This past week I travelled to Washington, DC, to speak on a panel with US colleagues on patent eligibility and to visit many friends and colleagues. It was great to meet so many people in real life again and to see that in Washington, as in London, life is returning back to normal. I had many great discussions around patent eligibility and how this is dealt with in the US and in Europe. What struck me was that more and more, the approaches on each side of the pond are getting closer to each other. In most cases, it seems the outcomes are likely to be the same, if not the way the law is applied. One aspect of this is whether machine learning is (and should) be seen as abstract math or as a field of technology/engineering.
While there are not that many AI patent disputes in front of the courts in the US, there are many more than in the UK. Recently, I found one dealing with the question of whether core AI is abstract and hence not patent-eligible (non-technical and not contributing to inventive step over here) or not in this blog post. In this case before a district court, Health Discovery Corp asserted a patent against Intel Corp, related to a development of Support Vector Machines (called SVM-RFE for short). The question on which the case turned was whether SVM-RFE, which was identified as the core concept of all claims, was an improvement in an abstract mathematical method or an improvement in computer technology. This corresponds closely with the notion in Europe that AI methods per se are to be treated as mathematical methods, themselves considered to be not inventions “as such” (see for example here and here). The court then asked the question as to whether any claim “contains an ‘inventive concept’ sufficient to ‘transform’ the claimed abstract idea into a patent-eligible application.” Here, the court was looking for innovation in the non-abstract application realm. Although some of the claims were limited to a particular field of invention or input data (e.g. gene expression), that was of no consequence. Similarly, the fact that the mathematical steps were recited in some detail or required running on a generic computer or outputting information on a generic output device did not help. Likewise, at the EPO, merely specifying input data that is technical in nature will not be enough (see here).
For sure, the way the concepts of patent-eligible subject matter are applied in the US and Europe are very different. The former asks whether the subject matter is inherently eligible for patenting, and the latter considers this to be the case for any computer-implemented invention. Instead, the European approach moves the filter to the question of obviousness: only features that contribute to the (non-abstract) technical character of the invention are considered to decide whether a claimed advance is obvious or not. That said, in most technical fields, it seems that our jurisdictions are converging as far as the likely end result of the patentability question is concerned. If your invention concerns a diagnostic method, you are likely to fare better in Europe. Your chances of success are better in the US if your invention relates to natural language processing (understanding or classifying text based on its content). But on the whole, it seems that the similarities are more significant now than the discrepancies.