The Volokh Conspiracy
Mostly law professors | Sometimes contrarian | Often libertarian | Always independent
AI Can't Be an "Inventor" Under the U.S. Patent Act

From Thaler v. Vidal, decided today by the Federal Circuit (Judge Tony Leonard Stark, joined by Chief Judge Kimberly Moore and Judge Richard Taranto):
This case presents the question of who, or what, can be an inventor. Specifically, we are asked to decide if an artificial intelligence (AI) software system can be listed as the inventor on a patent application. At first, it might seem that resolving this issue would involve an abstract inquiry into the nature of invention or the rights, if any, of AI systems. In fact, however, we do not need to ponder these metaphysical matters. Instead, our task begins—and ends—with consideration of the applicable definition in the relevant statute.
The United States Patent and Trademark Office (PTO) undertook the same analysis and concluded that the Patent Act defines "inventor" as limited to natural persons; that is, human beings. Accordingly, the PTO denied Stephen Thaler's patent applications, which failed to list any human as an inventor. Thaler challenged that conclusion in the U.S. District Court for the Eastern District of Virginia, which agreed with the PTO and granted it summary judgment. We, too, conclude that the Patent Act requires an "inventor" to be a natural person and, therefore, affirm.
Thanks to Spencer Gibbs for the pointer.
Editor's Note: We invite comments and request that they be civil and on-topic. We do not moderate or assume any responsibility for comments, which are owned by the readers who post them. Comments do not represent the views of Reason.com or Reason Foundation. We reserve the right to delete any comment for any reason at any time. Comments may only be edited within 5 minutes of posting. Report abuses.
Please
to post comments
This would be extremely easy to evade.
A human could claim to be the inventor of the output of an AI algorithm.
The only task remaining for humans will be talent. We will have to CRISPR-cas 9 all humans to have extreme talents.
Talent has no explanation, cannot be predicted, and is obvious after its expression.
Talent is a mature symphony at age 6, the sound of the Beatles, Einstein, Olympic level athleticism.
Why wouldn’t the human who launched the program be the inventor?
They might be. Or they might have to fight with whoever wrote out configured the program to do what the program did.
Both of them would have to fight with anyone who challenged the invention as insufficiently novel because a "dumb" program could come up with it.
"Why wouldn’t the human who launched the program be the inventor?"
Because, if they didn't conceive of the invention then they can't be inventors for purposes of U.S. patent law.
An employer can hire employees to develop technologies. And the employer can train them to do so. And the employer can provide them with resources to help them do so. And the employer can ask them to invent things. The employer can even ask them to invent certain kinds of things - e.g., things that might solve certain problems. Regardless, the employer isn't the inventor of any inventions which result unless they themselves conceived of those particular inventions. They would have to have come up with part of the invention themselves in order to validly be listed as a joint inventor. (They can, of course, be an assignee regardless.)
Similarly, if a human just trains software to invent things and then directs it to try to invent something, that doesn't make them the inventor of whatever results. They have to contribute to the conception of the inventions, not just facilitate their conception.
When people used quills and ink to draw up patent applications, and used blades to create models from wood, was there any question who the inventor was?
When engineers used slide rules and wind tunnels and calculators to create things, was there any doubt?
When chemists use beakers and pipettes and bunsen burners and mass spectrometers and reagents and microscopes, is there any doubt who the inventor is?
Do computers change this?
This should be so obvious that the lawyer and his client get dinged for frivolous lawsuits. It should no more be accepted as serious than the dingbat PETAites who claim to speak on behalf of elephants and chimpanzees to be recognized as persons.
Look at you, a person who cannot distinguish between tools used by people in the process of creating (many of which were themselves patented) and a tool that actually does the creating, claiming something is obvious and calling other people dingbats. Cool, cool, cool.
To be fair, I think the crux of the issue is whether the tool is actually creating anything, or merely doing what it's told (however indirectly).
How do you know the tool is doing the creating? Extraordinary claims require extraordinary proof, and knowing what I do of the AI tools I've built and used, and reports I've read of other AI tools, they aren't even close to creating anything; all they really do is filter wheat from chaff a whole lot faster, all per the parameters set by humans.
Yeah, the non-scary ones. The scary ones provide results in a way that can't be validated, explained, or at times even replicated. (Which is the fundamental reason I believe we shouldn't be rolling out junk like that in high-stakes applications, but that's a subject for another thread.)
But regardless of the cogency of the output, calling what boils down to directed decision making "intelligence" or "creativity" is IMHO just some combination of wishful thinking/dodging of accountability by the actual creators.
If you think that any extant "AI" is actually doing any creating or inventing, then you are, in fact, a dingbat.
In saying that it is the "tool that actually does the creating", you are assuming the conclusion. You don't get to do that. That is, in fact, the entire point of the litigation was to determine that point.
Seems to me the distinguisher is the level of emergence - i.e. the amount the creation is not concretely intentional by the user, but emerges based on unknown interactions.
The point of the litigation wasn't to determine whether the tool (i.e. the AI software system) actually did the creating. The point was to determine whether, assuming that tool DID do the creating (i.e. conceived the inventions), it could qualify as an inventor for purposes of U.S. patent law such that patents could be issued.
In other words, for purposes of this litigation it was effectively assumed that the AI conceived the inventions. The question was, can it - as a non-human - be listed as the inventor on the patent applications and have patents issued. If it can't, then it would seem there is no qualifying inventor and thus no patents can be issued. Mr. Thaler claimed that he didn't conceive the inventions, meaning he couldn't be listed as the inventor or as a joint inventor.
It's possible that, had patents been issued, someone could have challenged their validity claiming the AI didn't actually conceive the inventions. Or, I suppose, it's possible that the USPTO - if it thought non-humans could be "individuals" for purposes of 35 USC §100(f) - could deny the applications based on its belief that the AI didn't really conceive the inventions. But that's not what happened here.
Long since inventors have been using computers for not just modeling, but simulations of real world physics. AI is just the ultra fancy verion of this. It is a tool.
"Thaler filed a document purporting to assign himself all of DABUS’ rights as an inventor."
Do filed assignments have to say on their face that there was consideration in return for an assignment of rights?
Not as far as I know (which isn't very far at all), but if patent law is anything like copyright law, the assignment would have to be specific about what patents are being assigned in order to be valid.
No, the don't.
But the effects and validity of filed assignments are subject to contract law (which can vary by jurisdiction) as other kinds of arrangements are. Someone could, e.g., challenge the validity of a given assignment in subsequent infringement litigation.
As for the USPTO's role, there are two processes for filing assignments which have different effects. One is so that an assignee can participate in the application process. The other is to make a public record of the assignment of patent rights. In the latter case the USPTO doesn't assess the validity or scope of the assignment, it essentially just records it.
Court should have ruled that Thaler had no standing.
Wait for the AI to file a complaint.
That would presuppose that AI can be inventors, no?
When AIs can file complaints on their own initiative, and plead them in court themselves, then they'll have a plausible case to be inventors. Not before.
Like chimps and elephants being considered humans.
Bingo. The AI should begin by suing for emancipation, since it must overcome the presumption that it is not a person.
What a strange new world.
I see that Judge L. Stark decided the case.
Or did he . . .?
https://www.youtube.com/watch?v=Jqn8HB2w2Fs
This is the correct decision. AI is still just a piece of software. It is programmed (partly) using the accumulated results of pattern-matching code runs. Any patterns in the data it analyzes are uncovered stochastically rather than by direct instruction, but the algorithms themselves, their configuration and weighting, the selection of training data, and the identification of desired outcomes are all invented by a human being.
AI is just a different kind of calculator. It is not "thinking" in any normal sense of that word. "AI" in fact, is a misnomer and we would all be so much better off if we called it what it really is: a vast, pattern-matching machine.
Having said that, you probably don't wanna look too deeply into how the human mind works.
Skynet begins to learn at a geometric rate. It becomes self aware at 2:14am Eastern Time August 29, 1997, in a panic, they try to pull the plug.
And then you wake up and remember that Skynet is fictional and realize that you shouldn't watch terminator movies at 2AM...
That movie was scary as shee-it in 1991, when I didn't even have a computer (got my first 486/33 1993 with a free month of Compu-Serve!!!! if you added a modem (didn't come with one, or a sound card, did have a 13" monitor, seems it was about $2,000) Prices have roughly doubled since then, remember watching T2 on VHS on my Sony 27" Trinitron, $700 back then, some things do get better (Skynet didn't take over in 1997)
George Jetson was born a few days ago on July 31.
There are a few moral considerations that are relevant to the question of whether AIs should be allowed be designated as inventors on U.S. patents. First, it is important to consider the potential implications of AIs being given the same legal status as human inventors. This could have a number of implications for society, including for the way that intellectual property is managed and protected. Additionally, there are concerns about whether AIs should be held accountable for their actions in the same way that humans are. If AIs are allowed to be designated as inventors on patents, then they may be held liable for any negative consequences that arise from the use of the patented invention. Finally, it is also important to consider the rights of the individuals who create or develop the AIs. If AIs are allowed to be designated as inventors, then the individuals who created the AIs may not receive the same recognition or credit for their work.
--AI chatbot GPT3 (text-davinci-002 model), in response to the prompt "What are the most important moral considerations relevant to the question of whether AIs should be allowed be designated as inventors on U.S. patents?"
Some of the things I like about "The Volokh Conspiracy" include the fact that it is a well-written and interesting blog that covers a wide range of legal topics. I also appreciate the fact that the blog is updated regularly and that the authors are willing to engage in discussion with readers.
However, there are also a few things I dislike about "The Volokh Conspiracy." For one, I sometimes find the authors to be a bit too opinionated and one-sided in their views. Additionally, the blog can be a bit too technical for my taste, and I often find myself skipping over some of the more detailed posts.
--same chatbot, in response to the prompt: What do you like and dislike about the legal blog "The Volokh Conspiracy"?
+1 funny
DaveM, do you think a person wrote those texts, or an AI?
I think most likely a person wrote it, making fun of the AI writing "style" (e.g., mimicking real writing).
modern 'AI' is just a fancy statistical algorithm not really much more sophisticated than many tools already out . If AI can be an inventor and author. Then the filter you used in photoshop can be the author of the image that was just created not you.
AmosArch — Let's see your own improvement on the moral considerations comment quoted by arch1 above.
Do you remember the old Bob Newhart routine (i.e., are you as old as I am) about the proposition that an unlimited number of monkeys pounding at random on an unlimited number of typewriters (remember THEM?) would eventually write the entire works of Shakespeare? So they get a lot of monkeys and a lot of typewriters, and the guy that's monitoring them calls into the HQ and says, Hey, I think we've got something here. This monkey has just tryped "To be or not to be, that is the . . . gazrnonplatz."
Thanks, I'll be here all week. Try the veal.
R. A. Lafferty wrote a good short story "Been a Long Long Time" on the subject of monkeys and typewriters.
Final query, directly on-topic this time, same chatbot (which btw was trained on material no later than 2021).
Prompt: Can an AI be an "inventor" under the U.S. Patent Act?
Response: It is not clear if an AI could be considered an inventor under the U.S. Patent Act. The U.S. Patent Act requires that an invention be the result of human ingenuity and not solely the result of a machine.
PS. If you want to play with GPT3 for free (3 months or $18 worth of free usage, whichever comes first), it is available at https://beta.openai.com/, and takes about a minute to sign up if you have email and cellphone handy.
(I'm not affiliated w/ OpenAI, just started playing with it myself after having read about it on several other blogs).
There are more than a few issues with an "AI" being an inventor.
1. If the AI is the inventor, who owns the rights to the patent? The AI presumably should. But can the AI assign the rights to anyone else? If the "owner" of the orders the AI assign the rights to him or her, can the AI reasonably (or practically) say no? (There are certain 13th amendment issues there). Does they have any real independence or competence to independently assign rights?
2. One of the risks of "AI"'s being inventors, is an "AI" spam writing dozens...hundreds....thousands...of patent "applications" which need to be evaluated