With that lens, I welcome gradually phasing this stuff out, especially as we navigate into the unknown game-theory landscape AI-as-inventors brings.
I need a bit more depth and detail to believe that this doesn't destroy the pharma industry.
What would the empirical evidence even look like? It's not like the modern pharma industry existed before patents.
The complicating factor is that as time passes our ability to reverse engineer has grown, however I'm not sure that invalidates the need for patents, the question is whether the new patents are being assessed well from a novelty / inventiveness perspective
I always assumed that intellectual property was invented in order to protect against a specific use case:
If researching a new product is extremely cost intensive. But once a product is invented, it is easy to reverse engineer how the product works. Then the first firm will need intellectual property to put in the initial cost, otherwise they will not do so, as they know they will not have enough time to recoup their costs in the market before a competitor moves in with a copy-cat product without having to paid the initial costs.
1) [the stronger one] while the scenario/narrative is a compelling one (or maybe it just feels compelling as I’ve heard it so many times), if it doesn’t have experimental/data backing I have to abandon it.
2) [the weaker one, as it replaces a narrative with another narrative within a complex system] I’ll only give the highlights as the arguments are a lot more eloquently laid out in the book; part of it is comparing the force of “many inventor nodes building on top of many invention nodes” vs “inventor nodes (with more investment individually?) building on top of fewer invention nodes”, part of it is the game theory effect of companies collectively investing less (proportionally) in R&D as the ROI from lawyers under this regime has more power, part of it was that actually, the reverse-engineering-simplicity story was too overblown and that actually the friction + domain knowledge has a stronger effect than people think (they published a paper on this). There were others, but it’s been a while now!
But doing away completely with patents would certainly stifle companies’ willingness to invest in R&D. They’d rather wait for someone else to invent something they can copy.
I’m not in favor of the current patent landscape, but doing away completely with them would likely be throwing out the baby with the bathwater.
I'm not making a judgement on what the ideal situation is, more so explaining why the referenced study could have come to its conclusion.
Then there is the fact that when something is patented, that has a chilling effect on competition, making the market less efficient.
There are also a lot of really silly patents that end up benefitting no-one, not even their inventor, but only result in needless litigation. The recent lawsuit between Nintendo and PocketPair comes to mind.
While there are cases in which patent law can help individual people profit from their invention, once all consequences are tallied, the overall effect of patent law on society appears to be negative.
I’m very much for not allowing trivial patents, but that’s independent from whether the invention was made by AI or by a human. The nature of the inventor should be immaterial for assessing the (non-)triviality of an invention.
I believe in many countries, the standard for a wide range of IP is that if something is largely produced by AI systems, it can not be patented / copyrighted / trademarked. It seems that "a significant" contribution must have been done by humans, that's the word you'll see again and again.
But I am not sure how one could prove that something is produced mostly by AI, or mostly by human. Right now anyone could use AI models to do most of the work, and just say or make up documentation that it is (major) human work.
The idea that an AI could have some sort of property rights is a nonstarter, legally speaking. It's just as invalid of a legal idea as claiming that a tree could have a patent on the shape of its leaf.
So when people go to the patent office and say "I didn't make this! an AI invented this", the obvious response from the patent office is "cool, well only humans get rights, and if you didn't make it, you can't get a patent on it, so too bad". This isn't a judgement of AI.
Now, a lot of people come to presume that this means that anything that AI touches is not subject to any IP rights -- but that's not what this means at all. Humans are allowed to use tools to create things that they have IP rights to. Your typewriter itself can't hold a copyright to a book, but if you use a typewriter, you can still hold the copyright to the book.
Ultimately, whether or not the use of AI is disqualifying to a human inventor doesn't really have anything to do with AI -- it all hinges on whether or not the human meets the requirements of holding the patent.
With that said, AI contribution should always be disclosed in every medium that it participated in, including patents.
implies that if he provided his name as the inventor, the application may not have been rejected.
Also, if AI generated content cannot be copyrighted, they can't infringe copyright as well
You can prove something is created by AI by e.g. showing the transcripts, especially from the vendor side.
You cannot prove that something isn't created with AI, at least not if you require incontrovertible proof (outside of, like, working in some kind of verifiably AI-free clean room, or doing something that current models are provably unable to demonstrate). But you certainly might be able to prove it to the satisfaction of the legal system.
If AI generated content cannot be copyrighted, it does not follow at all that they can't infringe copyright; there is no deductive step there that I can think of.
I assume the idea is that the fault/blame lies with the human(s) that caused the AI to generate something that violates copyright. Going back to previous comments, the typewriter that generated a document didn't infringe copyright - the person using it did.
Why not? Content that isn't under copyright can certainly infringe copyright.
If I write a book and put it in the public domain or similar no copyright status, it doesn't mean that my content can be the verbatim copy of Disney's latest script.
In a legal context what is necessary is evidence, not a math/logic formal proof.
> Also, if AI generated content cannot be copyrighted, they can't infringe copyright as well
Because AI cannot pay fines, go to jail, or be assigned the rights of a human. However, a human who uses AI can. If you use AI to infringe copyright, you have infringed the copyright, not the AI.
The US ruled similarly to Japan, but years ago, from copyrights through patents... from my limited POV, the LLMs (specifically) and art models are just regurgitating stolen concepts... echo'ing Picasso's great artists steal! mantra. The US has already ruled this is legal (e.g. newspaper content isn't "stolen" when a genAI summarizes it for a 3rd-party user).
Having sat with published authors, discussing their work/book with LLMs... it is really an interesting perspective on "readers' perspective(s)" [human ¬].
No, in the US AI output is ineligible for copyright not because "art models are just regurgitating stolen concepts", but because only human created works are eligible for protection.
>only works created by a human can be copyrighted under United States law, which excludes photographs and artwork created by animals or by machines without human intervention
https://en.wikipedia.org/wiki/Monkey_selfie_copyright_disput...
Or to not use a comma, to cram two incomplete sentences together whatever #FUamBOT =P
When people start pointing out spelling/formatting (in comments, no less...)... I'll typically just keep participating in discussions, elsewhere.
~~~~
I recently purchased a GPU capable of running 16GB models (5070Ti), so definitely understand how easy it is to be susceptible to bot/AI comments. This stuff is really powerful/convincing. It replaced a decade-old machine, and runs Ollama/Qwen/Mistral insanely responsively.
But I'm still commenting pure humanly written. My PObox is listed in my profile, and I'll hand-write anybody back a similarly-efforted response card.
Formally speaking, it's not the case, though this is commonly misunderstood. Statistical models are definitionally ampliative, otherwise they wouldn't be statistical. One can argue about it until they're blue in the face, but it almost always comes down to a misunderstanding of what the models are, what the mathematics behind them is a description of, and what the underlying logical structures represent.
The thing is that the position and objection to these models isn't actually a substantial, reasoned position where the words have a direct meaning. Though it's dressed up like reason, it's not the point. It's a kind of metaphor. This actually does reflect the nature of intellectual property law. The legal framework is knowingly illogical at an object-level, because the end its seeking is completely divorced from the means. It has to be, because the idea of intellectual property is absolutely unjustifiable in-and-of-itself. It's just a useful legal fiction to make sure people are getting paid by commoditizing ideation. That's not a bad thing, it just means you have to be mindful that bottom-up reason will lead you astray when dealing with it.
The plantiff is Stephen Thaler: https://imagination-engines.com/founder.html
The law does not recognize the anthropomorphization of inanimate objects.
But, since the income of a patent office is determined by how many patents they approve, one can dream ...
And for a very simple reason: you could easily overwhelm any intellectual property bureau just by having your AI drown them in AI slop. Even if most of these patents get refused, just refusing a patent is a lot of work, I imagine.
And even if you did, it's entirely inanimate, how would it even exercise them?
To my knowledge, he has notched only one win (i.e., granted patent) in South Africa, where patents are only cursorily examined [1].
The last word in the US is from the Federal Circuit a couple of years ago [2]. Same basic outcome: only a human being can be an inventor.
That said, the new Director of the USPTO has indicated that inventors should feel free to use AI however much they want as long as a human name is on the patent. However, it should be stressed that the Director's guidelines have not been litigated yet.
[1] https://artificialinventor.com/patent/
[2] https://www.cafc.uscourts.gov/opinions-orders/21-2347.OPINIO...
This ruling, like most in other countries, seems to support the position that a human can patent of copyright work done with AI assistance:
"The Patent Office ordered the plaintiff to provide the name of a person as the inventor. The plaintiff refused to do so, and the application was rejected."
Not sure about patents in the US but irt copyright, only the parts that are not LLM output are copyrightable. All LLM output is automatically public domain.
So if you have a work that was done with AI assistance, only the pieces of that work that are human authored can be subject to copyright. The AI parts cannot, if there are any.
I think it's long past time we get rid of the silly idea of intellectual property all together. If AI has the potential to do any good in the world in its current form, its that.
That is not exactly true under US law. You're simplifying what the copyright office has said to the point where you're missing the key points of what they were trying to convey.
The copyright office has affirmed multiple times that whether or not you use an LLM is irrelevant. Copyright eligibility requires "sufficient human-authored expressive elements". It doesn't matter what tools you use -- an LLM, a troop of trained monkeys, etc.
Ultimately all that matters is whether or not the human creativity involved qualifies. Because copyright is ultimately a right that protects human creativity.
So yes, if you put "write me a book" into ChatGPT -- that clearly does not quality for copyright. "Write me a book" itself is not creative enough for copyright.
Now on the other hand, if you spend 1000 hours writing a book, and you run it through ChatGPT for suggestions and/or edits -- there is no reason why that LLM output would not qualify.
https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...
It varies a lot in other countries, but in most (if not all) an AI cannot hold a copyright.
The only two things that can transact with any legal system in any way are humans and groups of humans.
document.getElementById("consentModal").remove()
document.getElementById("tpModal").remove()This is how the reverse centaur operation works. LLMs suck and not work in increasingly bad ways, and the companies who sell them treat them as one would buy psychic services (read: entertainment). So they need a token human to person-wash this slop.
the likelihood of one single guy having the same data scraping & storage capabilities as the big players, years before them (i see info about DABUS back to 2018), is slim.
The same applies to image generation - they can generate images that almost certainly were not in the training data.
Only after the participant has completed their grift or extraction operation then they begin virtue signalling their ‘morals’. It is fake.
If you are here for asserting morals, this is the wrong industry.
AI is a tool, like your keyboard or your code editor.
Those can't own patents. That doesn't mean anything produced by those tools is public domain, it just means the attribution has to belong to a human.
They can't produce anything on their own. They have to be prompted which is initiated by humans at this point, so the patents can be owned by the initiator(human) not the tool.
Someone wrote some instructions. No agent harness ever simply decided to pursue its own interests.
We will know when we see it. I don't see it right now.
Does a gradient descent algorithm pursue its interest of minimizing error? Does a home automation assistant pursue its interests when it sets my thermostat? I'm not super interested in the definition of "consciousness" or "interests". However, a thermostat setpoint has effects that are visible in the real world. That's a thing that happened, regardless whether you consider it to have happened in "the pursuit of an interest".
I'm saying that LLMs are affecting the world. And sometimes those effects might be difficult or impossible to trace back to a particular prompt written by a particular human. Chatbot input and output doesn't have to be in the form of text i/o. You can put them in a for loop. Remember OpenClaw?
> We will know when we see it. I don't see it right now.
There might exist an incentive to make it hard to see.
I haven't been able to square this belief (This is what i believe too.) with what I perceive as so, so many people making projects, putting them on github and slapping an MIT/GPL license on them.
If IP rights can't be applied to generated code then how are they able to apply a such a license to them?
I've asked this before and the response was along the lines of people thinking their multiple prompting amounted to human creative process and therefore it was covered but ... how? Any lawyers around that can ELI5 it for us? Maybe links to a lawyer somewhere who did?
AI has all the IP rights of a pen, pencil, chalk, or crayon.
>The Office concludes that, given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.
https://www.copyright.gov/ai/Copyright-and-Artificial-Intell...
While as a species our key strength has been our intelligence and it’s been core to our identity, and computing has slowly over decades infringed on this forcing us to rewrite what it is to be human, I understand the defensive view.
I also see LLMs and other AI systems spit out complete nonsense that’s truly obvious to most people. But that doesn’t make any of these systems, in my opinion, incapable of creating or bridging novel new ideas that I would call far from obvious had we substituted a human in place of it. I didn’t look at the patents in question, plenty of obvious patents make it through anymore, so that could be the case here, but I believe AI isn’t far away if not already there of creating truly patentable inventions if someone were to push it.
Now if the invention also includes some real world work, or if the AI took a huge amount of tokens/money to reach the conclusion, ok. But otherwise an AI coming up with the idea at low cost should invalidate a patent of that idea (the AI not being trained on the patent of course.)