All AI-generated code should be Open Source
76 points by Yogthos
76 points by Yogthos
I personally believe the world would be a better place if all software was Free and Open-Source Software, regardless of how it was generated. If there's one wish I have with the current "AI" craze is that copyright as we know it goes away or at least is drastically reduced. Copyright terms of less than 20 years and no corporate ownership of copyright, only individuals.
I agree completely. Of all the AI doom scenarios, the deflation of copyright law is the one I welcome. I'd throw in patent law as well. I used to release all my code under the public domain, and only in recent years switched to the MIT license for pragmatic reasons (some countries like Germany don't recognize public domain dedications). The only form of intellectual property I think is reasonable is trademark law, as a prevention of fraud.
I completely agree. Trademark laws have some value in consumer protection, though they are still sometimes abused. Copyright in the US has clearly failed in serving creators and has long been a tool used by corporations to opress creators. Whether it's through the parasitic nature of things like record labels or going after hobbyist projects and archivists (I'm looking at you, Nintendo), it's a mess and not in the public's best interest.
(Seeing your avatar, I immediately feel a kinship with you. Lunar is a very important series to me.)
I'd throw in patent law
My name is on a few software patents, and I wholeheartedly agreed.
Patents serve the exact opposite purpose they are supposed to. In theory, they favor innovation by ensuring small players can make some profit out of their invention. Innovation/progress being driven by seeking profit is debatable to begin with, but even if you agree with this premise, it just does not work out at all in practice. Only giant corps can enforce their patents, which they more often than not have bought as weapons.
Have you ever met a "software patent consultant"? Their job is to find "holes" in competitors patents, to fill them with your own. Whatever your software can actually do does not really matter at all, if it can maybe one day theoretically do something, they will try to convince you to cover it in your patent. Also, patent organizations basically have no way of verifying you actually can do whatever you claim.
For other domains, it might be more nuanced, but for software, patents have to go away.
I bristled a bit at: "Open Source code written since RMS announced the GNU Project and the Free Software Movement."
Code sharing was alive and well before GNU. I fact RMS often quotes the "sharing culture" as something he wanted to revive.
Same! Code sharing was once assumed, based in part on the idea that code was assumed to not be subject to copyright. The GNU project codified values that were previously broadly held but upended by court decisions. What RMS brought was his zeal.
My general feeling is as the "sum of human knowledge" (for lack of a better term, it's all lossy interpretation of the training data ofc) and as an answer for the copyrighted training data problem, models should be in the public domain, and ideally the output from them as well. It's absurd to me to claim copyright and licensing on the resulting model considering all the labs almost certainly violated licenses on the training data, and incredibly hypocritical to cry about distillation.
The same thing should apply to LLM-generated output in general. It's absurd to allow AI companies to vacuum up all humanity's output and then claim to own the result.
While I agree with this article on the first premise (as lofty of a goal as it may be)
The second reason is legal: purely AI-generated code is not written by a human, meaning it should not be copyrightable
This is true, but non-copywritable does not mean open source or source available.
Like photo generated with Camara? How do you define written by human?
It's well defined in copyright law already but the specifics may need discussion in court in some cases. A camera is a tool and the majority of the choices in the output came from a human. If the camera was mounted on a device controlled by some form of AI, it would likely be considered not created by a human and thus uncopyrightable.
it would likely be considered not created by a human
I'm not criticizing you at all, but you refer to an important point. As I understand it, your "likely be considered" is true, and there's things of interest inside it. The USA legal system is adversarial, and a trial of some sort would be necessary to ultimately settle the question. Despite everyone's best efforts to stay inside the law, sometimes trials give an apparently bad result. ABC vs Aereo seems like an example. Aereo went to great lengths to stay inside what statutes and previous cases had laid out, but still lost, leading engineers to decide that using a very long antenna cable could constitute copyright infringement. So "likely" is the correct phrasing, and copyright is very odd.
For myself, beyond the copyright angle, there is a basic trust and auditability issue here. If companies are going to ship vibe-coded software to production, users need source access so obvious security regressions actually get caught. Look at Humble Bundle's recent redesign. They forced a half-baked migration straight to live users that stripped out app-based two-factor authentication and let you change passwords without entering the old one. Had that code been open or at least source-available during an opt-in beta, people would have flagged those glaring flaws before launch. Without any transparency or accountability, deleting my account was the only sensible move.
Under what license? If I have licensed code under BSD and then it's used by a GPL project then (1) that project has relicensed code they don't own, and more importantly I cannot use any changes to that code as it is nominally GPL.
On the other end: I release code under the GPL, and it is incorporated into BSD or MIT licensed project, and then a corporation that uses my GPL code via the copy that was relicensed under the BSD license?
Or even more extreme: an open source project that is open source through saying it is public domain code - now that project is ostensibly making code under any OSS license public domain, making it free to use by anyone.
Opensource is not a single thing, and people disagree about what they think the correct approach is.
As I have said before, the reason corporations are pushing chatbot code is to launder OSS code and avoid any of the obligations that come with those licenses. It's why they're happy to use chatbots trained on OSS code but do not make their own code available for training: they know what the chatbots are doing, and they don't want other people using their IP, but are happy to steal the IP of OSS projects.
Thanks for posting this. When reading this article, I was thinking the general thrust seems right, but something bugged me about it. I think you nailed it.
However, I also think that the sooner we move beyond copyright on software, the better. Copyleft was invented to prevent closing up of code written by people who wanted their code to remain open. If, in a post-LLM society, all code would be considered uncopyrightable, we no longer need copyleft or open source as it currently exists.
I am not sure about the end result though. One would be able to freely copy any and all software, but that no longer matters in this day and age with SaaS. Being able to inspect and modify and then redistribute software is way more important IMO. But again, everything moving to online silos means the software itself doesn't really matter that much anymore.
Copyleft was invented to prevent closing up of code written by people who wanted their code to remain open
No, copyright licenses like the GPL were created so that non-lawyers had a way of sharing software in a way that prevented companies from taking it and claiming that because it was public, it was public domain. Stallman started this because he did not like the fact that all software you got had no source code available, and wanted to be able to write code where the source was available without allow companies to take it and incorporate into their own software without giving it back That required significant legal work to ensure the license was sufficient to cover all the potential abuses of companies, and could be enforced (you can easily write unenforceable terms).
The important thing is that if you remove copyright then you cannot own source code, if you cannot own it, you cannot restrict how it is used. e.g. absent copyright you roll back the entire concept of open source code to what companies want to be able to do: claim that public source code is public domain, and so they are free to use it as they want and not ever release the code they derive from it.
The ability to copyright law is the only thing that gives any OSS license any ability to put requirements on users - even simple acknowledgement of the use of the code.
that public source code is public domain, and so they are free to use it as they want and not ever release the code they derive from it.
If it were truly public domain they should have less reason to keep it to themselves. If copyright on code were to abolished, at the same time we should require all code to be public.
no, if it's public domain there is an exceptionally strong reason to not disclose it. If it's disclosed (or leaks) anyone can use it, without restriction (much like they could now copy what would previously have been protected by gpl or whatever). Exactly what the GPL was created to prevent.
Hear, hear! I'd love it if the law said you can't provide software or software-based services without also providing the source code.
If I have licensed code under BSD and then it's used by a GPL project then (1) that project has relicensed code they don't own
Er, no, that's not how it works...
The other project can easily vendor, link to, and bundle your code while meeting the requirements of your BSD license. This is done literally all the time. From big tech companies, to little tech companies, to volunteer-run Linux distributions.
There isn't a requirement for a project to have only one license, nor for all parts to be covered under the same license.
Using code from one project in another does not somehow change the license of the first.
It is of course completely possible to "steal" open source code by incorporating it into a project without following the requirements set forth by the license, but that's just regular old law-breaking.
They can include that other code, and that code they include retains its original license.
If you look at large projects, you'll see different licenses on some files, and the projects include all of the licenses.
And you can only steal property, and things like source code are only property if copyright law exists. Copyright law was invented specifically for this purpose.
Good news: everything made by AI is in the public domain.
This comes up a lot, and every time I have to give the same reply: in Thaler v. Perlmutter, the human involved explicitly said that he considered the AI the sole creator and tried to get the copyright assigned to the model. This has zero relevance to the more common cases where the human will assert that they themselves are the author.
The decision explicitly says this:
The Copyright Office, in fact, has allowed the registration of works made by human authors who use artificial intelligence. See Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190, 16,192 (March 16, 2023) (Whether a work made with artificial intelligence is registerable depends “on the circumstances, particularly how the AI tool operates and how it was used to create the final work.”
https://law.justia.com/cases/federal/appellate-courts/cadc/23-5233/23-5233-2025-03-18.html
Public domain doesn't equal open source. You don't have to publicly share something that contains public domain materials. It just means that anyone else can also use the bits that are public domain in your software because it isn't copyrightable. They can still copyright the codebase as a whole so you can't freely use, modify, etc the whole software.
There's a few logical counterpoints to this:
Training on copyrighted materials is not necessarily something that copyright law prohibits. NYT vs OpenAI should hopefully help clarify this. I think there's a reasonable argument that training is fair use. Which would mean that training on GPL code doesn't automatically trigger anything in the GPL license regarding derivative works.
Copyright covers the expression, in general unless it's a reproduction, a lot of code won't be the same expression of things or is the simplest obvious way to do something, which means it may not be infringing in the first place. This isn't necessarily even a fair use question.
Even when AI generates some non-copyrightable code, there's still the question of whether there's enough human contribution to push it into being an expression of human creativity. This is hard to determine in bulk and if you get it wrong by assuming that all AI generated code is uncopyrightable then you risk being sued. E.g. if I take something public domain and paint over it, the work is copyrightable, even though the thing behind it is not. This is the concept of a work containing both copyrightable and non-copyrightable material.
Even if AI generated code isn't copyrightable, that doesn't mean there's any obligation to publish it or make it open source. Public domain and open source aren't the same thing, and nothing about a work being uncopyrightable means it has to be made publicly available.
Given how easy it is to produce code now, I don't think code by itself is an asset anymore. It can be open source, public domain, whatever license, nobody cares. Unfortunately. Of course some code keeps its value, like the Linux kernel or some deep-tech/research stuff. But I think it's like less than 1% of the volume. Everything else is replicable, and copyright is irrelevant.