It's not so impressive when you realise that South Africa is a non-examining patent issuer - patents are issued as long as all the forms are properly filled out.
A markov-chain based word-generator could just have easily gotten the patent issued, because the patent office does not examine the patent application for patentability, it only checks that the correct procedure is followed.
I don't know about you, but for me this feels like the gates of hell opening for patent trolls. AI can produce thousands of patents in a single day. As always patent trolls make no money from patents, but from the bureaucratic process. So a schema like this that allows virtually unlimited patents to be created will increase even more trolls misbehavior
Then patent offices will generate adversarial AI review examiners. Together the AI trolls and AI examiners might for the first time produce true innovation out of the patent-giving process
More likely to be the opposite. It would be very easy to explore prior art through tons of databases. On the other hand, AI has been touted for years to be on the verge of revolutionizing case law and this has yet to happen.
Considering the quality of current examinations in the approval process, I don't see how AI can do any worse.
Basically now instead of doing a proper technical analysis and search over already approved patents, seems they just approve and "let the courts handle it".
I understand a proper type of technical patent analysis in the approval process is extremely hard and resource consuming, but the way it's handled now is killing anyone not having a legal firm on their speed dial and cash to burn.
This can be done also without listing AI as the inventor. What was stopping them yesterday from generating random patents and is not stopping them today with this?
Nothing at all. The friction is that it costs a decent chunk of money to submit patents. It’s not worthwhile to submit gibberish ones that will never be accepted, or never hold up in court if accepted but challenged. But there is probably some potential market for an innovative AI able to produce patents worth more than the fees they incur, for legitimate or trolling purposes.
A patent fee should keep this behaviour to a minimum...
Patents already have quite a big fee, but perhaps it should be made larger to prevent this kind of abuse. If the large fee prevents inventors, then perhaps some scheme along the lines of 'The fee is $50 for the first 5 inventions, then $100k per patent after that" might be suitable.
I hate to be pessimistic, but I'm sure this would lead to situations where dummy companies are set up for registering the patent, and are then acquired by the patent troll, with all those assets.
> In IP geek circles, Manfred is legendary; he's the guy who patented the business practice of moving your e-business somewhere with a slack intellectual property regime in order to evade licensing encumbrances. He's the guy who patented using genetic algorithms to patent everything they can permutate from an initial description of a problem domain – not just a better mousetrap, but the set of all possible better mousetraps.
It is a great way to control the AI. Patents should be mandatory for them. If they file for "Mass extinction of human species devices" we will reveal their plans just in time!
Courts in and of themselves have no hard power. Law enforcement agents and other officers of the peace are the ones in charge of delivering on courts judgements.
You could use AI to produce thousands of patents in a single day and put your name as the inventor - I'm not sure having the AI system recorded as inventor adds much for patent trolls.
If AI just churns out a whole bunch of stuff with no known useful application that can be demonstrated in physical reality, it won't be able to patent that.
"[A] patent is not a hunting license. It is not a reward for the search, but compensation for its successful conclusion" - Brenner v. Manson [1]
In the above case, the inventor discovered a whole class of molecules. The Supreme Court said that you can't patent that because you have to show that each thing you patent can be used for something useful. Basically, the inventor couldn't just claim the whole class of molecules to keep anyone from using them and then figure out what they're good for later.
Isn't that effectively what this outcome does? The AI is listed as the creator, but the patent is held by the person that crated the AI. IANAL, but doesn't that mean the Dr Thaler has the legal rights to the patent?
South Africa has become the first country to award a patent that names an artificial intelligence as its inventor and the AI’s owner as the patent's owner.
Edit - I not claiming this is the correct result. Just looking for clarification on what this means, legally speaking.
No, what I'm saying is that the AI is a tool which its creators use to create the invention, nothing more. It doesn't get listed on the patent in the same way other tools don't.
Yeah, I think the key aspect of the story isn't that they somehow convinced the patent office, but more likely that the patent office in South Africa was the first one they tried that didn't bother to check if the people listed as inventors actually exist.
(Given that we don't have their side of the story and they are only listed as not having rejected it)
> That being said AI patents are an issue, but I don’t think they’re quite there yet.
The 1981 ZX81 Chess program that ran in 1kiB of RAM would have astonished a 19th century observer and would have qualified as "artificial intelligence" from their point of view. What's currently called "AI" is not that much different from that program. Claiming it has somehow any kind of personhood is just nonsensical, and probably a publicity stunt. Why not list SPICE or your favorite finite elements solver in your patent application as a contributor?
> What's currently called "AI" is not that much different from that program.
It's pretty significantly different to that program, both in terms of scope, complexity and output.
I'm going to assume by "person" we mean some kind of intelligence which appears to be self-aware, or that can interact with humans to make it difficult to distinguish between human and machine (what's a human in terms of thinking, if a machine can output the same thoughts?)
We aren't anywhere near there yet, but if you contrast the 1981 chess program's ability to be considered a "person" vs GPT-3's ability to be considered a "person", the second is substantially closer to that end position.
It's significantly different in degree, but not in kind.
Modern ML is vastly more complex, capable, and widely-applicable than the first chess programs, but it's not intelligent, and doesn't have any greater ability to be 'considered a person'.
An aeroplane is a much more complex, capable, and widely-applicable vehicle than a bicycle, but it's no more human or more generally intelligent.
This is the patent application, as far as I can tell. And as others have noted, this is obviously a political stunt—the article makes that clear as well.
Frustratingly, it doesn't seem like there's anything interesting or useful happening in the actual patent. It would not surprise me to find that they had been rejected in part because the idea seems to be either literally impossible to implement or basically uninteresting, depending on your interpretation.
Also, as a non-lawyer, this stunt requires talking out of both sides of one's mouth. They necessarily assert that they have the authority to speak on behalf of the machine listed as the inventor in order to assign ownership of the patent, so why not just assert oneself as the inventor?
No, because people are not objects or expert systems.That’s not even a hard concept. It’s like asking if you can be considered a plant because you drink water and grow taller…it’s not looking at the right details.
“Tools do stuff, humans do stuff, are humans tools?” Is abuse of transitive property in an illogical way.
Of course we're more than just that, but if we agree on the definition of tool to be "an object that can extend an individual's ability to modify features of the surrounding environment" (lifted from wikipedia) then wouldn't you agree that humans are being used on a daily basis exactly this way?
If we agree on the definition of a broom as something that can sweep the floor and say that humans can also sweep the floor, wouldn't you agree that humans are brooms often on a daily basis?
I think that definition of a broom may be too broad for me to be able to exclude anything from it but I'll concede on the definition of tool from wikipedia being similarly also too broad.
I think you’re missing the point if that’s what you’re getting from my posts. The issue isn’t picking the right definition from wikipedia - the issue is not being able to distinguish between an object and a person.
I still don't agree with your premise, people can be used as tools even when I can distinguish it from a person given other aspects (for example I may recognize that people can act autonomously and tools not). My initial question was precisely about semantics though and I think there's the possibility that my initial reasoning may not hold for some, more specific definition of tool.
OK. Someone had to to make the obvious point. At least it's concise.
Patents, as they exist, typically list an inventor, but the owner can be different. Mostly, the inventor is an employee and the owner is the employer. Where this can get hairy is FOSS. If FOSS invents something...
I think there might be other philosophical objections. If a NN "invents" something, perhaps its unpatentable because it is (by some definition) trivial or discovery of a natural law... IE, not a creative work for patent purposes.
Ultimately, patent law and the arguments underpinning it have many avenues for philosophical "attack." But, it's a legal framework, not a philosophical one.
This is silly. You may as well try to credit “calculus” as the inventor of a patent whose key innovation was calculated using a derivative. Some human crafted the problem for the AI, chose the parameters of a good solution, and verified the results. That human or humans are the inventors, not one of the tools they used.
I think the interesting question here is whether AI can be involved in what's non-trivial to a "person having ordinary skill in the art"?
If the point of this stunt is that the owner (operator?) of an AI may not understand how specifically the AI produced a solution ... fine I guess. But if we have black-ish box AIs that let us find good solutions we don't ourselves understand, surely those tools will be available to other practitioners as well.
Does everything that is findable by a common AI architecture become "obvious" b/c a person with ordinary skill can launch the AI which finds it in from a colab notebook?
A company might use it to eliminate any benefits of the actual human inventor, even if it's just the recognition as many job contracts are already pretty bad.
It would be interesting to see it turned the other way around. If a company owns all inventions made by employee event if it's done outside the working hours without company equipment due to the contract, do they own inventions made by AI running on a computer owned by employee?
Brought to you by the same government that had a leader who tried justifying spending public funds on a pool for his personal house as a "fire pool". I love SA, but their oversight overall on things like this is non-existent.
I'm not sure what this means - I thought Inventor meant owner or part owner in some sense unless the ownership has been transferred, presumably Dabus isn't a legal entity - so this is meaningless.
How is the invention of an AI any different from the result of any traditional optimization process?
Example: I searched for aerodinamics genetic algorithms and I found "Genetic algorithms applied to the aerodynamic design of transonic airfoils" https://arc.aiaa.org/doi/abs/10.2514/3.46810 (paywalled but that's not the point.)
Rename genetic algorithms into Multi Generational AI and is the Patent Office going to accept a patent for those airfoils with the AI as the inventor?
It's not so impressive when you realise that South Africa is a non-examining patent issuer - patents are issued as long as all the forms are properly filled out.
https://www.bowmanslaw.com/insights/intellectual-property/so...
A markov-chain based word-generator could just have easily gotten the patent issued, because the patent office does not examine the patent application for patentability, it only checks that the correct procedure is followed.
I don't know about you, but for me this feels like the gates of hell opening for patent trolls. AI can produce thousands of patents in a single day. As always patent trolls make no money from patents, but from the bureaucratic process. So a schema like this that allows virtually unlimited patents to be created will increase even more trolls misbehavior
Then patent offices will generate adversarial AI review examiners. Together the AI trolls and AI examiners might for the first time produce true innovation out of the patent-giving process
After said trolls force the hand of the government to invest in a sufficient adversarial network.
AFAICT, a sufficient AI review examiner would validate all inputs that are reasonably coherent, but not necessarily novel.
More likely to be the opposite. It would be very easy to explore prior art through tons of databases. On the other hand, AI has been touted for years to be on the verge of revolutionizing case law and this has yet to happen.
In my experience with patents, you don't really need something novel as much as you need unique words in your patent, its all about wordsmithing.
You forgot to mention the AI judges, when there is a dispute.
Isn't that just the objective function?
Considering the quality of current examinations in the approval process, I don't see how AI can do any worse. Basically now instead of doing a proper technical analysis and search over already approved patents, seems they just approve and "let the courts handle it".
I understand a proper type of technical patent analysis in the approval process is extremely hard and resource consuming, but the way it's handled now is killing anyone not having a legal firm on their speed dial and cash to burn.
> patent offices will generate adversarial AI review examiners
What if someone patents these examiners first? :)
This can be done also without listing AI as the inventor. What was stopping them yesterday from generating random patents and is not stopping them today with this?
Nothing at all. The friction is that it costs a decent chunk of money to submit patents. It’s not worthwhile to submit gibberish ones that will never be accepted, or never hold up in court if accepted but challenged. But there is probably some potential market for an innovative AI able to produce patents worth more than the fees they incur, for legitimate or trolling purposes.
Of course, then you need to understand what patents you have, which requires just as much work as drafting the patent in the first place :)
A patent fee should keep this behaviour to a minimum...
Patents already have quite a big fee, but perhaps it should be made larger to prevent this kind of abuse. If the large fee prevents inventors, then perhaps some scheme along the lines of 'The fee is $50 for the first 5 inventions, then $100k per patent after that" might be suitable.
I hate to be pessimistic, but I'm sure this would lead to situations where dummy companies are set up for registering the patent, and are then acquired by the patent troll, with all those assets.
Have it like income taxes... Ie., it's per citizen and companies can't be the inventor (or if they do, they have to pay the higher rate).
I'm sure I have read a Charles Stross novel in which the protagonist does this and becomes the wealthiest person on the planet
Accelerando
> In IP geek circles, Manfred is legendary; he's the guy who patented the business practice of moving your e-business somewhere with a slack intellectual property regime in order to evade licensing encumbrances. He's the guy who patented using genetic algorithms to patent everything they can permutate from an initial description of a problem domain – not just a better mousetrap, but the set of all possible better mousetraps.
That's the one! Thanks
> AI can produce thousands of patents in a single day
This is only a problem if the AI is litigious. Surely an AI with true agency would be keep the patent rights to themselves
It is a great way to control the AI. Patents should be mandatory for them. If they file for "Mass extinction of human species devices" we will reveal their plans just in time!
These patents will be enforced only by AI courts and any potential settlement needs to be in "energy points"
Courts in and of themselves have no hard power. Law enforcement agents and other officers of the peace are the ones in charge of delivering on courts judgements.
I, for one, welcome our new Robocop overlords.
If so, seventeen years later, nothing will be patentable.
You could use AI to produce thousands of patents in a single day and put your name as the inventor - I'm not sure having the AI system recorded as inventor adds much for patent trolls.
If AI just churns out a whole bunch of stuff with no known useful application that can be demonstrated in physical reality, it won't be able to patent that.
"[A] patent is not a hunting license. It is not a reward for the search, but compensation for its successful conclusion" - Brenner v. Manson [1]
In the above case, the inventor discovered a whole class of molecules. The Supreme Court said that you can't patent that because you have to show that each thing you patent can be used for something useful. Basically, the inventor couldn't just claim the whole class of molecules to keep anyone from using them and then figure out what they're good for later.
[1] https://en.wikipedia.org/wiki/Brenner_v._Manson
Yeah the rest of the world declined this because the AI
(1) isn’t a legal person
(2) wouldn’t be able to confer ownership and
(3) identify when the patent was being infringed.
This is silly. Great theater, but nothing else.
That being said AI patents are an issue, but I don’t think they’re quite there yet.
We're a long way off from Star Trek's Data / "Measure of a Man" scenario, when the AI can fight for itself.
I would suggest instead of viewing AI as the creator, it should be viewed as a tool, and the person/team who built the AI as the true creators.
Isn't that effectively what this outcome does? The AI is listed as the creator, but the patent is held by the person that crated the AI. IANAL, but doesn't that mean the Dr Thaler has the legal rights to the patent?
South Africa has become the first country to award a patent that names an artificial intelligence as its inventor and the AI’s owner as the patent's owner.
Edit - I not claiming this is the correct result. Just looking for clarification on what this means, legally speaking.
No, what I'm saying is that the AI is a tool which its creators use to create the invention, nothing more. It doesn't get listed on the patent in the same way other tools don't.
Yeah, I think the key aspect of the story isn't that they somehow convinced the patent office, but more likely that the patent office in South Africa was the first one they tried that didn't bother to check if the people listed as inventors actually exist.
(Given that we don't have their side of the story and they are only listed as not having rejected it)
> That being said AI patents are an issue, but I don’t think they’re quite there yet.
The 1981 ZX81 Chess program that ran in 1kiB of RAM would have astonished a 19th century observer and would have qualified as "artificial intelligence" from their point of view. What's currently called "AI" is not that much different from that program. Claiming it has somehow any kind of personhood is just nonsensical, and probably a publicity stunt. Why not list SPICE or your favorite finite elements solver in your patent application as a contributor?
Over time we realize things we thought required intelligence turn out to not.
The contention here is not about intelligence, but personhood.
> What's currently called "AI" is not that much different from that program.
It's pretty significantly different to that program, both in terms of scope, complexity and output.
I'm going to assume by "person" we mean some kind of intelligence which appears to be self-aware, or that can interact with humans to make it difficult to distinguish between human and machine (what's a human in terms of thinking, if a machine can output the same thoughts?)
We aren't anywhere near there yet, but if you contrast the 1981 chess program's ability to be considered a "person" vs GPT-3's ability to be considered a "person", the second is substantially closer to that end position.
It's significantly different in degree, but not in kind.
Modern ML is vastly more complex, capable, and widely-applicable than the first chess programs, but it's not intelligent, and doesn't have any greater ability to be 'considered a person'.
An aeroplane is a much more complex, capable, and widely-applicable vehicle than a bicycle, but it's no more human or more generally intelligent.
https://patentscope.wipo.int/search/en/detail.jsf?docId=WO20...
This is the patent application, as far as I can tell. And as others have noted, this is obviously a political stunt—the article makes that clear as well.
Frustratingly, it doesn't seem like there's anything interesting or useful happening in the actual patent. It would not surprise me to find that they had been rejected in part because the idea seems to be either literally impossible to implement or basically uninteresting, depending on your interpretation.
Also, as a non-lawyer, this stunt requires talking out of both sides of one's mouth. They necessarily assert that they have the authority to speak on behalf of the machine listed as the inventor in order to assign ownership of the patent, so why not just assert oneself as the inventor?
My hammer made a house, I merely own it
Yeah this seems like nothing more than a publicity stunt.
Corporations are also an artificial construct.
They are a legally recognised/defined artificial construct - which is quite important.
comprised of humans, which is also salient.
And other companies and other legal entities...
On the contrary, can a person paid to develop something be considered (partially) a tool?
No, because people are not objects or expert systems.That’s not even a hard concept. It’s like asking if you can be considered a plant because you drink water and grow taller…it’s not looking at the right details.
“Tools do stuff, humans do stuff, are humans tools?” Is abuse of transitive property in an illogical way.
Of course we're more than just that, but if we agree on the definition of tool to be "an object that can extend an individual's ability to modify features of the surrounding environment" (lifted from wikipedia) then wouldn't you agree that humans are being used on a daily basis exactly this way?
If we agree on the definition of a broom as something that can sweep the floor and say that humans can also sweep the floor, wouldn't you agree that humans are brooms often on a daily basis?
I think that definition of a broom may be too broad for me to be able to exclude anything from it but I'll concede on the definition of tool from wikipedia being similarly also too broad.
I think you’re missing the point if that’s what you’re getting from my posts. The issue isn’t picking the right definition from wikipedia - the issue is not being able to distinguish between an object and a person.
I still don't agree with your premise, people can be used as tools even when I can distinguish it from a person given other aspects (for example I may recognize that people can act autonomously and tools not). My initial question was precisely about semantics though and I think there's the possibility that my initial reasoning may not hold for some, more specific definition of tool.
OK. Someone had to to make the obvious point. At least it's concise.
Patents, as they exist, typically list an inventor, but the owner can be different. Mostly, the inventor is an employee and the owner is the employer. Where this can get hairy is FOSS. If FOSS invents something...
I think there might be other philosophical objections. If a NN "invents" something, perhaps its unpatentable because it is (by some definition) trivial or discovery of a natural law... IE, not a creative work for patent purposes.
Ultimately, patent law and the arguments underpinning it have many avenues for philosophical "attack." But, it's a legal framework, not a philosophical one.
Does anyone have a patent on using an AI to generate patents? Because these guys might owe those guys some money!
As a South African I guarantee you that getting something ludicrous verified by a bureaucrat is not remarkable.
This is silly. You may as well try to credit “calculus” as the inventor of a patent whose key innovation was calculated using a derivative. Some human crafted the problem for the AI, chose the parameters of a good solution, and verified the results. That human or humans are the inventors, not one of the tools they used.
The same logic, by some people's standards, would see all patents credited to God then, I suppose. Human exceptionalism?
Well, a lot of kings derived their legitimacy from god so it's not that far off...
I think the interesting question here is whether AI can be involved in what's non-trivial to a "person having ordinary skill in the art"?
If the point of this stunt is that the owner (operator?) of an AI may not understand how specifically the AI produced a solution ... fine I guess. But if we have black-ish box AIs that let us find good solutions we don't ourselves understand, surely those tools will be available to other practitioners as well.
Does everything that is findable by a common AI architecture become "obvious" b/c a person with ordinary skill can launch the AI which finds it in from a colab notebook?
I wonder if there is an advantage to a company in not having a named human registered against their patent.
The AI cannot be dragged into court to testify about the creation of the patent?
...yet?
The AI will spawn a linked subservient instance for that kind of stuff.... it will not take the kind of chances humans take...
A company might use it to eliminate any benefits of the actual human inventor, even if it's just the recognition as many job contracts are already pretty bad.
It would be interesting to see it turned the other way around. If a company owns all inventions made by employee event if it's done outside the working hours without company equipment due to the contract, do they own inventions made by AI running on a computer owned by employee?
Brought to you by the same government that had a leader who tried justifying spending public funds on a pool for his personal house as a "fire pool". I love SA, but their oversight overall on things like this is non-existent.
What does it even mean to have AI as the inventor, is that referring to the name of the file on the disk?
AI software isn't like any legal entity that currently exists. Software can be duplicated instantly, people and companies cannot.
I feel an NFT debate coming on...
I'm not sure what this means - I thought Inventor meant owner or part owner in some sense unless the ownership has been transferred, presumably Dabus isn't a legal entity - so this is meaningless.
How is the invention of an AI any different from the result of any traditional optimization process?
Example: I searched for aerodinamics genetic algorithms and I found "Genetic algorithms applied to the aerodynamic design of transonic airfoils" https://arc.aiaa.org/doi/abs/10.2514/3.46810 (paywalled but that's not the point.)
Rename genetic algorithms into Multi Generational AI and is the Patent Office going to accept a patent for those airfoils with the AI as the inventor?
Considering how far what's currently styled "AI" is from a real GAI, that's got to be a gimmick.
What happens when my code patents itself?
Isn't this the same as issuing a patent listing a microscope as inventor?
clickbait..
In a sense... IE, the professor behind this is obviously making a clickbait-ey point. The article is just reporting on it.
This is not real.