This is a question that came up in discussion after last month’s newsletter. Indeed, when you look at data comparing success rates in Community Patent Classification class G06N (which broadly maps onto many machine learning technologies and their applications), patenting AI seems hard work. In particular, applications in this class have a 40% grant rate compared to an overall grant rate of 69% and refusal is also more likely (16% compared to 4% overall)*. So why do we see this difference in success rates?
One possible explanation is that the EPO see AI methods as abstract math and therefore requires something extra to acknowledge a technical effect and hence inventive step. This extra can come from a technical implementation or a technical application. While the former can be used to seek protection for core AI methods, at least getting broad claims in this area is indeed tricky. However, this applies to, in my experience, a minority of AI inventions, with the majority being related to a technical application. And, anecdotally and based on my experience, these applications, if well drafted, are no more difficult to prosecute than any other software application, which the EPO grants readily if the invention achieves a "technical effect" in a non-obvious way.
Could there be another explanation? In my practice, I see some AI cases that have been very well drafted but also a significant number, more than average, with some drafting issues. These issues generally seem to be linked to a lack of a complete understanding of the technology by the draughtsman. Given that it is only relatively recent (compared to the time it takes to train and gain experience in the patent profession) that AI has taken on the commercial importance it now clearly has, it is maybe not surprising that there is a scarcity of patent attorneys with technical expertise and a deep understanding of AI technology. This is certainly something that I see when hiring for my team. True AI expertise is hard to come by. Therefore, we have started to encourage our trainees to add AI technology to patent law in their study diet.
The issues we see range from clarity problems, for example, when certain terms are taken as understood when their technical meaning is not that clear cut, to the description being light on detail or even contradictory when it comes to explaining the core of the invention. This can leave the prosecuting attorney and the patent office with some guesswork as to what the invention actually is, not a good path to success. On the other hand, even in cases where the technical description is entirely accurate and sufficiently comprehensive, the description may be missing an appreciation of what technical effect or benefit is achieved by the invention and how this links to the features that define the invention. This is a particular issue at the EPO where a technical effect is left, right and centre of the assessment of inventive step. Again, fully appreciating technical effects and drawing these out from all but very patent-savvy inventors requires a true understanding of AI.
So what can be done about this? No doubt, as time passes more and more patent attorneys will have a realistic understanding of AI from their academic or engineering experience and background. In the meantime, in-house teams need to be more vigilant (and perhaps encourage and educate their business’ inventors in the critical review of patent applications). And for those drafting applications to be alive to the limitations of their understanding and make sure that in the end no technical stone is left unturned in the quest for a high-quality AI application. While there is a lot of hype and confusion regarding AI technology, it is fundamentally no more difficult or complicated than many other complex technologies. With an increased awareness of this in the patent profession, the quality of AI applications should improve, followed no doubt by improved grant rates in time.
Finally, if and when the EPO recognises AI technology for what it is, technology, as it has recently done for database technology, obtaining patent protection even for broad and fundamental AI technology should be no more difficult than for any other field of technology.
*For those with a penchant for statistics, see here, where these numbers come from.