AI and IP – what is your strategy?

This morning the UKIPO hosted a round table of industry and academic experts discussing how to protect AI generated inventions. This coincides with the UK Government’s consultation on potential changes to UK IP law on the protection of computer-generated works for the purpose of copyright, the protection of inventions by AI and exceptions to copyright for text and data mining.

A recording will be made available. In the meantime, standout discussion points included:

The challenges to traditional IP strategies. The adoption of AI in industries such as aviation and pharmaceuticals requires even highly sophisticated IP owners to reassess their IP strategies and to educate their R&D teams at all levels to understand the changes. These include the relative importance of trade secrets and the best ways to utilize patents, copyright and database rights for inventions relating to, assisted by or generated by AI.

The need for clarity and certainty over the subsistence and ownership of inventions by AI in order to attract investment. This is particularly important for inventions that are commercialised in the public domain, such as new alloys for aviation or new pharmaceuticals. Such inventions, if they can be copied, need monopoly protection to reward investment. For a new pharmaceutical the upfront investment may run to hundreds of millions of pounds.

The complexities of identifying whether and which humans are inventors where an AI is involved. Practically, AI projects involve many people and the legal requirements for inventorship diverge internationally. Some legal systems penalise misidentification of inventors (e.g. in the US, a patent fraudulently naming the wrong inventor may be unenforceable). There was discussion of whether misidentification could be detected and proved in practice (although one may expect litigants to try to prove misidentification).

How to enable and encourage companies involved in safety critical applications of AI to share data. Real world data from failures of safety critical systems (e.g. in aviation) are very sparse. Companies seeking to train AI to predict failures of such safety critical systems need to enlarge their training data with synthetic data and, ideally, real world data from competitors. There was discussion of potential regulatory requirements for data sharing (and, no doubt, IP rights in data, competition law and sector-specific considerations will also be relevant).

The panel was:

• Adrian Weller, Programme Director for AI, The Alan Turing Institute (chair)
• Ryan Abbot, Professor of Law and Health Sciences, University of Surrey
• Caroline Gorski, Group Director of R² Data Labs, Rolls-Royce
• Philip Guildford, Chief Operating Officer, Department of Engineering, University of Cambridge
• Dan Sola, Founder, Archangel Imaging
• Jason Rice, Assistant General Counsel, Patents, GSK
• Michael Prior, Deputy Director of Patents Policy, Intellectual Property Office

Comment

Critical to the UKIPO’s consultation is whether the UK can make unilateral changes to IP law that will continue to attract investment in AI to the UK. According to the UK Government’s National AI Strategy, the UK enjoys the third highest private investment in AI in the world and the Government is looking at reforms to IP and privacy laws to keep the UK a global AI “superpower”.

Protection for computer-generated works and text and data mining (TDM) exceptions are strong candidates for unilateral action. The UK introduced both before its European neighbours and has wide leeway to make changes. The UK Government’s decision not to implement the broader TDM exceptions in EU’s Directive on Copyright in the Digital Single Market has currently put the UK at a disadvantage for R&D, especially commercial R&D.

By contrast, it will be hard for the UK to change patent protection unilaterally. There may be scope to protect inventions by AI in the UK: the laws of Monaco and Cyprus do not require a human inventor; the laws of Australia have recently been interpreted (subject to appeal) as allowing patents for inventions by AI; and one of three of the judges in the UK Court of Appeal in the “DABUS” case held there was no impediment under UK law (see our summary here). However, it is doubtful that companies will pursue patent protection in the UK alone if that means disclosing their invention for the purposes of major markets in which patent protection is not available and where trade secrets afford viable alternative protection.

The UK must also work hard on enabling data sharing. This may require adjustments in the law (including IP and competition law) and governments may encourage sharing through technical and practical support for data markets. Again, the UK needs to remain competitive with the EU, which has been focusing on these issues for some time, and the rest of the world.

In the meantime, companies pursuing AI should undoubtedly follow the lead of panellists Jason Rice (GSK) and Caroline Gorski (Rolls-Royce) and work with their R&D departments to review their IP strategies for the law as it stands.