The fading of summer is often one of those times to finish off all of those tricky projects that have built up. This is always a never-ending race as just as one thing is signed off external events find something new for you to do. In this case the UK Intellectual Property Office (UKIPO) has decided to mark the end of the summer with the publication of its long-trailed AI patentability guidelines.

Originally announced as part of the conclusion to the UKIPO's first AI and intellectual property consultation that ran from September 2020 to March 2021 - which itself led to further work (discussed in previous bulletins) by the UKIPO exploring the appetite for changes to UK patent and copyright law around AI innovation.

Whilst, in respect of patents at least, changes to the law seem to be off the cards for now, the UK government is still keen to "unleash the power of AI" and these new Guidelines very much seem to focus on what is possible in terms of AI patents.

Meet the new approach
As with the European Patent Office before them the UKIPO is keen to emphasise that in principle AI innovation is dealt with in exactly the same way as any other computational/mathematical method or system. Many of the cases cited in the new Guidelines will be familiar to those of us who handle what are traditionally called "computer implemented inventions (CII)" at the UKIPO.

As always at the UKIPO the key consideration as to whether a CII is patentable remains "what is the technical contribution" (and is it really technical in nature). Also, as always, those not familiar with the rather particular way the patent world uses the word "technical" will, perhaps, be a little bemused at some of the outcomes.

Nevertheless, the Guidelines provide a nice discussion of some of the leading case law and, more importantly, do not shy away from tackling some of the more thorny issues around AI patentability. These include the patentability of training processes, training data (and associated processes) and even "core AI" - along with some brief discussion around what training data you may (or may not) need to disclose in an AI patent application.

The issues are addressed using some recent AI decisions where possible along with (non-binding) example "scenarios" of what (potentially) is and what is not patentable with reference to the more established non-AI case law.

A view ahead
Whilst the overall approach to AI patentability set out in the Guidelines is very much using the "old" CII system there are still some interesting, and perhaps surprising, directions hinted at - which may make for some interesting future cases. We intend to take a deeper dive into some of these in upcoming bulletins.

In any case these new Guidelines provide a good overview of what the UKIPO practice is (and to some degree how it got there) and demonstrate the wide range of possibilities for protection of AI innovation in the UK patent system. The art of the possible may, after all, provide applicants with valuable protection for their AI innovation.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.