InLaw: Kilburn & Strode look at the ‘if and when’ of patenting AI-related healthcare inventions

Although artificial intelligence (AI) – in essence, machine learning – is by no means a new concept, the growth of computer processing power has, in recent years, helped AI make the transition from an active field of academic research to a practical tool in uncountable engineering and scientific applications. AI is increasingly being employed throughout the healthcare sector and we see a number of emerging areas of immense commercial potential – for example, the use of AI to shorten drug development cycles or predict protein structures.

Kilburn & Strode’s Nick Noble, Sarah Lau, and Alexander Korenberg look at a number of patent-related AI issues specific to the healthcare world

Can you patent AI inventions?

Fundamentally, AI is software. Contrary to common misconceptions, the practical applications of software, AI or otherwise, can be protected by patents and this is true as much in healthcare as in any other field of technology. AI healthcare inventions can be patented in principle but what form this may take and the possible boundaries and pitfalls of patenting in this space are, is beyond this short note and is highly fact specific.

So, you can patent AI healthcare inventions. But should you?

Should I patent my AI healthcare invention?

This question must be approached from a business perspective of using patents to protect investments in research and development and to provide or enhance competitive advantage. You might expect patent attorneys to simply say ‘yes’.  However, for AI-related inventions, there are sometimes reasons why it might make sense to not go down the patent route. Here are some of them:

You don’t want to disclose your AI method. In some circumstances, it may be better to protect your approach as a trade secret. Since patenting requires that you disclose the invention in the patent application, if protecting your edge involves trade secrets, patenting is out of the question. Trade secrets can be a powerful way to guard a technological edge but are not without challenge. Often, significant investments in infrastructure and processes are required to ensure that trade secrets stay secret. Also, in some areas, in particular where AI is used as a medical device, regulatory requirements may mean that secrecy is impossible

The speed of development of AI technology is so high compared to the three plus years that it can take to grant a patent application that a patent application could become obsolete before a patent is even granted for it. Of course, even if other new technology has been developed, a patent that protects core aspects of your approach may still provide a valuable defence against imitation by others. The skill of your patent attorney in future-proofing your patent application is critical here

The European Patent Office is developing a consistent and predictable approach for handling AI-related inventions. However, given that its wide application is a relatively new phenomenon, many examiners (and patent attorneys) do not yet really understand the technology very well. This can make the process of getting a patent harder and possibly less predictable. However, this is not so much a reason to not patent, rather it serves as a note of caution to carefully select those projects for patenting that are of sufficient commercial importance. It also suggests that it’s worth investing in high quality advice from a patent attorney with a track record not only of the legal and procedural aspects of obtaining a patent but also a true understanding of AI technology

Getting a patent granted for your AI method is one thing but demonstrating that another party is using it can be a further challenge. Whereas with a patented product you might make test purchases from a suspected infringer, a competitor’s AI method may be performed completely on their servers or in the cloud. Unless the competitor publicly discloses their method, a patentee’s options for identifying infringement may be scant and/or unfeasible. Luckily however, with AI being a relatively young field, the drive to publish one’s technology is strong, and can be crucial to attract and retain talent, so that evidence of use may in fact be surprisingly easy to come by. What is more, when AI is used as or in the context of a medical device, the regulatory hurdles that need to be cleared will involve detailed disclosure of how the AI method works, to similar effect

In situations where it is appropriate to seek patent protection for an AI-related invention, the benefits can be significant. Secured patent rights can help cement a company’s market position and can significantly boost a company’s worth – especially for early-stage start-up companies relying on AI-related inventions that have not yet been turned into a product.

The balance as to whether to patent or not is more likely to favour patenting in fields where publication is the norm, either because it is demanded by talent or because it is a requirement of regulatory processes. On the other hand, where AI is used in internal processes, such as in supporting the identification of drug targets and candidate compounds or the evaluation of toxicity and other aspects relevant to drug development, the balance may favour strong trade secrets over patents.

AI patents – it’s all in the timing

The fast pace of developing, implementing, and publishing new AI approaches in the healthcare field combined with the need for patent applications to be filed before their contents are publicly disclosed, can result in AI patents being filed before their inventions are sufficiently refined. This can manifest itself in the form of methodological errors, lack of detail and/or the use of low quality or limited data to illustrate the benefits of an approach. Although not invariable fatal, this can result in applications being refused, patents being granted which are not commercially relevant, or granted patents which are more susceptible to revocation than they might otherwise have been. What is perhaps even worse, due to the way the patent system works, a prematurely filed poor application may scupper the prospects of protecting later high-value developments once all the details are worked out.

Often, delaying the filing of the patent application could have resulted in a better outcome. For example, the inventors could have added details to their methodology or confirmed their findings with higher quality datasets. The problem of premature filing is, of course, not restricted to the AI field. For example, in life sciences, data generation can be slow and expensive, tempting applicants to file too soon.

However, the push for premature filing is one that the AI healthcare field appears particularly susceptible to – mirroring concerns expressed by the scientific community about poor-quality, premature scientific publication in AI-based healthcare technology.

Deciding on the right time to file an AI application is seldom an exact science. Sometimes one’s hand is forced – for example when it becomes time to obtain regulatory approval or when a product launch looms. Otherwise, the urge to file and publish early – and so before competitors – needs to be weighed carefully against the need to make sure that the approach is near enough to being finalised and that, where needed, appropriate data can be included.

So, what about the data, you ask?

As is now widely understood, any AI technology is only ever going to be as good as the data that was used to feed it. This raises the question of how the data, possibly the most valuable ingredient of any one AI technology, can be protected.

Data sets cannot be protected by patents and indeed are not well served by other existing intellectual property rights, be that copyright or even the dedicated database right we have in the EU and in the UK (the reasons for that being beyond the scope of this article).

So, what is one to do? The answer is strong protective measures that limit access to the data only to those authorised to access it and strong contracts with external parties that require access to your data. Luckily, while the patent system cannot be used to protect data, it also does not require the disclosure of the data set used to obtain AI technology as part of the patenting process.

It is therefore perfectly feasible to protect a valuable data set by strong secrecy while seeking patent protection for the resulting AI technology.

Conclusion

Patenting AI healthcare inventions is possible and, often if not always, a powerful commercial tool.

We continue to see significant growth and development of AI in the healthcare space and Kilburn & Strode’s dedicated AI team, which spans the tech, healthcare, and life-science sectors, is well-placed to help navigate the challenges of obtaining protection in this complex and rapidly evolving field.