At the end of last year, the High Court heard an appeal (MR COREY KAIZEN REAUX-SAVONTE vs COMPTROLLER-GENERAL OF PATENTS, DESIGNS AND TRADE MARKS) brought by a private applicant against a decision by the patent office to refuse an application related to an Artificial Intelligence able to evolve without human intervention. While the case does not turn on any details related to AI, it is a useful read in relation to AI patents as it makes some comments on the relevance of theoretical technical contributions that are not supported or disclosed in any technical detail in the application. Further, it drives home, very clearly, the importance of making a complete case at the patent office in light of the limited review an appeal from the patent office will contemplate.
The decision can be accessed here for those interested in its finer detail and is both a light read and salutatory case to consider for anyone with an interest in patenting AI. For those with insufficient time on their hand, I provide a summary of the main points below.
The purpose and remit of an appeal
The judge was careful to emphasise that it is the hearing officer, not this court, who is the primary evaluator of whether the statutory test for patentable subject matter has been fulfilled. The court cannot overturn such a determination unless there has been an error of law or the evaluation is sufficiently clearly erroneous. The court must ask, having analysed the points made on appeal, whether it is in a better position correctly to apply the statutory test to this application than the hearing officer was, and only to interfere with the decision if it is sufficiently clear that it is.
In the present case the judge found that rather than being able to point to an error of law or approach in the hearing officer's decision, the appeal is impermissible for inviting this court to re-evaluate the matter afresh, taking a more generous approach. This is a reminder of the importance of making ones case at the patent office as first instance as fully and completely as possible, as the court will only investigate whether the decision applies the law correctly and reasonably evaluates the case in front of the patent office. In that respect, the court’s approach is in line with the very strict approach the EPO’s boards of appeal now take in relation to change from one’s first instance case at the EPO.
The application and first instance decision
The application related to a computer implemented method involving an AI able to evolve without human intervention through the use of an Artificial Intelligence Genome, an Artificial Intelligence Genome Organizer and an Artificial Intelligence Genome Controller. The judge found that the application describes a function implemented in software by using analogies from the field of biology which do not have a clear or unambiguous meaning in the context of computer science or artificial intelligence and that an "AI genome" is not a term of art in this field. It was agreed that this description is no more than using an analogy to describe characteristics of the functionality which is implemented in software. That was said to be clear from the specification as a whole and the way in which the appellant has described this as a "self-contained, modular, and hierarchical code structure" with the organiser as a "manifest file" and the controller being a "program". Each of these either are, or are characteristic features of, computer programs.
Given the use of biological features as an analogy, the judge held that the use of biological analogies therefore need to be treated with caution to determine what the actual subject matter of the invention is. Further, the judge held that, in the application, the "genome" is also said to be used to create "digital and physical entities" and that it is not specific as to whether software is to create new physical entities and potentially contemplates the production of purely digital entity. It is also said to be unclear how it is that the genome is to "create a new genome" or how that alleged functionality could be achieved. No working example is given in the application of that being done whether for any digital or physical entity. Finally, the judge found that it is also unclear from the specification how the alleged "genome" could be used to produce any "physical" entities, although the description refers to the possible use in robotics.
At first instance, the patent hearing officer asked the appellant to explain in his own words where he said his invention lay and to explain what, in substance, he said his contribution was and took that into account in evaluating whether the statutory test was satisfied. The judge found this was justified in the context of the circumstances of the case.
The result of that discussion at first instance was that the contribution made by the invention is computer code, a sort of hierarchical or a modular arrangement of computer code that facilitates evolution over time so that the computer code is able to modify, adapt, change and improve over time in the same way that biological code evolves through reproduction. The hearing officer concluded that this contribution was not technical because there was no technical effect outside the computer, the alleged effect of enabling the healing and reproduction of AI genomes in an analogous way to biological genes in nature (said to be allowing the creation of physical systems outside a computer) being purely theoretical and not supported by any detail described in the application. Further, the organisation and structuring of AI code was found not to operate at the architecture level of the computer, nor to make the computer operate in a new way, so no technical effects were to be found at that level.
The appeal decision and its wider significance
While the specific technical details and the level at which they were disclosed in the refused application clearly explain some of the reasons for this negative decision, the judge made a number of observations that provide useful insights into how the UK patent office and the English courts apply the exclusions to patentability to AI related inventions. In the judges words:
“The first signpost [technical contribution outside the computer] cannot be satisfied […] by pointing to a potential purely theoretical impact which it is not shown has been contributed by the invention.”
“The statutory exclusion does not cease to apply merely because a particular structure of a computer program can be applied over a wider area than mere data manipulation and may be instantiated in the context of robotics especially where, as here, the application does not provide sufficient guidance to a skilled person as to how that should be done.”
And also, specifically in the context of the possible application to robotics mentioned in the application:
“If a patent application refers to the merely theoretical possibility of applying an alleged invention to which the statutory exclusion does apply to aspects of robotics but does not provide sufficient information about how that should be done, the mere possibility of that being implemented in a way that the patent has not enabled does not mean that the patent application will escape the excluded subject matter objection. The statutory exclusion focuses on the contribution made and disclosed in the patent application under consideration, not on the contribution which may be made (possibly by others) using the subject matter of the application to create products or system to which an excluded subject matter may not apply.”
One aspect of assessing whether an invention is excluded or not is to consider whether the contribution is made at the level of architecture of the computer or where the effect is produced irrespective of the data being processed or the applications being run. In that respect, the judge commented that:
“It is not good enough to avoid such an objection to say that, because characteristics of software are described at such a high level of generality as may render it applicable over a wider field, such satisfies the requirement of operating at the level of architecture.”
Another area in which a technical contribution may arise is when a computer programmed with the software in question operates in a new way. Unsurprisingly, the judge endorsed the first instance ruling that a “computer system operating on new code does not imply that the system works in any way differently to how it would with the old code”. In other words, it is not enough to show that the code itself is new or represents a new sequence of instructions.
As a result of these findings, the appeal was dismissed.
While the judge had already indicated in the order made after the hearing the leave to appeal would unlikely be granted, the question of leave to appeal and also costs were dealt with in a subsequent decision that formally dismissed the appeal. As to leave to appeal, this was not granted, it being said that the case would not be suitable to develop the law. Equally, leave to amend the application was not granted as no reasonable prospect of finding patentable subject matter was found. As to cost, the patent office asked for its cost in defending the appeal and the judge awarded costs of GBP 8000 against Mr Reaux-Savonte on the basis that this sum was not particularly high for a case in this subject area involving a one-day hearing and that the explanation for the costs provided by the patent office was reasonable. The order as to leave to appeal and amend, and cost, can be found here.
While the outcome of this case may be not entirely surprising given the nature and technical detail of the disclosure in the application in question, the fact that the judge went into some detail to explore whether the alleged technical contribution can help to meet the statutory test for patentable subject matter nevertheless makes this case useful reading and offers some lessons to heed.
This case illustrates that applicants should resist the temptation to define technical subject matter by analogy (whether biological, anthropomorphic or otherwise) but rather to define the disclosed and in particular claimed subject matter in precise terms that can safely be assumed to be known to a person skilled in the field of the application.
Watching out for the use of analogies is particularly relevant for AI related inventions, where the complexity of the technology can make the use of analogies particularly tempting and the often anthropomorphic view of AI in common parlance can provide a false sense of clarity that does not withstand careful technical analysis.
If you have any questions or would like to discuss this topic further, please contact Alexander or your usual Kilburn & Strode advisor.