Patenting AI and computer simulation – an interview with Partner Mike Jennings

An interview with Mike Jennings of AA Thornton – Patenting AI and computer simulation

Partner Mike Jennings is a regular speaker on patenting computer-implemented inventions (CIIs).  In this interview, we asked Mike for an update on European patenting of artificial intelligence and computer simulation.

AA Thornton has a substantial practice protecting CIIs.  Clients include some of the world’s leading developers of computer-implemented inventions – applying computer simulation and monitoring technologies, quantum computing and artificial intelligence to solve problems ranging from healthcare to 3D printing, security, subterranean exploration and resource management.


Q: I know you have been involved in patenting computer-related inventions since the 1990s, so you must have seen major changes of law as well as technology in that time. Has patent law kept pace with the advances in technology?

Mike: I would say European patent case law evolves at quite a slow pace, but that is usually preferred by established industry players and it allows patent attorneys to offer good predictions to our clients.

I started working on software patents in 1992, and the threshold test for patentability has changed in a number of countries since then; but I have to admit that European patent attorneys have been luckier than our colleagues in the US as we have enjoyed a decade of consistency and predictability at the EPO. We have a slightly different but nevertheless clear and workable test at the UKIPO, and we have obtained large numbers of patents for computer-implemented inventions in both offices.

As you know, the US courts and USPTO had a major shift in 2014 following the US Supreme Court’s decision in Alice v CLS, and started rejecting a lot of claims to computer-implemented inventions while waiting for more guidance on which types of invention remained patentable. In 2018, the EPO issued revised examination guidelines based on their recent case law and the USPTO’s Director Iancu took initiative in the US and set his office the task of clarifying what subject matter is eligible for patent protection. The USPTO added to its examples of patent eligible subject matter in January 2019, and this was quickly followed by some helpful proposals from the reinstated US Senate Judiciary Subcommittee on IP. The JPO also added some AI examples to its Examination Handbook in 2019 and highlighted the requirement for an enabling disclosure which will be an important issue for AI cases. The USPTO added more examples in October and China’s National IP Administration office has also issued new guidelines (effective from 1 February 2020) that set out a similar approach to the EPO approach.

That brings us to 2020, with artificial intelligence accelerating the pace of innovation in many industries. Major patent offices have offered encouragement to applicants who wish to protect new and inventive technical applications of AI technologies, as long as the claims are limited to the specific technical purpose. With the IP5 offices working closely together  and similar guidance being issued by patent offices around the world, it seems to me that we have the best ever opportunity for harmonisation of international law on CIIs – with technical solutions accepted and computer-implemented business and administrative methods rejected.

But we now face the question of whether we should be able to patent innovative solutions that are proposed by an AI system, instead of a human inventor. AI systems are now solving complex problems that were neither solved nor fully defined by human inventors, so the question of whether we should grant patents for these solutions is an important one that needs to be discussed. To change our inventorship requirements would need a change of legislation.

Q: I was going to ask you about AI inventorship later, but since you have raised it… What are the EPO and UKIPO saying about these inventions? And do you think the law needs to change?

Mike: It is important to differentiate between the applications and implementations of AI technology that are being invented by people, which often are patentable, and outputs from an AI system for which no human inventor can be identified.

The EPO’s Patent Law Director has been very clear that they cannot grant patents without human inventors being named, and the EPO made a point of announcing some recent refusals of patent applications naming an AI system as the inventor. This was added to the EPO’s website homepage on 20 December before detailed reasons were published on 27 January. The UKIPO also refused the UK counterpart applications. These decisions were entirely predictable, as current European and UK patent law required the offices to refuse these applications.

I do think there should be a debate about the possibility of a change of legislation, but it is essential in my view that we hear a wide range of industry views on this. For some people, it is a fundamental principle that new and inventive solutions should be patentable regardless of their origin, whereas others want a wider debate including consideration of the economic impact of removal of the requirement for human inventorship before any legislative change. We have been discussing this within CIPA’s computer technology committee, and in AIPPI we just began an international study of AI inventorship. WIPO launched a public consultation on AI and intellectual property and published a draft issues paper in December, and I encourage industry and all interested people to contribute to the debate.


Q: You mentioned that there is a slightly different test for patentability in the UK and Europe. So how close are we to harmonisation between the EPO and UKIPO on software and AI patentability?

Mike: The EPO applies the European Patent Convention and EPO Board of Appeal case law that focusses on new technical effects to decide whether a computer-implemented invention has technical character. The UKIPO takes account of the leading EPO appeal decisions but applies a patentability test defined in UK national case law. The two are definitely closer than a decade ago and will often produce the same result, but their different approaches can produce different outcomes for claims at the threshold of allowability – for example UI inventions have a higher success rate at the EPO than UKIPO. As I mentioned, EPO practice on computer-implemented inventions has been stable and predictable for a decade, and UKIPO practice has also been clear and predictable for almost as long. Both offices accept that computer simulation can provide the required technical effect for patentability, and the UKIPO’s latest comments in January 2020 included a confirmation that claims to the use of AI in a specific technical process are not excluded from patent protection. However, we currently recommend EPO filing for core AI algorithms that control a computer or neural network’s internal operations.

Although they differ slightly in their approaches, I congratulate each of the EPO and UKIPO for achieving a high level of internal consistency within their own office and for explaining their approaches. Consistency and transparency gives us highly predictable outcomes – and industry loves predictability !


Q: I understand you have praised the EPO for their updated patentability Guidelines – particularly their identification of which AI inventions can be patented. What do you think they did well, and is there still room for improvement?

Mike: The EPO team carried out a major review of their patentability Guidelines in 2018, including reviewing around 1000 Board of Appeal cases, and the Guidelines are a clear and helpful abstraction of the principles set out in the case law. In relation to AI, I think they did well by explaining that the requirement for technical character and the established approach to assessment of mathematical methods would be applied to machine learning algorithms and computational models just like other mathematical methods, and by giving us lots of examples of technical applications of AI that could qualify.

Patents are being awarded for ‘specific technical applications of machine learning and other mathematical methods such as classification and genetic algorithms, when the claims are functionally limited to the specific technical purpose, but patents will not be granted for algorithms if claimed in abstract form – isolated from any technical purpose. The EPO’s helpful examples of specific technical applications include control of technical systems, image and video analysis and enhancement, healthcare applications, encryption of communications, and simulation of technical processes. The new Guidelines also note that generating a training set and training a system can contribute to the technical character of a classification method when relevant to the technical purpose of the method.

However, we have asked for a bit more guidance and examples of what “specific technical implementations’ of AI methods the EPO considers to be patentable. The EPO noted that a mathematical method may be patentable if it is adapted for a particular technical implementation (in the sense that its design is motivated by technical considerations of the computer system that will perform the method). But the EPO’s 2018 Guidelines only offered one negative and one positive example of a ‘specific technical implementation’ and it seems possible to interpret this requirement for adaptation to the hardware system quite narrowly. I believe the EPO will add to their examples in the next version of the Guidelines, but there is still work to do here to help applicants and others understand what adaptations and implementations will be accepted.


Q: You mentioned that the EPO Guidelines identified computer simulation in their examples of patentable inventions. When you spoke at the Technology Patent Network Europe last summer, you commented on an EPO Board of Appeal’s referral of computer simulation questions to the Enlarged Board of Appeal. Do you want to comment on this today?

Mike: Yes, simulation of the behaviour of technical items or a technical process has been accepted as patentable for more than 10 years, and is one of the EPO’s positive examples of uses of AI. The section of their examination Guidelines on simulation, design and modelling  notes that “computer-implemented simulation methods cannot be denied a technical effect merely on the ground that they precede actual production and do not comprise a step of manufacturing the physical end product”. This is consistent with UK and German national law but the referring Board of Appeal wants to change this.

I was disappointed with the referral, and in particular the way in which the referring Board chose to criticise another Board of Appeal’s 2006 decision T1227/05. This related to an Infineon circuit simulation invention and the decision set out a clear and workable test that has been followed in many cases. It noted that specific technical applications of computer-implemented simulation methods were technical methods themselves, and involve technical effects regardless of whether the claims incorporate the physical end product. Note that this “specific technical applications” wording now appears in the EPO’s Guidelines for mathematical methods including AI and has been applied consistently for 14 years. The UK courts also confirmed patentability of computer simulation in 2012, by accepting claims to Halliburton’s method of drill bit design using simulation, without requiring the claims to recite actual manufacture of drill bits.

As we noted in CIPA’s Amicus Curiae brief for the Enlarged Board’s consideration in referral G1/19, Europe’s decade of consistency on computer simulation patentability has been achieved by applying the reasoning of T1227/05 and focussing attention on the technical purpose of a claimed computer simulation.

It will be helpful to hear from the Enlarged Board, but my initial reaction was that the referred case T489/14 could have been rejected without a referral because the claims to computer-implemented modelling of pedestrian crowd movement include vague mental considerations (determining “dissatisfaction”, “inconvenience” and “frustration”) to determine each pedestrian’s preferred next step towards a destination.

Part of my disappointment with this referral is that it encourages the Enlarged Board to consider scrapping a clear and workable patentability test that has provided industry with consistency and predictability for a decade. Secondly, the referring Board chose to disregard important parts of T1227/05. Thirdly the referring Board’s desire to see claims explicitly limited to a “change in or measurement of a physical entity” would avoid direct infringement unless a claimed simulation is performed by the same party in the same country as the operation which changes the entity – ignoring the fact that simulation and manufacture are both technical steps but may be separated in time and place and be performed by different parties. This would take the law backwards to a time before 2006 and create an inconsistency with Germany’s 1999 Verifikation case and the UK’s 2012 Halliburton drill bit design case.

It is ironic that an EPO Board of Appeal has chosen to question established EPO practice just after the USPTO endorsed that practice by including a similar computer simulation claim within its January 2019 examples of patent eligible subject matter. However, if the EPO Enlarged Board upholds current EPO practice but adds some guidance for particular types of simulation invention, and if the US courts or US senate support the USPTO’s recent initiatives, we will still have an opportunity for greater harmonisation than we have had for many years.


Q: Do you have any predictions on how the EPO’s Enlarged Board will answer the referred questions on computer simulation?

Mike: It is probably not helpful to offer detailed predictions so close to a decision being made, but I was greatly encouraged by the consistency between the Amicus Curiae briefs that were offered to the Enlarged Board. Numerous companies and industry bodies such as IP Owners Association, IP Federation, FEMIPI and EFPIA, and organisations such as CIPA, EPI, FICPI and AIPPI all supported the current EPO approach and the reasoning set out in T1227/05, in preference to the comments in T489/14.

It is also worth noting that the Enlarged Board invited the EPO President to provide comments and the EPO President obliged with a helpful and detailed explanation of current practice. I am sure the Enlarged Board will give all of these briefs their careful consideration before answering the referred questions, and I look forward to receiving their guidance.


Q: Thank you Mike. It sounds as though you are optimistic that the Enlarged Board will not shift the law too far away from current EPO practice, and enthusiastic about the USPTO’s recent efforts. I guess we must still wait and see how the US courts and legislators will steer US law.

Mike: Yes, we must wait a little longer for the EPO’s Enlarged Board, and the US legislative reform will take time, but we have seen some positive decisions in the US courts and it was encouraging to see that the recommendations for reform from the US Senate Judiciary Subcommittee came from republican and democratic Senators and were supported by members of the House of Representatives. So I am optimistic.


Q: Thanks for answering my questions and sharing the historical perspective as well as bringing us right up to date. You are clearly passionate about your work.

Mike: I think you have to be passionate, or change career. But yes, it has been a privilege to work in this area at a time of such rapid technological change, and I have been very fortunate with my roles and the people I have worked with. I learnt a great deal from my time at IBM, where we worked on many of the technologies that have enabled the current AI explosion, and at AA Thornton I am still learning and have the opportunity to work on a vast range of inventions, including advanced computing and communication technologies, imaging and robotic control, and a range of sensing and measurement technologies. Clients are using computer simulation to save costs and improve operational efficiency in 3D printing and to control operations in harsh and inaccessible environments, and using AI to redefine every industry from healthcare to automotive. They are even starting to disrupt the digital computing world with quantum computing. So yes, I am enjoying myself.


Mike is a Partner at AA Thornton in London. Before joining AA Thornton, Mike spent more than a decade within IBM’s IP law team, where he handled software patent protection for inventors from around the world, as well as software patent licensing and protection of inventions in communications, microelectronics, computer architectures, data storage, etc. Mike joined AA Thornton in 2007, becoming a Partner in 2008, and has seen the firm’s computer-related patent work grow rapidly since then. Mike specialises in European and UK invention protection and EPO oppositions, and has an exceptional success rate at EPO appeals. He is a member of CIPA’s Computer Technology Committee and of the EPO’s Standing Advisory Committee’s working group on quality. He contributed to AIPPI’s international study on computer-implemented inventions in 2017, and was invited to join the EPO’s discussions on AI patenting in 2018. He was invited by CIPA to educate UK and US patent attorneys on software and AI patenting in 2018, 2019 and 2020 and is contributing to CIPA’s and AIPPI’s discussion of AI inventorship.


Category: Latest Insights | Author: Mike Jennings | Published: | Read more