The landscape of intellectual property and invention is facing unprecedented upheaval, with artificial intelligence promising to both liberate and potentially overtake human inventors.
But what may seem like a minefield of IP questions might actually be easier to navigate than it first appears. The trick will be ensuring appropriate legal protections through diligence and creativity in patent filing.
Over the last several years, as AI and large language models have become more widely accessible and available, there has been an increase in concern about the integrity of the patent process. If AI is capable of writing entire novels and screenplays, what’s to stop it from parlaying such creativity to invention, coming up with the next miracle pill or revolutionary bit of tech?
It’s a question that’s been nagging at both patent offices globally and inventors themselves, who have a pretty good understanding of the potential of AI to transform their work.
AI has already shown itself to be a boon for inventors, doing things like offering them the ability to perform multiple complex experiments at once, or pulling together new molecules that humans had not yet thought of.
In the world of software development in particular, AI has the ability to do things like write code incredibly quickly, solving issues that may take inventors months to even discover. Tools like GitHub CoPilot, Anthropic’s Claude and Cursor have been getting a lot of attention for their ability to write software — and because that skill is rapidly improving.
We know that legally, patents are only going to be granted to human inventors, but what about inventions that are assisted in their development by AI? What happens then?
The U.S. Patent and Trademark Office’s February guidance maintains that if AI is used to assist in developing an invention, only humans who have made a significant contribution to the development of that invention can be named as inventors on a patent application for the invention. It’s a decision that’s largely in line with other nations — the German Federal Court of Justice ruled in an appeal brought by Stephen L. Thaler in June that AI-generated inventions can be patented in Germany, but a human inventor must be named in the patent application.[5]
That is, it should be noted, the same standard by which patent inventorship determinations are already determined for all inventions, whether or not AI played a role in its creation.
A lab assistant, for example, may not be named as an inventor on technology just because they did some of the legwork on it. Maybe they wrote code, ran experiments to test the inventor’s theories or performed other work to help the inventor. But those activities alone do not meet the standard of a significant contribution.
There is a standard of creative and developmental input that’s required to be considered a co-inventor. An inventor must be a driving force behind the invention — the one who requested that particular code be written or that certain experiments be performed.
It’s the same reasoning with AI: While the AI may assist with the ultimate development of an invention, it must still be prompted to find a specific answer or run a specific simulation. The work being done by AI is generally a distillation of what would ordinarily be laboratory practices.
All of which to say, for businesses and individuals trying to patent something partially developed with assistance from AI — whether it’s software that’s been coded by AI or a drug that’s been discovered as a result of the use of AI — recordkeeping and diligent documentation is of paramount importance when seeking patent protection.
A successful application will need to document human contributions to the invention. For example, if AI was used in the development of a new invention, the human inventor’s role in creating, training and utilizing the AI model should be clearly documented; while this isn’t a current requirement of the USPTO, in the early days of any new technology, it’s best to document its use in case it’s required later. This documentation should demonstrate how the human inventor’s input was essential to the development of the invention.
For example, if any particular training data was used to train an AI model, then it is advisable to disclose any information about such training data that is necessary to enable the use of such training data. If the training data is publicly available, then it may be sufficient to refer to it by name.
If, however, the training data was created or customized for use in the invention, then it may be helpful or necessary to disclose additional information about the training data. If human skill or ingenuity was used to create or modify the training data, then it may be helpful to disclose the role that humans played in the preparation of the training data to justify listing certain people as inventors.
If the invention performs a useful function using a machine learning model, it may be useful to conduct experiments that demonstrate the ability of the model to perform that function reliably, and to document such experiments and their results in the patent application.
Otherwise, the patent application might be rejected for lack of enablement, or the scope of the claims might be limited to the scope of the specific uses of the model that are clearly proven in the patent application. In this way, patent applications for AI-related inventions may resemble traditional patent applications for chemical and biological applications in their inclusion of experimental methods and results. Describing such experiments may also provide further support for the assertion that specific people qualify as inventors.
A successful patent application will also need to clarify how AI is used to implement the invention, especially if AI is required for recreating the invention. Was it a tool for assisting in the coding process, or did it generate substantial parts of the code autonomously? The more autonomous AI’s role, the more critical it becomes to highlight the human aspects of the invention.
The need to disclose the use of AI in the inventive process may depend on the relationship of AI to the claims in the patent application and the scope of protection being sought. For example, if AI was used to create a design for a product — e.g., a chemical or a circuit — but such a product can be sufficiently enabled using a traditional description, then it might not be necessary to disclose the use of AI in the inventive process to satisfy the enablement requirement.
However, the process by which the invention was made might still be useful to describe in order to demonstrate that there was a significant human contribution to establish human inventorship.
Even if it is not strictly necessary in a particular case to describe the use of AI in order to satisfy the enablement requirement, it still might be useful to disclose how AI was used to make the invention if doing so enables a broader claim scope. For example, if the patent’s claims hold a class of chemical structure, and the use of AI enables a wide range of molecules having such a structure to be discovered, then disclosing the use of AI in the patent’s specification might justify affording the claim with broader scope than if AI were not disclosed.
If the invention uses AI in its operation, such as antivirus software that uses a machine learning model to detect viruses and to continuously update that model based on new information, then the patent application will need to describe the use of AI in order to enable the claims.
There is a degree to which inventors need to be careful of overreliance on AI; given AI’s current capabilities, it seems highly likely that elements of the process that are genuinely discovered by AI will be deemed to be obvious, and thus threaten the patentability of the invention.
The same standards of invention apply, whether AI has been used or not, so ensuring the patent office has no confusion over both human and AI contributions is key.
To be clear, anthropomorphizing AI as a contributor is the wrong tack — AI is a tool, just like any other computing system. Its power makes it an asset, but as in the early days of software development, as the landscape builds, inventors need to be more descriptive rather than less in order to ensure that patent officers understand, and also that appropriate patent protection is afforded.
This means that more than likely, filing a patent that involves AI right now will be more expensive, as the diligence and time needed is likely greater than a mechanical invention. Though, coincidentally, inventors looking for assistance in creating their patent application may be able to turn to AI to help.
As technology continues to advance, staying informed and adaptable will be key to securing IP rights in the age of AI.
This piece originally published in Law360.
The post Navigating The Minefield Of Patenting AI-Generated Inventions first appeared on Blueshift IP.