New machine learning technologies may have far-reaching, unexpected effects on IP and patent law

Like most people, I expect that artificial intelligence and machine learning will have a host of near-term positive effects on pharma companies, most notably in accelerating the process of developing drugs and identifying the patients who are most likely to benefit from them. As AI enables pharma companies to sort through, analyse and contextualise preclinical and clinical results more quickly, we should see an increased pace of new drugs and therapies being developed to help patients.

Based on the US patent laws as they currently exist, drug companies should see a corresponding expansion of their patent portfolios, but policymakers will need to wrestle with the question of whether these new AI-assisted inventions are sufficiently non-obvious to merit patent protection. While such inventions may surpass what is obvious to a person having ordinary skill in the art, we may reach the point when these inventions cannot be said to be non-obvious to a person skilled in the art, and who has access to machine learning that collates, analyses, and draws conclusions at a rate beyond the capacity of any person.

Non-patented forms of intellectual property, such as software, trade secrets, and know-how, may become exposed to somewhat greater risk as AI and machine learning are used more broadly. For example, diagnostics companies often develop and rely on algorithms to process and analyse biospecimens, and identify biomarkers and disease states, and potentially connect the patient to a potential therapy. In many cases, these algorithms are protected as trade secrets but are not patented. In a pre-AI world, such trade secret protection was fairly secure as long as the company did not disclose the algorithm and limited access to raw data.  With the availability of AI, there is a risk that, even having taken typical precautions, the diagnostic company’s algorithm could be reverse engineered and exploited by the company’s competitors.

Lastly, it should be said that data will be the primary fuel for the engine of AI and machine learning.  As a result, I expect that data regarding patients, drugs, and disease states will become more valuable, and those with access to large stores of data (e.g., a hospital system with access to patient health histories, and EHRs) will be in position to command premium prices for access to this data. However, because mere data is generally not subject to patent protection, it may be challenging for data owners to monetise their data on an ongoing basis, because once the data becomes publicly available, it can generally be used by anyone, without a royalty obligation.  As a result, data owners should share their data only pursuant to thoughtfully-drafted confidentiality and data access agreements.

In these data access agreements, the data owner should insist on strong and clear covenants in which any data recipient would agree to exactly how it may and may not use the data. The data owner should consider the manner and duration for which each partner – including affiliates, consultants, and collaborators – will need to access and use the data, and draft covenants around those activities as well. Reverse engineering and other activities that may be facilitated through the use of AI should be specifically considered and addressed. Remedies for breach should be spelled out in the contract, including, where appropriate, injunctive relief and the availability of consequential damages.

Of course, contractual covenants may not provide complete relief to an entity whose proprietary data has been made publicly available, so data owners should also take care to limit disclosure to trustworthy, reputable recipients who have implemented adequate data protection procedures.

Cullen Taylor is life sciences partner, Northern Virginia, at Hogan Lovells