Intellectual property and artificial intelligence in the UK: what might the future hold?

AI is a hot topic for the current UK government. Last year, the UK Intellectual Property Office (UKIPO) published a call for views on the current and future suitability of the IP framework to deal with AI technology. The UK government has recently published their response. Here we discuss a selection of highlights from this.

Patents

The section on patents was divided into the following major topics: aims of the patent system (does it incentivise development of AI in the UK, something that our government is very keen on), AI as an inventor (something that got a lot of press recently with the DABUS case where the High Court upheld the UKIPO’s decision to deem two applications withdrawn for failure to name a natural person as an inventor - see our blog Inventive AI: can machines innovate?), conditions for grant of a patent (whether the current conditions cause problems for AI inventions, with impacts on stimulation of innovation and ethical oversight), and infringement (whether the current system is fit to deal with the possibility of AI infringing a patent).

In relation to the aims of the patent system, the main take home message was that all types of IP rights should be considered together when assessing whether the system as a whole adequately supports innovation, with patents already contributing positively to this (e.g. by helping to secure investment, encouraging disclosure, etc.). Much has been written elsewhere in relation to AI as an inventor so we will not dwell on this. The main concern raised was about transparency in the innovation process, and whether or not allowing AI to be named as an inventor would be detrimental to this by encouraging innovation to be kept secret or human inventors to be inappropriately named. The government has promised to consult later this year on policy options for protecting AI generated inventions.

Regarding the conditions for grant of a patent, the main concerns raised were that patent exclusions in the UK (or more specifically, the way that the UKIPO interprets these, which is somewhat less clear and friendly than the practice at the EPO - see our blog How to patent AI technology at the EPO) currently make it difficult to protect or even to obtain a search report for some types of AI invention (in particular, developments in the AI system itself –“core AI inventions” - rather than inventions that apply AI to another technical field), and that there may be some practical and strategic/commercial problems associated with the disclosure of training data. In response to this, the government has committed to: (i) publishing enhanced IPO guidelines on patent exclusion practice for AI inventions, following a review and comparison with EPO practice, and (ii) investigate the possibility of establishing a deposit system for data used to train AI systems disclosed in patent applications.

The former would certainly be most welcome as far as we are concerned. As to the latter, a way to deposit training data could allow applicants to keep a patent specification shorter and thereby save fees. However, the practicalities of how this is implemented could have a significant impact for users. Indeed, in some fields the data can be very sensitive (e.g. medical data comes to mind), and further the data is likely to be of a significantly more heterogenous type than e.g. sequence listings. The introduction of sequence listings similarly removed the needs for excessively long specifications but came with additional constraints including formal requirements, further deadlines and opportunities for deficiencies to cause a lapse of the application. All such practical difficulties and additional constraints are likely to be even more prominent with regards to AI training data in many cases. Additionally, there are cases where it is arguable whether full disclosure of the training data is in fact necessary to fulfil the disclosure requirements. In such cases the requirements for any training data to be deposited could reduce flexibility for the applicant.

Copyrights, designs, trade marks and trade secrets

As mentioned, this post is highly biased so we will whizz through the main points here, focussing on those that frequently come up with applicants considering a mixed strategy involving combinations of patents, copyrights, and trade secrets (which is the most common scenario for AI innovators).

In relation to copyright, an interesting point was raised in relation to whether UK law should provide copyright infringement exceptions for the use of copyright works in the AI training process and/or improve licensing provisions for this. Access to training data is of course extremely valuable to AI developers, which of course conversely means that training data is an extremely valuable asset for many companies. The government has promised to investigate how copyright owners currently license works for use with AI, and to consult on how licensing or copyright exceptions could be improved to support innovation. A framework that clearly and explicitly deals with this could be valuable both for the owners of copyrighted data and for AI developers wishing to make use of this data.

The current protection of the software code itself as literary work was deemed adequate, and existing license mechanisms sufficient to allow exploitation of these. We would add that although software licenses do provide workable solutions that allow co-existence of even open software licenses and patents relating to inventions that may be embodied in this software, the situation is not always clear to many applicants especially when it comes to “free for non-commercial use” licenses. The government has also committed to consult on whether to limit copyright in original works to human creations (including AI-assisted creations), possibly in combination with a related right for AI-generated work.

Turning to trade marks, the take home message was that the current state of AI technology does not warrant a change in the law, but that the situation should be monitored for example if AI fundamentally changes the way that consumers shop. Regarding designs, the question of whether current provisions in relation to ownership of computer-generated work are fit for purpose was raised (according to the Registered Designs Act 1949, the author of a design generated by a computer is the person who made the arrangements necessary for the creation of the design – who this may be becomes a lot less clear for designs autonomously generated by AI). However, any changes in this regard were deemed premature by the government.

Finally, in relation to trade secrets, the government acknowledged comments that the practicalities of ensuring trade secret protection subsists can be complicated especially where disclosure of some information is necessary for e.g. safety or data protection reasons. Nevertheless, the government deemed the current provisions to be adequate. We would agree, although we note that the above-mentioned practicalities are easily overlooked or underestimated. Trade secrets often form a valuable part of a solid and considered overall IP strategy and we would certainly encourage this aspect to be managed from the onset.