AI patentability & case law review 2021

AI patentability & case law review 2021

This year AI case law kicked off with this England & Wales High Court appeal from a patent office decision by an individual against the refusal of their patent application by the patent office. While the technical details of the application in suit were somewhat vague, the court made some useful comments on the application of the UK tests for patentability on speculative subject matter, as well as reminding us on the limitations of an appeal to errors in law (rather than the re-evaluation of facts). Hot on the heels of the High Court, the EPO Boards of Appeal issued their first decision of the year (and in some time) in an AI-related case, reported here. The board made some concerning comments about supervised learning but looking at the facts of the case, it simply seems to be a case of a limited technical concept not going beyond simple administrative rules. Hopefully this case will not make mischief for AI patent applicants in the future.  

With no new cases to get stuck into, here I looked at whether the EPO guidelines got it right in terms of applying the case law to the patentability of fundamental AI innovation (inventions of general applicability and not limited to any specific technical application). I concluded that the wide-spread hardware-focussed approach in first instance examination is probably not the full answer and that the actual case law seems to be more inclusive. Some on the ground experience from the EPO later in the year seems to support this notion, applying principles laid down in an interesting appeal decision we dissected here. The details of the case provide a clear framework for what is needed for algorithmic efficiency to contribute to a technical character in the context of its implementation in a computer. While we are pushing the envelope for fundamental AI at the EPO, the UK patent office decision reported here suggests that the IPO still is unsure whether an algorithm implemented in actual, physical and fixed, hardware can take an invention out of the exclusions from patentability. Ultimately, the decision comes to find that it does (phew).  

Finally, at the tail end of 2021, the EPO Boards of Appeal gave us another AI-related decision to mull over, this time on the “specific technical application” route to patenting AI. The board provided us with some guidance as to what is needed for supervised learning to make a technical contribution in a specific application. Inventive step came down to what an alleged technical contribution of improved performance was linked to. Unfortunately for the applicant, the board found that any potential performance improvement would depend on the quality of human assessments as part of the training data, rather than any features of the algorithm or input data itself. As a consequnce, a technical contribution was not acknowledged. Fortunately for the rest of us, this case provided another data point on how the EPO deals with this kind of subject matter.  

I will be here next year again, looking out for case law relevant to patenting AI in the EPO and in the UK. With more and more cases making their way through the examination and appeal pipe line, I suspect it will be a busy year.  

Read previous article | Read next article

Follow Alexander Korenberg’s AI musings

AI musings