AI patentability and sufficiency: new UK guidance

The UKIPO has today published guidance on patenting artificial intelligence (AI) inventions, together with some useful example scenarios. It will be essential reading for patent specialists working on patenting “applied AI” and “core AI”, including training methods and datasets. The guidance focusses on patentability but also touches on sufficiency of disclosure.

The guidance should be a useful companion to the EPO’s Guidelines of Artificial intelligence and machine learning. The new UK guidance cites and discusses key EPO decisions, but puts more weight on UK case law, noting “decisions made by the UK courts relating to the Patents Act 1977 are binding on our practice, whilst the European Patent Office (EPO) Board of Appeal decisions are considered strongly persuasive.” It does not emphasise the discrepancies between UKIPO and EPO practice or offer specific guidance on resolving them.

The explanations and examples of “applied AI” and “core AI” will be useful to patent specialists and may also prove an effective guide to their clients wishing to understand for themselves the broad outlines of what is patentable and what is not. Given the lack of UK case law on the patentability of AI the detailed examples (albeit non-binding) are very welcome.

The discussion of sufficiency is brief, reciting the principles set out by Kitchin J in Eli Lilly & Company v Human Genome Sciences Inc [2008] EWHC 1903 (Pat) (31 July 2008) and noting a specific EPO decision (T 0161/18 (Äquivalenter Aortendruck/ARC SEIBERSDORF) in which the claimed neural network could not be reworked without disclosure of suitable input data.

The guidance is not about the patentability of inventions by AI. That is an issue to be considered by the UK Supreme Court and which, ideally, will be the subject of harmonised international patent law reform. This is being discussed (again) as I type, in WIPO’s sixth discussion of AI: “Frontier technologies – AI Inventions”.