Microsoft has claimed that something like 30% of the code for Windows is AI-generated. (I have not verified that they even said that, maybe that’s totally fake.) Last i checked AI-generated stuff cannot be copyrighted. Does this mean that significant portions of Windows are un-copyrightable, and MacroSlop couldn’t do a damn thing if someone leaked those parts?
And then, reservations about including slop in a software project aside, what could this mean for things like ReactOS? They’ve been careful to avoid including any MS-owned code, but nobody owns slop code.
If you can get at the 30% of the code without any human modification, then that may be in practice public domain.
Now you have to figure out which 30%, without any human modification, and have the ability to prove the specific code is AI generated.
This is why the companies do not sweat the detail that the output of the LLM might not qualify for copyright, the only way it is a problem is if they label it as such and pass it through without modification.
In the cases proving only humans can hold copyright, the human that wanted copyright had essentially bragged that the works in dispute were 100% not created by humans. One a human has done almost anything to the output, that modification is subject to copyright.
So if the company keeps the details secret and/or mixes it with human output, the copyright issue is moot.
Rules are enforced differently for corporations than they are for the rest of us.
Even if it’s not copyright eligible, they still don’t have any obligations to release the sources
Oh, I disagree. The GNU GPL, which is one of the most widely used open source licenses, is a copyleft license. In practice, this means that any software which uses GPL code (a derivative work) also falls under the GPL and must abide by its terms.
In purely private (or internal) use—with no sales and no distribution—the software code may be modified and parts reused without requiring the source code to be released. For sales or distribution, the entire source code needs to be made available to end users, including any code changes and additions—in that case, copyleft is applied to ensure that end users retain the freedoms defined above.
Only if GPL-licensed parts are used in a program (and the program is distributed) must all other source code of the program be made available under the same license terms.
https://en.wikipedia.org/wiki/GNU_General_Public_License
Therefore, if a code generating model uses any code derived from GPL projects, the generated code is also GPL and must be made available if the code is to be distributed. The GPL specifically obligates release of the source code.
Considering how widely and indiscriminately the scraping has been done to feed training data into the bots, the burden should be on Microsoft to prove that any generated code they use is not derived from any GPL code.
All Microsoft has to do is give the presiding judge a few bags of money, and it will be determined that any suit that accuses Microsoft of using GPL code has no standing.
Depends on the country. Companies haven’t been able to do this in EU
I’ve been on top of these topics for years and I’ve somehow never thought of it like this. It’s not like they can remove some of the training material from the finished model. If they fucked upand included open-source training data, it should all be open-source and they can eat shit and get wrecked.
That’d be true if it was just copypasting GPL code, which is not really true except on very simplistic high level view of “it’s seen GPL code, this it knows to recreate similar code”
Which is same bullshit arguments used against people working on foss products when they’ve worked on related p*opietary solutions before (see windows code/wine for example)
Also I have a controversial take here; GPL is a bullshit proprietary license due to that cancer of a requirement. Any license that limits you in any way using the code you just downloaded is cancer just as any other proprietary code. Obviously it’s working “fine” in the crapitalist world we live in, but it’s still shit
Also I have a controversial take here; GPL is a bullshit proprietary license due to that cancer of a requirement. Any license that limits you in any way using the code you just downloaded is cancer just as any other proprietary code. Obviously it’s working “fine” in the crapitalist world we live in, but it’s still shit
You overlook that without a license, you do not have the right to either copy the source code, or copy the binaries, or copy the software.
It is right that copyright is used by corporations to make further their interests and make more money. But that is not why copyright was invented - it was invented to provide income to writers, artists, musicians, composers, and that was extended to software developers.
The GPL is not supportive of capitalism in itself - it is more like a hack within the existing legal system that is geared to the interests of corporation. But the copyright law it is based on stems from the social contract to support artists, creators, and inventors.
And with AI, corporations are trying to undo that contract.
Things will become interesting since AI seems also capable to decompile binary code to understandable source code. Which would make it possible to re-implement a lot of proprietay interfaces.
There’s still someone mapping out the features, designs etc b4 the AI goes to create it. Also saying 30% of code was AI generated hides cost of making the correct input: Planning, already existing code, Testing & Feedback loop, Reviewing. Its like saying Roofers aren’t needed because 100% of nails were done with a nailgun.
Some tasks are easy with AI, others aren’t but even when the code itself was 100% autonomous AI generated, the features, design was mapped out by someone is probably sufficient to retain copyright.
An image created by a five word prompt does not have a creative depth in creation like an image created by AI but with many refinement steps to get something truly unique.
I think this would be a case where copyright can only be held over some parts of whatever is being made.
Like Zarya of the Dawn. The text and panel layouts were done by a human author, but all the visuals were done by AI. The author got it copyrighted, bragged about the AI part, and had that copyright revoked. The author owns the dialog and layout and everything else that’s not a picture, but nobody owns the visuals.
Even if it can’t be copyrighted anymore, it’s still under trademark. And it’s a shit OS, so why bother rebranding a forked version?
Doesn’t trademark just protect the name “Microsoft Windows®” and slogans (“Just Bing it®”)?
And yeah, i don’t think anyone who knows enough about computers to make an operating system wants to just fork Windows. I’m more curious about how much MS could be hurting themselves by having a robot write important code.
Let me give you an example that can help answer your question.
I, a shitbag “entrepreneur”, find and open source project I like that does 95% of what I want. I proceed to feed the code, design docs, and anything else I can find from the project to an AI. I then ask the AI to write the application from scratch using what I fed it as “inspiration”, and add the 5% functionality I want. Now, let’s say the AI gets it pretty close to working. Close enough that I can hire a dev to clean it up and get it going.
At the moment, that project is not covered by the open source license since the original code is gone and I used my own resources to “develop” it. I state “at the moment” because there is much debate around this.
AI gets it pretty close to working. Close enough that I can hire a dev to clean it up and get it going.
This simply just isn’t how that works. 95% working can very, very easily be 100% shit. Because of the lack of professional guidance and structure, the last 5% could be very difficult or impossible without a complete restructure and rewrite or the whole code base.
This has been true since outsourcing code become popular. So many times I’ve been asked to “just finish the last 5%” just to have to explain that the last 5% is actually 150% the amount of work (now I have to untangle this mess).
At the moment, that project is not covered by the open source license since the original code is gone
That’s not how most open source licenses work. Which is really the crux of the issue. If the new code was based on the open source code in any way it is now open source (at least for GPL code).
If you took a popular open source project, looked at the code and rewrote it by hand, the resulting code is open source. Even if you wrote it in a totally new programming language. I don’t see how AI reading and reproducing the code is much different in this sense.
If you took a popular open source project, looked at the code and rewrote it by hand, the resulting code is open source. Even if you wrote it in a totally new programming language.
I completely agree with you BUT this is the mess that is literally playing out now. We need legal clarity because the scenario I wrote is not theoretical, it is happening now.
Also, I am sad to say that I have already seen Claude Fable 5 spit out code to spec that a mid level dev got working in less than 30 minutes by fixing a few connection assumptions the model made. Afterwards, every iteration has been done with the less powerful Claude Opus.
I refuse to believe that the LLM designed everything from scratch and I can almost guarantee that whatever it “knows”, it pulled from open source projects. It also explains why it is so good at Java, Bash, and Python, and absolute garbage at PL-SQL.
They’re probably safe until some governmental organization builds up enough motivation and money to try and investigate it. And given the reaming they gave the IRS I doubt anyone is up for it.
Beyond that, I can’t imagine the sheer legal nightmare it would be, given that MS will hire the best fucking lawyers money can buy. How do you quantify how much is AI generated and how much isn’t? How much involvement of AI makes it “AI code”? You could spend years in court arguing about where that particular line is.
It’s a fun hypothetical, but unlikely to matter to any real degree, at least for a company their size. Maybe it could be used against some startup making dumb public statements.
