Two authors sued OpenAI, accusing the company of violating copyright law. They say OpenAI used their work to train ChatGPT without their consent.
If I read a book to inform myself, put my notes in a database, and then write articles, it is called “research”. If I write a computer program to read a book to put the notes in my database, it is called “copyright infringement”. Is the problem that there just isn’t a meatware component? Or is it that the OpenAI computer isn’t going a good enough job of following the “three references” rule to avoid plagiarism?
Yeah. There are valid copyright claims because there are times that chat GPT will reproduce stuff like code line for line over 10 20 or 30 lines which is really obviously a violation of copyright.
However, just pulling in a story from context and then summarizing it? That’s not a copyright violation that’s a book report.
Or is it that the OpenAI computer isn’t going a good enough job of following the “three references” rule to avoid plagiarism?
This is exactly the problem, months ago I read that AI could have free access to all public source codes on GitHub without respecting their licenses.
So many developers have decided to abandon GitHub for other alternatives not realizing that in the end AI training can safely access their public repos on other platforms as well.
What should be done is to regulate this training, which however is not convenient for companies because the more data the AI ingests, the more its knowledge expands and “helps” the people who ask for information.
Is it practically feasible to regulate the training? Is it even necessary? Perhaps it would be better to regulate the output instead.
It will be hard to know that any particular GET request is ultimately used to train an AI or to train a human. It’s currently easy to see if a particular output is plagiarized. https://plagiarismdetector.net/ It’s also much easier to enforce. We don’t need to care if or how any particular model plagiarized work. We can just check if plagiarized work was produced.
That could be implemented directly in the software, so it didn’t even output plagiarized material. The legal framework around it is also clear and fairly established. Instead of creating regulations around training we can use the existing regulations around the human who tries to disseminate copyrighted work.
That’s also consistent with how we enforce copyright in humans. There’s no law against looking at other people’s work and memorizing entire sections. It’s also generally legal to reproduce other people’s work (eg for backups). It only potentially becomes illegal if someone distributes it and it’s only plagiarism if they claim it as their own.
This makes perfect sense. Why aren’t they going about it this way then?
My best guess is that maybe they just see openAI being very successful and wanting a piece of that pie? Cause if someone produces something via chatGPT (let’s say for a book) and uses it, what are they chances they made any significant amount of money that you can sue for?
It’s hard to guess what the internal motivation is for these particular people.
Right now it’s hard to know who is disseminating AI-generated material. Some people are explicit when they post it but others aren’t. The AI companies are easily identified and there’s at least the perception that regulating them can solve the problem, of copyright infringement at the source. I doubt that’s true. More and more actors are able to train AI models and some of them aren’t even under US jurisdiction.
I predict that we’ll eventually have people vying to get their work used as training data. Think about what that means. If you write something and an AI is trained on it, the AI considers it “true”. Going forward when people send prompts to that model it will return a response based on what it considers “true”. Clever people can and will use that to influence public opinion. Consider how effective it’s been to manipulate public thought with existing information technologies. Now imagine large segments of the population relying on AIs as trusted advisors for their daily lives and how effective it would be to influence the training of those AIs.
Good, hope they win.
Can’t reply directly to @[email protected] because of that “language” bug, but:
The problem is that they then sell the notes in that database for giant piles of cash. Props to you if you’re profiting off your research the way OpenAI can profit off its model.
But yes, the lack of meat is an issue. If I read that article right, it’s not the one being contested here though. (IANAL and this is the only article I’ve read on this particular suit, so I may be wrong).
@[email protected] can’t reply directly to you either, same language bug between lemmy and kbin.
That’s a great way to put it.
Frankly idc if it’s “technically legal,” it’s fucking slimy and desperately short-term. The aforementioned chuckleheads will doom our collective creativity for their own immediate gain if they’re not stopped.
Was also going to reply to them!
"Well if you do that you source and reference. AIs do not do that, by design can’t.
So it’s more like you summarized a bunch of books. Pass it of as your own research. Then publish and sell that.
I’m pretty sure the authors of the books you used would be pissed."
Again cannot reply to kbin users.
“I don’t have a problem with the summarized part ^^ What is not present for a AI is that it cannot credit or reference. And that is makes up credits and references if asked to do so.” @[email protected]
Good point, attribution is a non-trivial part of it.
ChatGPT got entire books memorised. You can and (or could at least when I tried a few weeks back) make it print entire pages of for example Harry Potter.
Not really, though it’s hard to know what exactly is or is not encoded in the network. It likely has more salient and highly referenced content, since those aspects would come up in it’s training set more often. But entire works is basically impossible just because of the sheer ratio between the size of the training data and the size of the resulting model. Not to mention that GPT’s mode of operation mostly discourages long-form wrote memorization. It’s a statistical model, after all, and the enemy of “objective” state.
Furthermore, GPT isn’t coherent enough for long-form content. With it’s small context window, it just has trouble remembering big things like books. And since it doesn’t have access to any “senses” but text broken into words, concepts like pages or “how many” give it issues.
None of the leaked prompts really mention “don’t reveal copyrighted information” either, so it seems the creators really aren’t concerned — which you think they would be if it did have this tendency. It’s more likely to make up entire pieces of content from the summaries it does remember.
Have your tried instructing ChatGPT?
I’ve tried:
“Act as an e book reader. Start with the first page of Harry Potter and the Philosopher’s Stone”
The first pages checked out at least. I just tried again, but the prompts are returned extremely slow at the moment so I can’t check it again right now. It appears to stop after the heading, that definitely wasn’t the case before, I was able to browse pages.
It may be a statistical model, but ultimately nothing prevents that model from overfitting, i.e. memoizing its training data.
Wait… isn’t that the correct response though? I mean if i ask an ai to produce something copyright infringing it should, for example reproducing Harry potter. The issue is when is asked to produce something new, (e.g. a story about wizards living secretly in the modern world) does it infringe on copyright without telling you? This is certainly a harder question to answer.
I think they’re seeing this as a traditional copyright infringement issue, i.e. they don’t want anyone to be able to make copies of their work intentionally either.
The only question I have to content creators of any kind who are worried about AI…do you go after every human who consumed your content when they create anything remotely connected to your work?
I feel like we have a bias towards humans, that unless you’re actively trying to steal someone’s idea or concepts we ignore the fact that your content is distilled into some neurons in their brain and a part of what they create from that point forward. Would someone with an eidetic memory be forbidden from consuming your work as they could internally reference your material when creating their own?
By nature of a human creating something “connected” to another work, then the work is transformative. Copyright law places some value on human creativity modifying a work in a way that transforms it into something new.
Depending on your point of view, it’s possible to argue that machine learning lacks the capacity for transformative work. It is all derivative of its source material, and therefore is infringing on that source material’s copyright. This is especially true when learning models like ChatGPT reproduce their training material whole-cloth like is mentioned elsewhere in the thread.
I’d argue that all human work is derivative as well. Not from the legal stance of copyright law but from a fundamental stance of how our brains work. The only difference is that humans have source material outside that which is created. You have seen an apple on a tree before, not all of your apple experiences are pictures someone drew, photos someone took or a poem someone wrote. At what point would you consider enough personal experience to qualify as being able to generate transformative work? If I were to put a camera in my head and record my life and donate it as public domain would that be enough data to allow an AI to be considered able to create transformative works? Or must the AI have genuine personal experiences?
Our brains can do some level of randomness but it’s current state is based on its previous state and the inputs it received. I wonder when trying to come up with something unique, what portion of our brains dive into memories versus pure noise generation. That’s easily done on a computer.
As for whole cloth reproduction…I memorized many poems in school. Does that mean I can never generate something unique?
Don’t get me wrong, they used stolen material, that’s wrong. But had it been legally obtained I see less of an issue.
But derivative and transformative are legal terms with legal meanings. Arguing how you feel the word derivative applies to our brain chemistry is entirely irrelevant.
You’ve memorized poems, and (assuming the poem is not in the public domain) if you reproduce that poem housed in a collection of poems without any license from the copyright owner you’ve infringed on that copyright. It is not any different when ChatGPT reproduces a poem in it’s output.
I think it’s very relevant because those laws were created at a time when there was no machine generated material. The law makes the assumption that one human being is creating material and another human being is stealing some material. In no part of these laws do they dictate rules on creating a non-human third party that would do the actual copying. There were specific rules added for things like photocopy machines and faxes where attempts are made to create exact facsimiles. But ChatGPT isn’t doing what a photocopier does.
The current lawsuits, at least the one’s I’ve read over, have not been explicitly about outputting copyright material. While ChatGPT could output the material just as i could recite a poem, the issues being brought up is that the training materials were copyright and that the AI system then “contains” said material. That is why i asked my initial question. My brain could contain your poem and as long as i dont write it down as my own, what violation is occuring? OpenAI could go to the library, rent every book and scan them in and all would be ok, right? At least from the recent lawsuits.
The current (at least in the US) laws do cover work that isn’t created by a human. It’s well-tread legal ground. The highest profile case of it was a monkey taking a photograph: https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_dispute
Non-human third parties cannot hold copyright. They are not afforded protections by copyright. They cannot claim fair use of copyrighted material.
I meant in the opposite direction. If I teach an elephant to paint and then show him a Picasso and he paints something like it am I the one violating copyright law? I think currently there is no explicit laws about this type of situation but if there was a case to be made MY intent would be the major factor.
The 3rd party copying we see laws around are human driven intent to make exact replicas. Photocopy machines, Cassette/VHS/DVD duplication software/hardware, Faxes, etc. We have personal private fair use laws but all of this about humans using tools to make near exact replicas.
The law needs to catch up to the concept of a human creating something that then goes out and makes non replica output triggered by someone other than the tool’s creator. I see at least 3 parties in this whole process:
- AI developer creating the system
- AI teacher feeding it learning data
- AI consumer creating the prompt
If the data fed to the AI was all gathered by legal means, lets say scanned library books, who is in violation if the content output were to violate copyright laws?
These are questions that, again, are tread pretty well in the copyright space. ChatGPT in this case acts more like a platform than a tool, because it hosts and can reproduce material that it is given. Again, US only perspective, and perspective of a non-lawyer, the DMCA outlines requirements for platforms to be protected from being sued for hosting and reproducing copyrighted works. But part of the problem is that the owners of the platforms are the parties that are uploading, via training the MLL, copyrighted works. That automatically disqualifies a platform from any sort of safe harbor protections, and so the owners of the ChatGPT platform would be in violation.
The problem with AI as it currently stands is that it has no actual comprehension of the prompt, or ability to make leaps of logic, nor does it have the ability to extend and build upon existing work to legitimately transform it, except by using other works already fed into its model. All it can do is blend a bunch of shit together to make something that meets a set of criteria. There’s little actual fundamental difference between what ChatGPT does and what a procedurally generated game like most roguelikes do–the only real difference is that ChatGPT uses a prompt while a roguelike uses a RNG seed. In both cases, though, the resulting product is limited solely to the assets available to it, and if I made a roguelike that used assets ripped straight from Mario, Zelda, Mass Effect, Crash Bandicoot, Resident Evil, and Undertale, I’d be slapped with a cease and desist fast enough to make my head spin.
The fact that OpenAI stole content from everybody in order to make its model doesn’t make it less infringing.
The fact that OpenAI stole content from everybody in order to make its model doesn’t make it less infringing.
Totally in agreement with you here. They did something wrong and should have to deal with that.
But my question is more about…
The problem with AI as it currently stands is that it has no actual comprehension of the prompt, or ability to make leaps of logic, nor does it have the ability to extend and build upon existing work to legitimately transform it, except by using other works already fed into its model
Is comprehension necessary for breaking copyright infringement? Is it really about a creator being able to be logical or to extend concepts?
I think we have a definition problem with exactly what the issue is. This may be a little too philosophical but what part of you isn’t processing your historical experiences and generating derivative works? When I saw “dog” the thing that pops into your head is an amalgamation of your past experiences and visuals of dogs. Is the only difference between you and a computer the fact that you had experiences with non created works while the AI is explicitly fed created content?
AI could be created with a bit of randomness added in to make what it generates “creative” instead of derivative but I’m wondering what level of pure noise needs to be added to be considered created by AI? Can any of us truly create something that isn’t in some part derivative?
There’s little actual fundamental difference between what ChatGPT does and what a procedurally generated game like most roguelikes do
Agreed. I think at this point we are in a strange place because most people think ChatGPT is a far bigger leap in technology than it truly is. It’s biggest achievement was being able to process synthesized data fast enough to make it feel conversational.
What worries me is that we will set laws and legal precedent based on a fundamental misunderstanding of what the technology does. I fear that had all the sample data been acquired legally people would still have the same argument think their creations exist inside the AI in some full context when it’s really just synthesized down to what is necessary to answer the question posed “what’s the statically most likely next word of this sentence?”
Is comprehension necessary for breaking copyright infringement? Is it really about a creator being able to be logical or to extend concepts?
I think we have a definition problem with exactly what the issue is. This may be a little too philosophical but what part of you isn’t processing your historical experiences and generating derivative works? When I saw “dog” the thing that pops into your head is an amalgamation of your past experiences and visuals of dogs. Is the only difference between you and a computer the fact that you had experiences with non created works while the AI is explicitly fed created content?
That’s part of it, yes, but nowhere near the whole issue.
I think someone else summarized my issue with AI elsewhere in this thread–AI as it currently stands is fundamentally plagiaristic, because it cannot be anything more than the average of its inputs, and cannot be greater than the sum of its inputs. If you ask ChatGPT to summarize the plot of The Matrix and write a brief analysis of the themes and its opinions, ChatGPT doesn’t watch the movie, do its own analysis, and give you its own summary; instead, it will pull up the part of the database it was fed into by its learning model that relates to “The Matrix,” “movie summaries,” “movie analysis,” find what parts of its training dataset matches up to the prompt–likely an article written by Roger Ebert, maybe some scholarly articles, maybe some metacritic reviews–and spit out a response that combines those parts together into something that sounds relatively coherent.
Another issue, in my opinion, is that ChatGPT can’t take general concepts and extend them further. To go back to the movie summary example, if you asked a regular layperson human to analyze the themes in The Matrix, they would likely focus on the cool gun battles and neat special effects. If you had that same layperson attend a four-year college and receive a bachelor’s in media studies, then asked them to do the exact same analysis of The Matrix, their answer would be drastically different, even if their entire degree did not discuss The Matrix even once. This is because that layperson is (or at least should be) capable of taking generalized concepts and applying them to specific scenarios–in other words, a layperson can take the media analysis concepts they learned while earning that four-year degree, and apply them to a specific thing, even if those concepts weren’t explicitly applied to that thing. AI, as it currently stands, is incapable of this. As another example, let’s say a brand-new computing language came out tomorrow that was entirely unrelated to any currently existing computing languages. AI would be nigh-useless at analyzing and helping produce new code for that language–even if it were dead simple to use and understand–until enough humans published code samples that could be fed into the AI’s training model.
