AI cannot be patent 'inventor', UK Supreme Court rules in landmark case

which is curious because an AI won't request a patent, so what benefit does he expect for applying on it's behalf? some infinite lifetime patent?

Yes, dude is on a quixotic quest to get his AI recognized as an inventor, and is basically being denied in every jurisdiction.

TFA links to an article about him being rejected in the US, https://www.reuters.com/legal/us-supreme-court-rejects-compu... . The interesting/surprising part for me is that Lawrence Lessig and some other academics wrote briefs supporting his claim.

There is an intersting wrinkle here, I haven't seen it mentioned so far.

In the UK a patent is invalid if you put an inventor on it who didn't make a significant and definable contribution to the invention.

If a patent is AI generated, and then labelled with a human inventor it may be argued that it is invalidated by the humans lack of inventive input.

I suppose this will vary from jurisdiction to jurisdiction, but for the most part, the idea of assigning a patent to a machine system seem precluded by the more general inability of machine systems to have legal rights at all other than the right of property to not be defaced or destroyed by anyone other than the property owner. Rights to ownership and control in general are typically limited to humans and collective entities such as governments and private corporations created to disaggregate the interests of communities of humans from the specific individuals making up that community at one particular time.

There have been some exceptions, such as Leona Helmsley bequeathing part of her estate to her dog, but just as with infants or mentally unfit humans who may possess rights of ownership but not understand them well enough to make appropriate decisions regarding disposition, that money was put into a trust.

Putting aside the philosophical and legal theory questions regarding what kind of entities can or should have rights and the ability to own property, there are more practical questions that would need to be answered. If you really wanted to assign a patent to an automated system that generated an idea, how would you do it? What is the boundary of this system separating it from its environment? We have a fairly clear answer to this when the system in question is a biological organism. But a software system? Is it the process ID of the first process spawned from the executable file you used to instantiate the system before querying it? Presumably that was killed at some point well before you ever filed the application. The server or servers it ran on? Those also may be ephemeral. Some content-addressing tagging system identifying a released version of the software? Well then the specific instances you ran to generate the idea shares ownership with a bunch of other instances that didn't do jack. The source code? Now it's shared with even more potential instances that don't exist yet and may never exist.

Do you snapshot the entire state machine consisting of register and memory contents across all hardware servers as they happened to exist at the moment of creation? Then what? As long as it is possible to physically realize this system by restoring the snapshot to currently powered on hardware with a sufficient similar processor architecture and operating system the system lives forever? But if that operating system or architecture disappears then it dies and the rights expire? Does the snapshot itself retain some kind of legal rights even if it is never physically realized and only exists as a file? Seemingly no, as even a potentially intelligent system that is in a state of hibernation has no interests and can't make decisions.

So now we have a further question. Even if we can somehow define an acceptable boundary of what even constitutes the entity that has rights, we have to define at what point it is still considered to be alive. It's not clear for even a single process on a single server that is getting pre-empted and potentially paged out anywhere from a few times a day to billions of times per second, but much less clear for a distributed system that relies upon replaying log-structured files shared across an unreliable network. Humans get death certificates when a physician determines the heart has been stopped long enough that it is unlikely to ever come back. But for a distributed system that relies on log consistency, it can always potentially come back as long as some combo of hardware and software exists that can interpret the state transition commands to restore its state and someone with the power to plug servers together, turn them on, and hit enter makes the decision to do it. The lifetime of such a system is arbitrary.

Maybe this is a good test for what isn't novel. If an LLM trained on our existing creative output can be prompted invent some patentable solution to a problem, then maybe that shouldn't be patentable. Tada, a much less stupid patent system.

So, if I tell Apple Computers at the end of last millennium 'you should make a music player' then I would have been the inventor of all the facets of the iPod?

Sometimes, but rarely, recognising the problem _is_ the invention; but outside of that I'd say the person in this process hasn't invented something and whatever development the computer generated should thus be unpatentable.

They'll at least get first-to-market benefits.

AI isn't legally an autonomous entity (... yet?)

That would be perjury?

Yeah a trivial bypass, anybody in say eastern Europe or Russia would immediately think about this path, that's one of million ways communism destroyed morality of whole region for generations to come. Change things a bit without affecting meaning and voila.

I see some dark path in patent world, corporations are amoral by principle so they will try to milk this even if just to defend themselves from everybody else doing it.

> In any case, most of our systems, and definitely how we manage intellectual property, are basically obsolete and require wholesale revisitation

I completely agree, but don't underestimate the power of the ones who own most of it to lobby in their favour for the decades to come.

That combined to a trend towards relocation of supply chains, more expensive energy, resources etc. makes me think that we will see the value of physical capital increase in proportion.

Please explain how agency is "just a matter of time". This is a pretty big statement you're just assuming to be true. I have seen no indication that current autonomous systems could achieve agency, or that anybody even considers that a development goal.

Also, I think it seems clear that LLMs and other generative models are poised to strengthen the moat of status quo beneficiaries. I'm not sure if this decision weakens that impact, but generative models are not the tools of the "little guys".

As someone who believes that patents and copyrights should be abolished, this is a great first step. If AI can do to innovation what it did to Chess where it made even centaur matches pointless, within a short time a significant amount of inventions will be unpatentable and public. In 10 years we can work to abolish the minority created before AI took over.

> he near future looks like a heterogeneous mixture of natural and artificial intelligences interacting on a level footing, enforced by legal protections.

using LLMs as tools already covers this.

But to your wider point, now that google have licensed most things, or have arranged some sort of deal with large IP creators, the whole "boo hiss copyright" movement has died down. This means that unless its in the interest of google, or some other exceptionally rich corp to water down IP laws, nothing much will change.

I suspect that the only thing will change is how data is licensed for LLM creation.

The person filing the patent has a lot of prior work in trying to get AI stuff legally recognised: https://www.wired.co.uk/article/stephen-thaler-ai-copyright-...

On other hand it cannot own property or earn money. So it is pretty pointless to sue one...

set anti-marketing && anti-neolanguage on:

" If the statistical database prepared by the team can not be rewarded the patent, can the team that configured the data that fed the vectorization of such database therefore also not be guilty of infringement ? "

There is no Johnny5, HAL, Skynet, etc. It is not expected that it will be seen in the coming decades, probably not even in this century at the very least.

You are thinking of wrong problem. It is not about use of AI in new patent. But if the AI can be the author.

Later of which would mean that AI would have some type of personhood like companies have... And currently courts consider that it should not as of now.

i think the idea of patent is to share knowledge needed to reproduce the invention. pitty that this is mostly unenforced.

Why? If the AI can write it, just ask for it yourself.

I'm not very excited for heaps of AI code to get released. I basically consider that already happened.

Such code would be a trade secret though. If a cola drink's secret sauce recipe was just a single plant in the Amazon, the name of that plant is in the public domain and cannot be copyrighted, and the plant itself probably can't be patented. All they have is knowing what it is and not telling anyone.

First, this case is a nothing burger. It only recognizes the obvious fact that you can't list the "name" of some program in the list of inventors for a patent.

Secondly, most code is typically protected by copyright, not patents. And copyright is actually not awardable for automatically generated code, except if the code is a derivative of an otherwise copyright able work. So compiled machine code is protected even though it was generated by a computer program, because it is a derivative of the human-written source code.

However, if you write a program which generates all C programs of 10 lines or shorter, you would not be able to claim copyright for those programs (though someone else who manually wrote one of them would still be able to have copyright over their version - provenance is important in copyright law).

Now, I don't think this has been tested in court yet, but it's very likely that code produced by ChatGPT or Copilot would be considered a derivative of the prompt provided, and thus would be protected by copyright just as much as a compiled binary.

I’ve been basically doing that all year and generated so much revenue that i dont care if someone copies and pastes anything i got AI to output

AI generated images on pitch decks, AI organized slides on said deck, investment achieved

AI generated songs on flashy websites, AI assisted code bases, revenue just fine

the philosophy is entertain the humans!