US patents boss cannot stress enough that inventors must be human, not AI The US Patent and Trademark Office this week repeated loud and clear it will only accept patent applications that list actual real humans as the inventor and not AI. That said, officials signaled they are willing to consider inventions developed with the assistance of artificial intelligence, though the human inventors listed …
It can talk about upholding whatever it wants, it has done nothing but undermine the very idea of patents since the beginning of the 20th century.
By never verifying prior art and letting the judicial system take care of issues, the USPTO is just a rubber stamp for multi-billion-dollar behemoths and serves no longer the ideal of the individual inventor getting recognized for an idea.
On top of that, the USPTo has long forgotten the idea that a patent could only be granted to an idea that was functionally described in full, thus, when the patent expired, others coulld pick up the idea and make use of it. Today, patents are granted on simple ideas (round corners, anyone?), and their functionality is no longer a subject of examination.
Experience has already shown that AI is a big liar when when it comes to being consistent. Several researchers who haven't published recently have asked AI to give them a biography, which includes the fact that they have dies. One going as far as stating the graveyard where the very much alive reasearcher is buried.
If the USPTO refuses a patent based on the major part having been the inventor, how will that AI be trusted in court?
From https://itsartlaw.org/2015/12/23/state-of-the-art-introduction-to-patent-law/
The second requirement that needs to be satisfied to obtain a patent is nonobviousness. Nonobviousness is a difficult concept in patent law that is a lot more confusing than it might appear. Simply put, nonobviousness is the requirement that someone would not be able to easily create the invention merely by looking at what has already been invented.
If a purely mechanical process is used to derive an otherwise patentable idea, even if it involves random number generation of setting initial weights for a deep learning model, from mcurrently available information, then it is obvious and cannot be patented.
As soon as we get general AI that can invent anything, then the idea and requirements of Patents are dead.
Which leads us to the "Clayton Test" (thank you JRC):
If, when given a prompt that expresses the problem to be solved, an AI proposes a solution that is substantially similar to the one described in the patent, then the patent is automatically invalid on the grounds of lack or original thought.
That would be a great test to use before a patent is awarded, or to be used if a patent is disputed.
>then it is obvious and cannot be patented.
So a hull design produced by CFD can't be patented because it's a purely mechanical process?
I can see some algorithms being trickier. I can patent an algorithm for detecting corners in an image (or I can't cos someone already did) so why can't I patent a trained model to do the same?
Really, not much has changed: Inventors and joint inventors must be natural persons, and those persons must pass the standard Pannu test, which among other things requires them to have made substantial contributions to the designs.
So companies can not be listed as inventor as they are not natural persons?
"The US Patent and Trademark Office this week repeated loud and clear it will only accept patent applications that list actual real humans as the inventor and not AI."
Considering they patent anything not patented earlier (and sometimes even that) as long as someone pays for it, this is absolute hypocrisy, they don't care an iota.
Pure lipstick on a pig.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
