually figured out how to do that? No, we haven't. So humans are still learning the same way, but with a new tool to condense and summarize some information. Kinda like a textbook in school. But we don't treat those as human beings with human rights do we?
- johnnyworker 3y ago
  > Especially when the activity X in question is "learning". No such thing, "learning" in a vacuum describes nothing here. Might as well ask why I am allowed to make noise, e.g. speak, but when I install 5000 watt speakers on every square meter of the planet suddenly it's a problem, and roll my eyes at the inconsistency of not being allowed to "do X more efficiently with a tool".
- allturtles 3y ago
  I don't find it at all absurd. Order-of-magnitude differences in quantity often become differences in quality. It's legal in most places in the U.S. to own a semi-automatic weapon that fires x bullets/minute, but not a fully-automatic one that fires 10*x bullets/minute. It's legal for me to send water into my sewer by flushing the toilet, but not for me to pump the larger volume of water in my sump pump into that same sewer. It's legal for a child to throw a handful of sand from the beach into the ocean. Do you think it should be legal for anyone to build and operate a sand throwing machine that methodically throws all of the sand on the beach into the ocean?

dspillett 3y ago

Yes, and if we reproduce something significant that way we have (depending on licences etc.) committed copyright infringement. The definition of significant is a very grey hairy one though – there are many obvious examples where you would say it is definitely copying from memory (so plagiarism rather than research or accidental similarity) or copying directly, and many examples where we'd all agree that the result is so likely to be independently produced that it isn't an issue, but even more examples where there is room for interpretation and disagreement. This isn't as well-defined for humans as you might think, so can't be well-defined for LLM techniques by comparing them to human agents. My argument against the current hoovering up of data under various licences for AI training, which they claim can#'t reproduce anything verbatim, is CoPilot. If there is no risk, then why did they only use public repositories and none of their own private ones? Surely they think their code contains good training material, unless they think their own code is gobbledygook. Or back in terms of licences: if it can't breach the GPL family, then it can't breach their own commercial licensing arrangements. Would OpenAI have a problem with humans for using some of their code/documents/other in this way?

akolbe 3y ago

Well, there is also the fact that if you train LLMs on LLM output the quality degrades very quickly. It is not a good thing to do.

batmenace 3y ago

Yes, humans operate similarly. We gain knowledge through learning and assimilation of experiences. But there is a cost associated with each new input, in one form or another. You can read and reproduce the NYT articles or their content and build based on what you have learned. But aside from piracy, you have to somehow pay a cost associated with access (ads, subscriptions etc). The LLM does not do so in an equal way.

layer8 3y ago

Humans can be sued for plagiarism, so AI should be as well.

CatWChainsaw 3y ago

You don't think perfect recall and sheer scale make this apples to pumpkins? What a disingenuous argument.