Patent law is falling behind the AI invention revolution
The inventive process has been upended by AI. For patent holders, their advisers, courts and policymakers, this poses challenging questions that demand answers
By Sharaz Gill
In June, Google outlined an ambitious vision. In an interview marking the creation of a programme to roll out AI tools for scientific research, the Head of Google Research Yossi Matias described AI as a collaborative research partner capable of generating hypotheses, proposing experiments and accelerating the pace of scientific innovation.
While that vision remains theoretical in many respects, it is clear AI is moving beyond document drafting and data analysis towards active participation in the research process itself. For the patent community, this immediately raises a familiar question: can an AI system be an inventor?
To my mind, though, the more significant questions arise once AI becomes an accepted part of the inventive process. If an AI system proposes the technical solution that ultimately becomes the invention, what constitutes the human inventor’s inventive contribution? How should inventorship be established when the path to the invention is recorded in a series of prompts and AI-generated responses rather than laboratory notebooks and discussions between researchers? More fundamentally, if AI routinely identifies promising technical solutions that might otherwise have remained undiscovered, does this alter the way in which patent law should assess inventive step?
As AI becomes embedded within research and development, it has the potential to reshape not only how inventions are created, but also the legal assumptions on which the patent system has long relied. The real debate is therefore about whether patent law’s established concepts of inventorship and obviousness remain fit for an era in which human ingenuity is increasingly augmented by machine intelligence.
There is a narrative in David Mitchell’s novel Cloud Atlas that has always stayed with me. Robert Frobisher becomes amanuensis to Vyvyan Ayrs, an ageing composer whose creative powers are fading. At first, the relationship appears clear: one man is the composer; the other assists. But the distinction becomes less stable as Frobisher’s own musical creativity begins to enter the work. The assistant is no longer merely recording another person’s ideas but composing himself.
This literary scenario captures something important about AI-assisted invention. Who may be named as an inventor is largely settled. How human inventorship should be assessed once AI becomes an integral part of the inventive process is not.
Consider a researcher seeking to develop a new pharmaceutical compound. Rather than designing candidate molecules through conventional experimentation, they ask an AI system to identify the compounds that are most likely to inhibit a particular biological pathway. The AI proposes a shortlist of candidates, one of which is selected, synthesised and ultimately proves to be both novel and inventive.
Most patent systems would almost certainly conclude that the AI is not the inventor. But does it necessarily follow that the researcher conceived the invention? Or has the researcher simply recognised the significance of a solution generated by another source?
Patent law has long distinguished between those who contribute to the conception of an invention and those who merely verify, implement or reduce it to practice. AI introduces an unfamiliar dynamic. Unlike laboratory equipment or conventional software, generative AI may contribute at precisely the stage that has traditionally been regarded as the inventive one: identifying a technical solution to a defined problem.
None of this necessarily requires a redefinition of inventorship. Human judgement remains essential. The researcher must define the technical problem, assess the plausibility of the AI’s proposals, reject unsuitable alternatives and decide whether the suggested solution merits further development. These are not necessarily trivial contributions. Indeed, on occasion, they may themselves amount to the exercise of inventive skill.
As AI assumes a greater role in scientific research, though, patent offices and courts may increasingly be required to determine where the human contribution ends and the AI-assisted process begins. That question is likely to become one of evidence rather than law.
On the record
Traditionally, the route to an invention might be reconstructed from laboratory notebooks, internal presentations, emails, draft invention disclosures and the recollection of the researchers involved. These materials usually show experimental results and project milestones, but they rarely capture every intermediate step in the reasoning process.
Prompt-based AI systems leave a more detailed trail. A prompt may show how the technical problem was framed. An AI response may identify candidate solutions. Later prompts may reveal how the researcher challenged the output, refined the constraints, rejected unsuitable pathways or steered the system towards a more promising result.
That record may help to resolve inventorship questions. It may show that the named inventor exercised genuine technical judgement: framing the technical problem, imposing meaningful constraints, rejecting false leads and recognising why a particular pathway would solve the problem. Conversely, it may show that the decisive technical contribution first appeared in an AI-generated response, with the human role limited to selection, verification or routine implementation.
The record may show whether the named human inventor made the kind of contribution that patent law recognises as inventorship. If not, the resulting risk may appear as an inventorship defect, an entitlement dispute, an ownership issue or even, in some jurisdictions, a validity problem.
Inventive step: is it all about the journey?
Once prompts and outputs reveal the route by which an invention was reached, they may invite questions not only about inventorship, but also about inventive step. Inventive step turns on whether the route from the prior art to the claimed invention was one that the skilled person would have taken; and the route is almost always reconstructed after the fact by people who already know where it led. AI will complicate that analysis.
Patent law does not ask whether the actual inventor found the invention difficult. In Europe, inventive step is assessed objectively, by asking whether the claimed invention would have been obvious to the notional skilled person starting from the prior art and faced with the objective technical problem.
Even so, the route to the invention can matter indirectly. If the prior art presents the skilled person with a clear and definite path and following that path would have led to the invention with a reasonable expectation of success, inventive step becomes harder to defend. EPO case law sometimes describes this as a “one-way street”.
English law addresses overlapping issues through the “obvious to try” analysis, particularly where progress is made through routine experimentation and the skilled person would have been motivated to pursue a route with a reasonable expectation of success. The modern statement of that approach is the UK Supreme Court’s decision in Actavis v ICOS, which treats the routineness of the research path as one factor among many rather than as a test in itself.
When AI shows the way…
AI raises the difficult question of what happens when that route is identified not by the skilled person unaided, but by an AI system. Suppose that the prior art is complex and there are many possible directions. A human researcher might not see any clear way forward. An AI system, however, identifies one candidate approach as unusually promising. The researchers follow that path and arrive at the claimed invention. Has AI merely assisted the inventors? Or has it shown that the invention lay on an obvious route once the available information was properly analysed?
Again, the answer is not straightforward. Both the “one-way street” and “obvious to try” analyses assume that the relevant route was apparent to the skilled person at the priority date. They do not normally ask what might have been discovered by a system capable of processing information at a scale and speed beyond ordinary human capacity. An AI-generated recommendation may therefore be evidence of a possible route, but it does not necessarily follow that the route was obvious in the traditional patent law sense.
The anomaly becomes clearer if two research teams arrive at the same invention by different routes. One team proceeds conventionally: reviewing the literature, rejecting possible approaches and eventually reaching the claimed invention after arduous experimentation. The other starts from the same prior art but uses an AI system that identifies the relevant candidate solution as especially promising. The invention is identical. The prior art is identical. The technical problem is identical. What differs is the route by which the invention is reached.
Patent law would normally resist giving legal significance to that difference. Inventive step is not assessed by asking how hard the actual inventor had to work. Yet AI-assisted research creates an uncomfortable truth:
If AI-generated pathways are treated as part of the ordinary toolkit of the skilled person, the first team’s invention could be treated as obvious even though it required considerable human ingenuity.
If AI is ignored, the second team may obtain protection for an outcome that may have required little more than formulating the right query, selecting a suggested pathway and performing confirmatory work.
Neither result is entirely satisfactory. Patent law seeks to avoid rewarding mere perseverance, luck or routine verification. It also seeks to avoid denying protection merely because a particular inventor happened to use a powerful tool.
It is therefore important to consider whether the availability of AI assistance changes the ordinary capabilities of the skilled person. Patent law has always attributed ordinary tools and knowledge to that person. Scientific databases, sequence-searching tools, simulation software, computational modelling and machine learning techniques have all, in appropriate fields, become part of the ordinary research environment.
The skilled person reloaded
AI is the next stage in that development. If researchers in a particular field routinely use AI systems to search literature, generate candidate molecules, model technical systems or propose experimental pathways, it must surely be artificial to exclude those tools from the capabilities of the skilled person. But AI differs from many earlier tools. It may not merely help the skilled person to find, calculate or verify; it may propose hypotheses and identify promising routes that a human researcher might not have independently conceived.
The United States may already have opened the door to this argument, even if in a different doctrinal context. The USPTO now characterises AI systems as instruments analogous to laboratory equipment, computer software or research databases – that is, tools that assist the human inventor rather than co-invent with them. This characterisation was made to preserve human inventorship and for that purpose, it works. But laboratory equipment, software and research databases are precisely the kinds of tools that patent law may attribute to the notional skilled person. The analogy that rescues inventorship may therefore put pressure on inventive step. If AI is an ordinary research tool in the hands of the inventor, it may become harder to explain why it is not also an ordinary research tool in the hands of the skilled person.
The solution is unlikely to be uniform across all technologies. The relevant question will be fact specific: at the priority date, would the skilled person in that field ordinarily have had access to AI tools capable of identifying the route to the claimed invention?
For patent owners, the lesson is straightforward. Once AI participates in the route to the invention, it becomes part of the invention story. That story needs to be understood, documented and, where necessary, managed.
The skilled person has always been a legal fiction. AI may not dissolve that fiction, but it will force us to update it. To paraphrase Neo, in The Matrix, the skilled person may soon need AI tools. Lots of AI tools.
Sharaz Gill is Head of Portfolio Management at Sisvel
The opinions expressed within this article are the author’s and do not necessarily reflect the views of Sisvel. The content is for informational purposes and should not be taken as legal advice.
