AI Just Designed New Life. Patent Law Has a Problem.
I recently watched a DW discussion about an experiment that I suspect we will be talking about for a long time.
Researchers at Stanford University and the Arc Institute trained an AI model called Evo on genomic data and used it to generate entirely new viral genomes.
The numbers are worth considering.
The AI generated hundreds of thousands of designs. Researchers selected 285 of them to physically create and test in a laboratory.
Sixteen worked.
They were functional bacteriophages, viruses that infect bacteria, capable of infecting their targets and reproducing. They had not previously existed in nature.
To be clear, these were not viruses designed to infect humans. There are also very good reasons for conducting this kind of research. Bacteriophages have potentially important applications in areas such as antibiotic resistance, where we badly need new ways of dealing with bacteria that no longer respond to conventional antibiotics.
Unsurprisingly, much of the discussion around the research has focused on biosecurity.
As someone who works in intellectual property, however, I found myself thinking about a different problem.
Who invented them?
Can AI be an inventor?
Under the approach currently taken by most major patent systems, an AI system cannot itself be named as an inventor. Inventorship remains tied to a human being.
We have already tested this proposition through the DABUS cases.
DABUS was an AI system identified as the inventor in patent applications filed across multiple jurisdictions. Courts and patent offices were subsequently forced to confront a question that had previously belonged largely to academic debate: can a machine legally be an inventor?
Broadly, the answer has been no.
But experiments such as Evo make me wonder whether we have been concentrating on the easier question.
The difficult question may no longer be whether AI can be an inventor.
It may be determining what the human actually invented.
Where does an AI-generated invention actually happen?
Consider what happened with Evo.
The AI generated the genomic sequences. Humans selected a small number from the enormous number of possibilities it produced. Humans then physically created those sequences and tested them.
Some worked.
Somewhere in that process, machine-generated information became a functioning biological invention.
Where exactly did the invention occur?
Was it when the AI generated the sequence?
Was it when a researcher recognised that a particular sequence was worth testing?
Was it when the sequence was physically created?
Was it when researchers discovered that it actually worked?
Or was the invention the product of several human contributions surrounding an idea whose underlying structure originated with the AI?
These are no longer particularly abstract questions.
They go directly to inventorship, ownership and ultimately whether a patent is valid.
The human contribution may become the real patent question
The traditional debate about AI and patent law tends to focus on the machine.
Can AI invent?
Can AI own a patent?
Can AI be listed as an inventor?
I think the more consequential question may eventually concern the human standing beside it.
What did that person actually contribute?
If a scientist uses AI as a sophisticated tool while conceiving and directing the invention, the answer may be relatively straightforward.
But imagine progressively removing the human contribution.
The AI generates 100 possible compounds and the scientist chooses one.
Then 10,000.
Then a million.
Another AI ranks them.
An automated laboratory synthesises the most promising candidates.
Robotic equipment tests them.
The experimental results return to the model, which generates the next generation of candidates.
The process repeats continuously.
At what point is the human still doing the inventing?
That is where I think the next phase of the AI inventorship debate becomes considerably more difficult.
From AI-generated inventions to invention at machine scale
There is another consequence that receives much less attention.
AI doesn’t merely change who or what contributes to an invention.
It potentially changes the volume at which invention occurs.
Human R&D has historically been constrained by human capacity. Researchers need time to formulate hypotheses, design experiments, analyse failures and develop the next idea.
AI can radically compress parts of that process.
Combine generative models with automated laboratories, robotics and increasingly capable scientific AI systems and it becomes possible to imagine an R&D environment producing and testing potential inventions continuously.
That would give us something quite different from the occasional AI-generated invention.
It would give us invention at machine scale.
And that creates an entirely different problem for companies and their IP advisers.
When invention becomes abundant, what becomes scarce?
Imagine that an AI-enabled R&D operation can generate 100,000 potentially useful inventions.
Nobody is filing 100,000 patent applications.
Someone still has to determine which inventions matter.
Someone needs to know how each invention was generated.
Someone needs to establish the human contribution.
Someone needs to determine who owns it.
Someone needs to search the prior art and assess novelty and inventive step.
Someone needs to decide whether patent protection, trade secret protection or simply doing nothing makes the most commercial sense.
And someone needs to make those decisions quickly enough for the business to retain an advantage.
This is why I think AI could change something more fundamental than the mechanics of inventorship.
The bottleneck may move from invention to invention capture.
What is invention capture?
I use “invention capture” to describe the process through which an organisation identifies potentially valuable inventions, records how they were created, establishes the relevant human contribution and ownership, evaluates their patentability and commercial importance, and determines which should enter the IP portfolio.
For AI-generated inventions, that process could eventually look something like this:
Generation → Provenance → Human Contribution → Validation → Novelty → Patentability → Ownership → Filing Strategy
Each stage matters.
Provenance matters because a company may need to establish how an invention arose.
Human contribution matters because patent systems continue to revolve around human inventorship.
Validation matters because an AI can generate enormous numbers of theoretically interesting outputs that may not work.
Novelty and patentability matter because generating something new to the model does not necessarily mean generating something new to the world.
Ownership matters because increasingly complicated combinations of employees, external researchers, AI providers, datasets and automated systems could sit behind a single invention.
And filing strategy matters because even the largest companies have finite patent budgets.
The more inventions AI produces, the more important these filters become.
The competitive advantage may change
For much of industrial history, the difficult part was producing the invention.
That assumption sits underneath much of the modern IP system.
But what happens when generating potential inventions becomes extraordinarily cheap?
The valuable capability may shift.
Companies may compete not simply on who can invent the fastest, but on who can recognise which machine-generated inventions actually matter.
Who can validate them.
Who can establish ownership.
Who can document the human contribution.
Who can identify the defensible ones.
And ultimately, who can capture and protect them before competitors do.
That means IP strategy potentially moves much further upstream.
It becomes part of the architecture of AI-enabled R&D itself rather than something that happens after researchers arrive at the patent attorney’s office with an invention.
That, to me, is where this gets really interesting.
We have spent the last few years asking what happens when AI creates.
We are going to have to start figuring out what happens when AI invents.
And if invention itself eventually happens at machine scale, the companies that win may not necessarily be those capable of generating the most ideas.
They may be the ones capable of identifying, capturing and protecting the right ones.
Author
Visharad Venugopal Mannadiar
Founder of Brandguard
Certified Intellectual Property Valuer (AMAVI)
About the author
Visharad is a certified IP valuer and intellectual property advisor focused on the intersection of artificial intelligence, intellectual property, and strategic defensibility.