EPO updates its guidelines with a section dedicated to AI

The EPO has updated its official guidelines to include a specific section on how the office is likely to assess patent applications directed towards artificial intelligence (AI). A preview of the update can be found on the EPO website and will come into force on 1 November 2018.

The update to the guidelines provides further clarity on how the EPO’s existing legal framework will be applied to AI inventions. Generally, the update confirms that the same rules that are applied to all computer-implemented inventions will apply to inventions involving AI. The rules stipulate that mathematical methods per se are “devoid of technical character” and thus are not patentable when considered in isolation. However, inventions that use mathematical methods remain patentable if they provide a technical solution to a technical problem. The EPO’s guidelines now state that AI and machine-learning algorithms are considered to be mathematical methods. Therefore, an invention that uses AI or machine learning must solve a technical problem in order to be patentable, in the same way as any other computer-implemented invention.

The definition of what constitutes ‘technical’ has evolved out of the case law of the EPO Boards of Appeal. For some inventions, there may be no case law providing a clear indication as to whether a problem solved by an invention is considered to be technical by the EPO. This can create uncertainty for applicants. However, the EPO has now given specific examples of how AI inventions will be assessed. In one example, the update indicates that the use of a neural network in a heart-monitoring apparatus for the purpose of identifying irregular heartbeats makes a technical contribution and thus would be patentable. In another example, classifying text documents solely in respect of their textual content will be viewed as being non-technical in nature and therefore unpatentable.

The new section on AI addresses the question of how the EPO’s statutory exclusion to patentability will be applied. However, any invention must still be non-obvious in comparison to the prior art in order for a patent to be granted. On the one hand, it seems that an invention that uses a new AI algorithm to tackle a technical problem would stand a good chance of being patented. On the other hand, an invention that applies a known AI algorithm to an existing technical problem will be more difficult to protect. Nevertheless, it may remain possible to obtain patent protection for these inventions depending on the specifics of the prior art and the strength of the arguments put forward during the patent application process.