Although I’m a firm believer that most AI models should be public domain or open source by default, the premise of “illegally trained LLMs” is flawed. Because there really is no assurance that LLMs currently in use are illegally trained to begin with. These things are still being argued in court, but the AI companies have a pretty good defense in the fact analyzing publicly viewable information is a pretty deep rooted freedom that provides a lot of positives to the world.
The idea of… well, ideas, being copyrightable, should shake the boots of anyone in this discussion. Especially since when the laws on the book around these kinds of things become active topic of change, they rarely shift in the direction of more freedom for the exact people we want to give it to. See: Copyright and Disney.
The underlying technology simply has more than enough good uses that banning it would simply cause it to flourish elsewhere that does not ban it, which means as usual that everyone but the multinational companies lose out. The same would happen with more strict copyright, as only the big companies have the means to build their own models with their own data. The general public is set up for a lose-lose to these companies as it currently stands. By requiring the models to be made available to the public do we ensure that the playing field doesn’t tip further into their favor to the point AI technology only exists to benefit them.
If the model is built on the corpus of humanity, then humanity should benefit.
OpenAI hasn’t disclosed the datasets that ChatGPT is trained on, but in an older paper two databases are referenced; “Books1” and “Books2”. The first one contains roughly 63,000 titles and the latter around 294,000 titles.
These numbers are meaningless in isolation. However, the authors note that OpenAI must have used pirated resources, as legitimate databases with that many books don’t exist.
Should be easy to defend against, right-out trivial: OpenAI, just tell us what those Books1 and Books2 databases are. Where you got them from, the licensing contracts with publishers that you signed to give you access to such a gigantic library. No need to divulge details, just give us information that makes it believable that you licensed them.
…crickets. They pirated the lot of it otherwise they would already have gotten that case thrown out. It’s US startup culture, plain and simple, “move fast and break laws”, get lots of money, have lots of money enabling you to pay the best lawyers to abuse the shit out of the US court system.
For OpenAI, I really wouldn’t be surprised if that happened to be the case, considering they still call themselves “OpenAI” despite being the most censored and closed source AI models on the market.
But my comment was more aimed at AI models in general. If you are assuming they indeed used non-publicly posted or gathered material, and did so directly themselves, they would indeed not have a defense to that. Unfortunately, if a second hand provided them the data, and did so under false pretenses, it would likely let them legally off the hook even if they had every ethical obligation to make sure it was publicly available. The second hand that provided it to them would be the one infringing.
If that assumption turns out to be a truth (Maybe through some kind of discovery in the trial), they should burn for that. Until then, even if it’s a justified assumption, it’s still an assumption, and most likely not true for most models, certainly not those trained recently.
the AI companies have a pretty good defense in the fact analyzing publicly viewable information is a pretty deep rooted freedom that provides a lot of positives to the world
They are not “analyzing” the data. They are feeding it into a regurgitating mechanism. There’s a big difference. Their defense is only “good” because AI is being misrepresented and misunderstood.
I agree that we shouldn’t strive for more strict copyright. We should fight for a much more liberal system. But as long as everyone else has to live by the current copyright laws, we should not let AI companies get away with what they’re doing.
They are not “analyzing” the data. They are feeding it into a regurgitating mechanism. There’s a big difference. Their defense is only “good” because AI is being misrepresented and misunderstood.
I really kind of hope you’re kidding here. Because this has got to be the most roundabout way of saying they’re analyzing the information. Just because you think it does so to regurgitate (which I have yet to see any good evidence for, at least for the larger models), does not change the definition of analyzing. And by doing so you are misrepresenting it and showing you might just have misunderstood it, which is ironic. And doing so does not help the cause of anyone who wishes to reduce the harm from AI, as you are literally giving ammo to people to point to and say you are being irrational about it.
Yes if you completely ignore how data is processed and how the product is derived from the data, then everything can be labeled “data analysis”. Great point. So copyright infringement can never exist because the original work can always be considered data that you analyze. Incredible.
No, not what I said at all. If you’re trying to say I’m making this argument I’d urge you (ironically) to actually analyze what I said rather than putting words in my mouth ;)
Copyright infringement (or plagiarism) in it’s simplest form, as in just taking the material as is, is devoid of any analysis. The point is to avoid having to do that analysis and just get right to the end result that has value.
But that’s not what AI technology does. None of the material used to train it ends up in the model. It looks at the training data and extracts patterns. For text, that is the sentence structure, the likelihood of words being followed by another, the paragraph/line length, the relationship between words when used together, and more. It can do all of this without even ‘knowing’ what these things are, because they are simply patterns that show up in large amounts of data, and machine learning as a technology is made to be able to detect and extract those patterns. That detection is synonymous with how humans do analysis. What it detects are empirical, factual observations about the material it is shown, which cannot be copyrighted.
The resulting data when fed back to the AI can be used to have it extrapolate on incomplete data, which it could not do without such analysis. You can see this quite easily by asking an AI to refer to you by a specific name, or talk in a specific manner, such as a pirate. It ‘understands’ that certain words are placeholders for names, and that some words can be ‘pirateitfied’ by adding filler words or pre/suffixing other words. It could not do so without analysis, unless that text was already in the data to begin with, which is doubtful.
There are law offices that exist specifically to fuck with people over patent and copywrite law.
There’s also cases where people use copywrite and patent law to hold us back. I can’t find the article but some religious jerk patented connecting a sex toy to a computer via USB. Thankfully someone got around this law with bluetooth and cell phones. Otherwise I imagine the camgirl and LDR market for toys would’ve been hit with products 10 years sooner.
I’ve never really delved into the AI copyright debate before, so forgive my ignorance on the matter.
I don’t understand how an AI reading a bunch of books and rearranging some of those words into a new story, is different to a human author reading a bunch of books and rearranging those words into a new story.
Most AI art I’ve seen has been… Unique, to say the least. To me, they tend to be different enough to the art they were trained in to not be a direct ripoff, so personally I don’t see the issue.
I think the the main difference is one being a human author and this is how humans function. We can not unsee or unhear things but we can be compelled to not use that information if the law requires so company secrets/inadmissible evidence in jury duty/plagiarism laws that already exist. And the other being a machine that do not have agency or personhood that has this information being fed to it ( created by other people ) for the sole purpose of creating a closed system for a company so it’s shareholders can make money. It’s this open for me but not for thee approach is the main problem people have. You have this proprietary “open ai” that microsoft invested 25 or so billion in so they can scrape other peoples work and charge you money for variations of it. I don’t mind abolishing ip or patent laws all together so everyone can use and improve chatgpt with whatever they have. If you yourself are hiding behind ip laws to protect your software and disrespecting other peoples copyright laws that’s what people see problematic.
Yes, this is my exact issue with some framing of AI. Creative people love their influences to the point you can ask them and they will point to parts that they reference or nudged to an influence they partially credit to getting to that result. It’s also extremely normal that when you make something new, you brainstorm and analyze any kind of material (copyrighted or not) you can find that gives the same feelings you desire to create. As is ironically said to give comfort to starting creatives that it’s okay to be inspired by others: “Good artists copy, great artists steal.”
And often people very anti AI don’t see an issue with this, yet it is in essence the same as the AI does, which is to detach the work from the ideas it was built on, and then re-using those ideas. And just like anyone who has the ability to create has the ability to plagiarize or infringe, so does the AI. As human users of AI we must be the ones to ethically guide it away from that (Since it can’t do that itself), just like you would not copy-paste your influences into a new human made work.
The for-profit large-scale media blender is the problem. When it’s a human writing Harry Potter fan fiction, it’s fine. When a company sells a tool for you to write thousands of trash “books” for profit, it’s a problem.
Which is why the technology itself isn’t the issue, but those willing to use it in unethical ways. AI is an invaluable tool to those with limited means, unlike big corporations.
ML algorithms aren’t capable of producing anything new, they can only ever produce a mishmash of copies of existing works.
If you feed a generative model a bunch of physics research papers, it won’t create a new valid physics research paper, just a mishmash of jargon from existing papers.
You say it’s not capable of producing anything new, but then give an example of it creating something new. You just changed the goal from “new” to “valid” in the next sentence. Looking at AI for “valid” information is silly, but looking at it for “new” information is not. Humans do this kind of information mixing all the time. It’s why fan works are a thing, and why most creative people have influences they credit with being where they are today.
Nobody alive today isn’t tainted by the ideas they’ve consumed in copyrighted works, but we do not bat an eye if you use that in a transformative manner. And AI already does this transformation much better than humans do since it’s trained on that much more information, diluting the pool of sources, which effectively means less information from a single source is used.
Banning AI is out of the question. Even the EU accepts that and they tend to be pretty ban heavy, unlike the US.
But it’s important that we have these discussions about how copyright applies to AI so that we can actually get an answer and move on, right now it’s this legal quagmire that no one really wants to get involved in except the big companies. If a small group of university students want to build an AI right now they can’t because of the legal nightmare that would be the Twilight zone of law that is acquiring training data.
AI is right-out unregulated in the EU unless and until you actually use it for something where it becomes relevant, then you’ve got at the lower end labelling requirements (If your customer service is an AI chat, say that it’s an AI chat), up to heavy, heavy requirements when you use it for stuff like sifting through job applications. The burden of proof that the AI isn’t e.g. racist is on you. Or, for that matter, using to reject health insurance claims, I think we saw some news lately out of the US what can happen when you do that.
OpenAI’s copyright case isn’t really good to make the legal situation any clearer: We already know that using pirated content to train stuff isn’t legal because you’re not looking at it legitimately. The case isn’t about the “are computers allowed to learn from public sources just as humans are” question.
Although I’m a firm believer that most AI models should be public domain or open source by default, the premise of “illegally trained LLMs” is flawed. Because there really is no assurance that LLMs currently in use are illegally trained to begin with. These things are still being argued in court, but the AI companies have a pretty good defense in the fact analyzing publicly viewable information is a pretty deep rooted freedom that provides a lot of positives to the world.
The idea of… well, ideas, being copyrightable, should shake the boots of anyone in this discussion. Especially since when the laws on the book around these kinds of things become active topic of change, they rarely shift in the direction of more freedom for the exact people we want to give it to. See: Copyright and Disney.
The underlying technology simply has more than enough good uses that banning it would simply cause it to flourish elsewhere that does not ban it, which means as usual that everyone but the multinational companies lose out. The same would happen with more strict copyright, as only the big companies have the means to build their own models with their own data. The general public is set up for a lose-lose to these companies as it currently stands. By requiring the models to be made available to the public do we ensure that the playing field doesn’t tip further into their favor to the point AI technology only exists to benefit them.
If the model is built on the corpus of humanity, then humanity should benefit.
As per torrentfreak
Should be easy to defend against, right-out trivial: OpenAI, just tell us what those Books1 and Books2 databases are. Where you got them from, the licensing contracts with publishers that you signed to give you access to such a gigantic library. No need to divulge details, just give us information that makes it believable that you licensed them.
…crickets. They pirated the lot of it otherwise they would already have gotten that case thrown out. It’s US startup culture, plain and simple, “move fast and break laws”, get lots of money, have lots of money enabling you to pay the best lawyers to abuse the shit out of the US court system.
For OpenAI, I really wouldn’t be surprised if that happened to be the case, considering they still call themselves “OpenAI” despite being the most censored and closed source AI models on the market.
But my comment was more aimed at AI models in general. If you are assuming they indeed used non-publicly posted or gathered material, and did so directly themselves, they would indeed not have a defense to that. Unfortunately, if a second hand provided them the data, and did so under false pretenses, it would likely let them legally off the hook even if they had every ethical obligation to make sure it was publicly available. The second hand that provided it to them would be the one infringing.
If that assumption turns out to be a truth (Maybe through some kind of discovery in the trial), they should burn for that. Until then, even if it’s a justified assumption, it’s still an assumption, and most likely not true for most models, certainly not those trained recently.
They are not “analyzing” the data. They are feeding it into a regurgitating mechanism. There’s a big difference. Their defense is only “good” because AI is being misrepresented and misunderstood.
I agree that we shouldn’t strive for more strict copyright. We should fight for a much more liberal system. But as long as everyone else has to live by the current copyright laws, we should not let AI companies get away with what they’re doing.
I really kind of hope you’re kidding here. Because this has got to be the most roundabout way of saying they’re analyzing the information. Just because you think it does so to regurgitate (which I have yet to see any good evidence for, at least for the larger models), does not change the definition of analyzing. And by doing so you are misrepresenting it and showing you might just have misunderstood it, which is ironic. And doing so does not help the cause of anyone who wishes to reduce the harm from AI, as you are literally giving ammo to people to point to and say you are being irrational about it.
Yes if you completely ignore how data is processed and how the product is derived from the data, then everything can be labeled “data analysis”. Great point. So copyright infringement can never exist because the original work can always be considered data that you analyze. Incredible.
No, not what I said at all. If you’re trying to say I’m making this argument I’d urge you (ironically) to actually analyze what I said rather than putting words in my mouth ;)
Copyright infringement (or plagiarism) in it’s simplest form, as in just taking the material as is, is devoid of any analysis. The point is to avoid having to do that analysis and just get right to the end result that has value.
But that’s not what AI technology does. None of the material used to train it ends up in the model. It looks at the training data and extracts patterns. For text, that is the sentence structure, the likelihood of words being followed by another, the paragraph/line length, the relationship between words when used together, and more. It can do all of this without even ‘knowing’ what these things are, because they are simply patterns that show up in large amounts of data, and machine learning as a technology is made to be able to detect and extract those patterns. That detection is synonymous with how humans do analysis. What it detects are empirical, factual observations about the material it is shown, which cannot be copyrighted.
The resulting data when fed back to the AI can be used to have it extrapolate on incomplete data, which it could not do without such analysis. You can see this quite easily by asking an AI to refer to you by a specific name, or talk in a specific manner, such as a pirate. It ‘understands’ that certain words are placeholders for names, and that some words can be ‘pirateitfied’ by adding filler words or pre/suffixing other words. It could not do so without analysis, unless that text was already in the data to begin with, which is doubtful.
Not to mention patent laws are bullshit.
There are law offices that exist specifically to fuck with people over patent and copywrite law.
There’s also cases where people use copywrite and patent law to hold us back. I can’t find the article but some religious jerk patented connecting a sex toy to a computer via USB. Thankfully someone got around this law with bluetooth and cell phones. Otherwise I imagine the camgirl and LDR market for toys would’ve been hit with products 10 years sooner.
I’ve never really delved into the AI copyright debate before, so forgive my ignorance on the matter.
I don’t understand how an AI reading a bunch of books and rearranging some of those words into a new story, is different to a human author reading a bunch of books and rearranging those words into a new story.
Most AI art I’ve seen has been… Unique, to say the least. To me, they tend to be different enough to the art they were trained in to not be a direct ripoff, so personally I don’t see the issue.
I think the the main difference is one being a human author and this is how humans function. We can not unsee or unhear things but we can be compelled to not use that information if the law requires so company secrets/inadmissible evidence in jury duty/plagiarism laws that already exist. And the other being a machine that do not have agency or personhood that has this information being fed to it ( created by other people ) for the sole purpose of creating a closed system for a company so it’s shareholders can make money. It’s this open for me but not for thee approach is the main problem people have. You have this proprietary “open ai” that microsoft invested 25 or so billion in so they can scrape other peoples work and charge you money for variations of it. I don’t mind abolishing ip or patent laws all together so everyone can use and improve chatgpt with whatever they have. If you yourself are hiding behind ip laws to protect your software and disrespecting other peoples copyright laws that’s what people see problematic.
Yes, this is my exact issue with some framing of AI. Creative people love their influences to the point you can ask them and they will point to parts that they reference or nudged to an influence they partially credit to getting to that result. It’s also extremely normal that when you make something new, you brainstorm and analyze any kind of material (copyrighted or not) you can find that gives the same feelings you desire to create. As is ironically said to give comfort to starting creatives that it’s okay to be inspired by others: “Good artists copy, great artists steal.”
And often people very anti AI don’t see an issue with this, yet it is in essence the same as the AI does, which is to detach the work from the ideas it was built on, and then re-using those ideas. And just like anyone who has the ability to create has the ability to plagiarize or infringe, so does the AI. As human users of AI we must be the ones to ethically guide it away from that (Since it can’t do that itself), just like you would not copy-paste your influences into a new human made work.
The for-profit large-scale media blender is the problem. When it’s a human writing Harry Potter fan fiction, it’s fine. When a company sells a tool for you to write thousands of trash “books” for profit, it’s a problem.
Which is why the technology itself isn’t the issue, but those willing to use it in unethical ways. AI is an invaluable tool to those with limited means, unlike big corporations.
ML algorithms aren’t capable of producing anything new, they can only ever produce a mishmash of copies of existing works.
If you feed a generative model a bunch of physics research papers, it won’t create a new valid physics research paper, just a mishmash of jargon from existing papers.
You say it’s not capable of producing anything new, but then give an example of it creating something new. You just changed the goal from “new” to “valid” in the next sentence. Looking at AI for “valid” information is silly, but looking at it for “new” information is not. Humans do this kind of information mixing all the time. It’s why fan works are a thing, and why most creative people have influences they credit with being where they are today.
Nobody alive today isn’t tainted by the ideas they’ve consumed in copyrighted works, but we do not bat an eye if you use that in a transformative manner. And AI already does this transformation much better than humans do since it’s trained on that much more information, diluting the pool of sources, which effectively means less information from a single source is used.
Banning AI is out of the question. Even the EU accepts that and they tend to be pretty ban heavy, unlike the US.
But it’s important that we have these discussions about how copyright applies to AI so that we can actually get an answer and move on, right now it’s this legal quagmire that no one really wants to get involved in except the big companies. If a small group of university students want to build an AI right now they can’t because of the legal nightmare that would be the Twilight zone of law that is acquiring training data.
AI is right-out unregulated in the EU unless and until you actually use it for something where it becomes relevant, then you’ve got at the lower end labelling requirements (If your customer service is an AI chat, say that it’s an AI chat), up to heavy, heavy requirements when you use it for stuff like sifting through job applications. The burden of proof that the AI isn’t e.g. racist is on you. Or, for that matter, using to reject health insurance claims, I think we saw some news lately out of the US what can happen when you do that.
OpenAI’s copyright case isn’t really good to make the legal situation any clearer: We already know that using pirated content to train stuff isn’t legal because you’re not looking at it legitimately. The case isn’t about the “are computers allowed to learn from public sources just as humans are” question.