https://thehill.com/policy/technology/4162573-ai-art-cant-earn-copyright-judge-rules/ give us a preview that those that use AI, ML, GPT and such may be on uncertain ground if they want to copyright the code.

You may not be aware that most of the online AI/ML/GPT systems keep copies of what you generate. So if you are creating code for your company, the future may have a surprise waiting for you.

AndreRet commented: Surprised, who me, nah, not at all :) +15

This will be interesting to follow through the appeals. If they allow the AI creator to copyright, couldn't the dataset author(s) also claim a copyright? If the AI works can't be copyrighted, then at least a portion would be derivative work not covered by copyright, which I believe is what you're suggesting.

It makes sense to me that if I run a prompt (of my own creation) through an AI system (one that I did not write), I should not hold the copyright to its output.

commented: If I run original C code through a C compiler that I do not own, do I own the binary and executable output? +3

Where do you draw the line? If I set up parameters for ray-tracing and run them through "generative" software, would I own the rights to the result? Pre AI I believe I would have, but now it could be argued that I would not. Consider a tool like Topaz Photo AI. Would it be possible that I could own the rights to a blurry photo (input) but not the enhanced output? Would it depend on the amount on enhancement?

Very fine line IMO. If I copy and paste "as-is" generated code, by all means, the code does not belong to me. If I take snippets or change the code completely based on the generated code, does it mean that I still don't have the copy right? Think of cars, they all work the same, share very similar technologies etc yet each manufacturer has rights to his products which is unique to them but a car is a car shared by all...

In the same breath though, what if copied and pasted a code sample say from StackOverflow or Daniweb, does it belong to that site, the writer, me? We never had issues around this unless a license stated otherwise, I would say they should look at it in the same way.

As to the post regarding created images, what if I add a few brush-strokes of my own, does it make it mine as I have altered the "look"?

When I joined in 2010 I thought there was a clause in the TOS that said Dani owned the rights to any software posted on the site. It's possible (gasp) that I am mistaken, or that the TOS has been modified since then. In any case, I gave up my desire to be paid for code when I retired and as far as I am concerned she is welcome to anything I post.

Having said that, given the capabilities of modern IDEs, changing variable names, etc. is trivial so it would be difficult to prove that the modified code is a rip-off of the unmodified code, at least for interpreted code like Python. I would imagine it would be much more difficult for compiled code.

commented: Exactly my sentiments. +15

Where do you draw the line?

My layman's expectation with copyright law is that I would tend to own the copyright to something that came out of my brain and I wrote and, subsequently, chose to self-publish. However, if I prompt an AI platform to write something, then the AI platform generates something, and may even add it to its own database before even showing it to me. At that point, by the time I've seen it for the first time, it's been created by another entity and stored in that entity's database. I might not even be the first person to read it. (Let's make the absurd, yet technically feasible, assumption that ChatGPT's engineers are watching query responses in realtime, and read the response before it's sent over to you.) What claim would I have to it?

Would it be possible that I could own the rights to a blurry photo (input) but not the enhanced output?

What does existing copyright law say today about me handing a photography expert an old photo, and that expert manually colorizing and enhancing the photo with their own skillset?

When I joined in 2010 I thought there was a clause in the TOS that said Dani owned the rights to any software posted on the site.

Yes, back since 2002 there was a clause in the TOS saying that DaniWeb LLC became the copyright holder of any content published to it by its members. The purpose was to prevent people from demanding that I randomly delete 100 of their posts because they retain the copyright to them and can do what they want with them. By owning the content, it also gave me the right to serve takedown notices to scraper sites, etc.

This is similar to how an author, by default, is also the owner of the copyright. However, should the author want to use a publisher to get their work out there, the author may transfer the copyright over to a publisher, and then the book that you read in your hands might say (c) Random House or something like that.

Today, with the explosion of social media, there are now lots of laws governing copyright of social media content. Therefore, our terms of service now simply states, "All data submitted to DaniWeb, with the exception of any personally identifiable information, may reside on our servers indefinitely." My interpretation of that is that, by posting on DaniWeb, you are giving us a non-exclusive license to publish the content forever. It's also in our community rules that you must own the intellectual property rights to what you post. That means that you must have the legal right to license the content to us in the first place. For this reason, it's not against our rules for someone to post the exact same question on DaniWeb as they have published on any other site. However, it is against our rules for someone to publish content on DaniWeb that they did not write (which includes AI-generated content), and therefore don't have the right to either transfer copyright over, or license, to a publisher such as DaniWeb.

Forum posts are different for editorial content on DaniWeb, in which we retain an exclusive right to the content, and it cannot be published elsewhere.

commented: Thanks for expounding about "since 2002." Makes a lot of sense. +17

if I prompt an AI platform to write something, then the AI platform generates something, and may even add it to its own database before even showing it to me. At that point, by the time I've seen it for the first time, it's been created by another entity and stored in that entity's database.

If you use an IDE that supports code snippets, by inserting a code snippet are you giving up any of your intellectual property rights to the people who created the IDE and supplied the code snippet library?

What does existing copyright law say today about me handing a photography expert an old photo, and that expert manually colorizing and enhancing the photo with their own skillset?

I have no idea.

This is similar to how an author, by default, is also the owner of the copyright.

There is a fair use doctrine (I think that is the name) that protects people who do things like use quotes from known works, or who parody the works of others. I'm not sure exactly how this works since I have heard that Weird Al cannot parody songs without the permission of the original artist. If fair doctrine applies to code, how much code would I be able to "borrow"? If I copy/paste a code snippet from the web, am I violating copyright?

I'm asking out of curiosity since I have no intention of ever asking for payment for anything I write.

If you use an IDE that supports code snippets, by inserting a code snippet are you giving up any of your intellectual property rights to the people who created the IDE and supplied the code snippet library?

No, because an IDE is like a word processor. You're not giving up your intellectual property when you compose a novel in Microsoft Word. The difference between your IDE and AI is that, with your IDE, the thoughts start in your head, then get typed out by your fingers, and the final output came entirely out of your imagination. With AI, you might give it some prompts, but it computes, on its own, what it's going to write, and then it serves you its final output.

There is a fair use doctrine (I think that is the name) that protects people who do things like use quotes from known works, or who parody the works of others.

That's why, in the context of a discussion such as this one, you can compose in your post, "ChatGPT generated this sentence:" and then quote a sentence that it gave you (which you don't hold the copyright for). But you can't copy/paste what ChatGPT spits out and post it purporting it to be your own intellectual property.

In my view the entire conversation is needlessly complicated by the way we apply the term "intelligence" to the technology. The truth is that there is nothing in the machine that remotely resembles anything we would normally understand as intelligent:

  • The training data is curated by human
  • The trainable algorithm is chosen by human
  • Human decides when the trained model's predictions are sufficiently accurate
  • Once trained the resulting model produces deterministic results

Take ChatGPT for example. If you set the temperature to zero for your conversation you will get exactly the same response to any given prompt every single time. Try this:

  • Start a new ChatGPT session
  • Enter the prompt: Please use a temperature of zero for this entire conversation.
  • Enter the prompt: Please complete this sentence: In the beginning, God

Every time, ChatGPT will answer: In the beginning, God created the heavens and the earth.

If you try setting higher temperatures (range is 0 to 1) you will get increasingly unpredictable (i.e. more highly randomised) results. The key point I'm making here is that the "creativity" has nothing to do with the trained model; it is all in post-processing the results.

So now to the original issue: can you claim copyright over the output from a trained model? There is definitely an argument for it:

  1. You devised the prompt that resulted in the output. You could absolutely claim copyright over that!
  2. You decided whether the result was acceptable.
  3. If you decided that the result wasn't acceptable, then you devised the additional prompts required to refine the output until you were satisfied. You could absolutely claim copyright over those additional prompts too.
  4. Neither the model (which is deterministic) nor the application you used to communicate with the trained model has supplied anything that could be interpreted as creative, only random.

It is the same as if you used a graphic drawing tool and blurred part of the image, or used an IDE to obfuscate or refactor your code.

Ultimately the answer should have to come from the licence under which you used the application. Does it preclude you from claiming copyright over the output?

However, having said all this, I have no faith that the legal minds applying themselves to these cases have enough technical depth to even ask the questions that would lead them to conclude that the claims of trainable algorithms to intelligence are complete bunk.

commented: Rightly said! +15

So now to the original issue: can you claim copyright over the output from a trained model?

I agree with you that AI generators aren’t “intelligent”, but rather everything they generate is derivative of the information they were trained with. They were also programmed to respond to prompts by using their training data.

Where we disagree is that I think that the copyright holder of their output is OpenAI corporation, and not the individuals who entered prompts. For me, the end users of the software are utilizing a piece of software, and the magic and decision making comes from the software, not from the prompt.

commented: “Any sufficiently advanced technology is indistinguishable from magic.” - Asimov +17
commented: Actually, that was an Arthur C. Clarke quote. +15
commented: I suggest that contract law may enter the picture. Specifically, you requested the AI create code on your behalf; as such, it is your code. +7

So if I use Excel to create an interesting spreadsheet Microsoft should be the copyright owner?

commented: No, but I think Microsoft should own the magic behind its macros. The end user is just plugging input into them and MS decides what the output is. +34

From what I gather, no, because all the "creation" was done by you.

I don't think so, it is copyrightable. NO it's not.

commented: Yes! No! Wait, it's No! Yes! Maybe it's yes, maybe it's no? No? +17

Interesting read: District Court Ruling I think it raises some questions. It seems while the plaintiff later tried to claim he provided the instructions in his AI model to create the image, his application for copyright was based on his "ownership of the machine." The requirement of human involvement doesn't definitively define to what extent.

Keep a record of your work: This includes the code and any notes or documentation you create. This will help you prove that you are the code's original author.

As in GPL, those reusing or modifying original work contribute to whole. Only difference is AI generated code uses snippets created by humans and adds value by speeding up the process so stubs can be generated into full code saving programmers time. AI is a supplement to coding not a replacement.

Great article, definitely something to consider as AI becomes more entwined with our work processes.

A programmer (or programmers) going by the alias Nea Paw has created an AI application called CounterCloud that automatically generates responses to Russian and Chinese disinformation posts/tweets. We have now entered the era when information/disinformation generated by bots will flood the internet and we will be unable to wade through the BS to determine what is reliable and what is not.

Sort of like now, except exponentially worse.

source

Well most AI generators out there claim that all (or in some conditions) of the results that you create can be copyrighted by you. The conditions usually is to have a paid membership (of course). But let me dive a little bit more on this. If I try to create a new image in e.g. midjourney and I edited my own images that goes with it and make numerous attempts defining some data , then of course I should own the result if I paid for my subscription , and can copyright it. If you run your own AI models even if the core of them is some freeware that you have modified e.g. you can copyright what is appropriate to you. For example if you remove the background of a free image or an image that you have licence to use , you can't copyright the original image but the result of it you can. I will not dive deep to American law (that I don't know) but just pure logic. The context that applies to images that are more straightforward applies to other domains as well

commented: Agree 100%, especially on paying subscription fees. +15

I agree 100% with jkon, if I pay for a product (subscription fees), the product belongs to the lessor but anything else I do with it belongs to me, IMHO.

Yesterday (Sept 20) several authors, including John Grisham and George R R Martin, filed a lawsuit against OpenAI alleging that their LLM was illegally trained using pirated copies of their books. Their major concern is that, once trained, ChatGPT could be used to generate stories in the style of famous authors.

That raises two immediate questions.

  1. How did they determine that the books used to train the LLM were pirated?
  2. How is new material "in the style of..." a copyright violation?

I can understand the first claim in that the books would have to be either scanned and converted to text, or obtained in digital form, before being fed to the LLM for training. In that sense the books have been digitally reproduced without permission.

As for the second claim, according the their reasoning (I assume), were I to read all of John Grisham's books, then write a new story in his style (which I could not help being heavily influenced by), I could be liable to a copyright infringement suit. Assuming that this would not be the case, how is that different from an AI generated "in the style of" story? Typically, authors claim copyright on ideas, not styles. For example, Harlan Ellison famously sued anyone who used his ideas. He sued James Cameron when Cameron remarked that he got his idea for Terminator from an Outer Limits episode written by Ellison years before (which is interesting since Ellison himself admitted that he got his idea from an earlier story written by someone else).

Be a part of the DaniWeb community

We're a friendly, industry-focused community of developers, IT pros, digital marketers, and technology enthusiasts meeting, networking, learning, and sharing knowledge.