Humans Can Patent AI-Generated Creations. Ownership Is Less Clear

The US Patent and Trademark Office’s recent guidance on inventorship for AI-assisted inventions partially resolves one of the thornier questions in modern intellectual property law: who can patent AI-generated inventions. The answer is now humans, provided that they “significantly contribute” to the output and “recognize and appreciate” the invention.

But the new policy merely clarifies how much AI involvement is permissible under existing inventorship rules, leaving open some important questions.

AI systems are increasingly capable of “inventing” valuable products that would be patentable if they’d been developed by humans. But courts have uniformly rejected attempts to patent these inventions on behalf of AI systems. And under longstanding inventorship rules, humans couldn’t patent these products either. The USPTO has been warned that unless humans are able to own AI-generated properties, innovation will likely be stunted and undermine economic growth.

The guidance incorporates these comments and opens a path for human ownership of AI-generated properties. Despite emphasizing that the patent system “is designed to encourage human ingenuity,” and that AI systems cannot be listed as inventors on patent applications, it concedes that AI can be used like any other tool to create an invention—so long as a human sits metaphorically in the driver’s seat.

To ensure uniform standards across different patent types, the guidance covers plant, design, and utility patents equally. It stresses that each named inventor on a patent application must have “significantly contribute[d]” to the “conception of the invention.”

Each must be responsible for the “definite and permanent idea of the complete and operative invention as it is thereafter applied in practice.” And at least one of the applicants must recognize and appreciate the invention for what it is. “Unrecognized accidental creation” will not suffice.

There is no change to a party’s duty of disclosure when filing a patent application. But patent examiners are now empowered to request additional material from applicants if there are grounds to believe the applicants did not significantly contribute to the invention. And examiners have a series of factors to consider when determining whether a natural person’s contribution is “significant.” These boil down to:

• A natural person isn’t an inventor if they recognize a problem and present it to the AI system to solve. The person must significantly contribute by showing how they constructed a prompt to guide the AI to the result.
• A natural person can’t reduce an invention to practice simply by recognizing and appreciating the AI’s output, especially if the output is obvious to a person of ordinary skill in the art. But if the person makes a significant contribution to the output, or experiments with it successfully, they may demonstrate significant contribution to reduction to practice.
• A natural person need not be present for every step of conception so long as the person develops an essential component from which the claimed invention is derived.
• Merely owning or controlling an AI system does not make a person an inventor. The person must significantly contribute to the conception of the invention.

Hits and Misses

The guidance is a welcome step in the right direction. By acknowledging humans can use tools to capture otherwise un-ownable properties, it builds on centuries of legal tradition. For instance, under the ancient law of capture, “fugitive” resources like wild animals or oil are owned by the first person to capture them. The policy reintroduces this concept for a new era, implicitly recognizing humans can draw ideas out of AI systems the same way they draw fish out of the sea or petroleum out of the earth.

The directive actually grapples with the law of capture’s central weakness: its perverse incentive to overharvest resources. By requiring an inventor recognize and appreciate the claimed invention, and actually develop the essential components of the prompt that triggers the invention, it thwarts trolls who might otherwise “farm” or “mine” inventions promiscuously.

Lawmakers should also embrace the ancient doctrine of accession, which allows indivisible property to be allocated between multiple owners according to the value of their relative interests. Accession would vest the patent for AI-generated inventions in the human with the leading interest in the invention, regardless of inventorship. But the guidance stops short of this outcome.

More clarity is needed in potential scenarios where both an AI developer and an AI user could credibly claim to be inventors of the same product. If an AI developer developed an AI system for the specific purpose of generating novel semiconductor designs, and a user leveraged it against his own knowledge of semiconductors to invent a new chip, inventorship would not be clear under this new guidance.

Owners, Creators, Users

With USPTO examiners now able to request proof of significant contribution from inventors, aspiring AI-assisted inventors must ensure they thoroughly understand the subject matter of the invention they seek to patent. Generic prompts like “make software source code that can interpret and process big data,” won’t cut it.

On the other hand, AI owners can’t simply assume that they own everything their models create, regardless of whether they developed it or not. The development of this field of law, like the development of AI itself, still has a long way to go.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law and Bloomberg Tax, or its owners.

Author Information

Arjun Padmanabhan is an associate attorney at Cole Schotz with focus on commercial and intellectual property litigation.

Write for Us: Author Guidelines