Comment by simonw

5 months ago

I feel like the more important question here is whether AI-generated code can be copyrighted.

Companies responsible for several billion dollars worth of software written over the past ~36 months would really like to know the answer to that one.

For many, many reasons, I suspect patents will become much more important now. After all, it's the ideas that matter now. Which I maintain, has always been the case, because "execution" is nothing more than a series of smaller ideas, except those typically needed money. How convenient for those with capital!

Patents have the drawback of being expensive and very slow to acquire, but having worked on a bunch, they are uniquely suited to be radically optimized by GenAI.

Also patents are very flawed in practice, but the only real protection that is left. Copyright is meaningless when, as people have done, you can reproduce entire saas products by feeding AI screenshots.

Intellectual Property as a whole has been in need for a revamp for a while now, but it's even more critical in the age of AI.

  • Patents aren't going to cover the vast majority of what exists as code because patents only cover inventions and not common every day code.

    > Copyright is meaningless when, as people have done, you can reproduce entire saas products by feeding AI screenshots.

    What do you mean?

    • > What do you mean?

      People are taking screenshots of existing products and feeding them to AI to reproduce them from scratch. From this site: https://news.ycombinator.com/item?id=47257803

      > Patents aren't going to cover the vast majority of what exists as code because patents only cover inventions and not common every day code.

      Agreed, but as we're finding out, everyday code is now cheap enough that there will be questions about how much of it is worth protecting... and, given the unclear stance on AI-generated code, maybe even if it is protectable at all.

      Yes, the bar for patents is way too high for every day code, but you can always get one by making the claims narrow enough. I think that will actually serve one of the underlying goals of patents by incentivizing people to build things that are actually novel and non-obvious rather than just a slight variation of another project.

      For projects where there is absolutely nothing technically novel (which is rarely true, cf "claims narrow enough") moats could be in hoarding data, network effects, and the like.

      Note, I'm not necessarily happy about all this, this is just how I think it could play out absent larger changes to IP laws.

      4 replies →

  • Patents are what allowed the industrial revolution to happen. No one is bringing a cotton gin from idea to design to manufacturing to market if the second you release it every manufacturing company in the world can start making their won.

Perhaps. Even that may not be important if the METR progress line continues much longer, because then all those billion dollars "worth" of software written over the past 3-ish years get re-invented for cents on the dollar.

Separately, I think code is more like an invention than a work of art, and should have been subject only to patent laws instead of (and not in addition to!) copyright laws. This doesn't really make much difference now, as AI doesn't (at least in the UK) have personhood for either copyright or patent law: https://www.briffa.com/blog/can-you-obtain-a-patent-for-inve...

I would assume that the same “substantial human authorship” criterion applies. Copyright is about human creativity, it doesn’t otherwise matter if something is art, prose, typesetting, or code.

  • Yep. Not just copyright but patents as well

    > The Supreme Court previously rejected Thaler's request to hear his argument in a separate case involving prototypes for a beverage holder and a light beacon concerning whether AI-generated inventions should be eligible for U.S. patent protection. His patent applications were rejected by the U.S. Patent and Trademark Office on similar grounds.

    https://www.reuters.com/legal/government/us-supreme-court-de...

It doesn't really matter as long as you keep physical control of the code and don't let others copy it.

  • That would effectively rely on the doctrine of trade secret rather than copyright. A major difference is that accidental or malicious disclosure of a trade secret usually ends the trade secret status, forever. In an alternate universe where computer source code had never been copyrightable, famous leaks (Microsoft Windows, 2004; id Quake, 1997) would have effectively open-sourced those codebases, and other companies could have openly and legally used them.

    As source code becomes more of a generated artifact of software development the way object code is an artifact of compilation, we might be moving toward a world where secrecy, constant forward motion, and moats become even more of an asset (vs plain IP protection).

    • > would have effectively open-sourced those codebases, and other companies could have openly and legally used them.

      It's actually better if we keep re-creating the wheel. Keeps more people employed.

  • Nor does it matter if code has no value.

    I do think what happens in this case is SCOTUS will ultimately rule that AI-built code is copyrightable while art is not. I'm sure there's some rationale thick enough for them.

    • It's strange how hard it is to think of a situation that could lead to that case. Who would bother filing an infringement lawsuit for code whose very existence proves that it can be derived by anyone from LLM prompts? What would the damages even be?

      Interesting world we live in. Soon it'll be faster to one-shot the tiny slice of functionality I need from Adobe CS than to navigate their subscription cancellation obstacle course.

      3 replies →

Can you imagine the chaos if suddenly all the slop code wasn’t owned by the company? Even though that result would be consistent with this ruling, it undermines the narrative the economy is now riding on, so there will likely be special exemption.