Insight
February 25, 2021

Five Tips For Life Sciences Companies To Protect Their AI Technologies

Introduction

Artificial intelligence (AI) has revolutionized many technology areas. As a few examples, it has already been instrumental in improving and enabling voice recognition algorithms, digital assistants, advertisement recommendation engines and financial trading applications.[1] Significant investment is being made for further development of this promising new technology, with R&D spending on AI predicted to reach $57.6 billion by the end of 2021.[2] Along with these R&D efforts, companies are also trying to protect and monetize their AI inventions, in some cases opting to seek patent protection. From 2002 to 2018, the number of AI patent applications filed with the United States Patent and Trademark Office (USPTO) more than doubled, from 30,000 to 60,000.[3]

These R&D efforts are no longer limited to software companies. Life sciences companies are making significant investments in AI technology as well. In 2019, over 60% of life sciences companies invested over $20 million into AI technologies.[4] AI and Machine Learning (ML) are being used to analyze data relating to the safety, quality and clinical effectiveness of certain treatments;[5] improve the methods of manufacturing for medical devices;[6] run in silico trials to find new drug candidates in lieu of costly in vitro trials;[7] and diagnose patients.[8]

Given that AI and ML – historically the domain of software companies – is a new frontier for life sciences companies, such companies may not be familiar with the various pitfalls and considerations that a company faces when trying to protect its AI inventions. We have assembled five helpful tips to consider:

Tip 1: Make sure you have permission to use the data

The lifeblood of most AI technologies is the data that is used to train the AI model so that the model can “learn” from the training data and generate useful predictions or inferences when presented with new data in future applications. In the life sciences context, this training data often includes or is derived from analysis of highly sensitive patient information, such as the patient’s personally identifiable information, medical history, and even their biological materials or DNA. Life sciences companies should be careful to ensure they receive proper permission to use the data they collect for purposes of training an AI model. Failure to receive proper permission could result in significant liability and other consequences associated with noncompliance with data privacy rules and regulations. Life sciences companies should familiarize themselves with the data privacy rules applicable to the types of data they are collecting and then develop a consent form, including all proper disclosures and terms, to be signed by all patients from whom they are receiving data. Standard consent forms used to collect patient samples for clinical trials or in other clinical settings often do not address the patient’s consent to use the samples to develop AI models for commercial use. The disclosures and terms will vary depending on the data collected. For example, if life sciences companies are collecting biological data from patients, the consent form should require that the samples be deidentified (e.g., identified using a unique code rather than personally identifiable information) and give the patient the option of permitting the company to use the information for research and commercial purposes while acknowledging no rights in any commercial value derived from the samples.

Tip 2: Get IP assignments from everyone contributing to the AI technology

Most life sciences companies understand that an assignment of intellectual property (IP) rights is required from any individual that contributes to the company’s core products and candidates. Failure to fully own and control the IP assets a company purports to own can be a big problem if discovered during diligence for a corporate transaction or enforcement proceeding. Historically, the individuals required to assign IP to life sciences companies are the scientists and engineers involved in the R&D associated with the core products and candidates. But for AI technologies, the universe of contributing individuals may be broader than expected. For example, any of the following individuals could, in certain situations, be considered the creator of an AI technology: the individuals that select the data to be acted on by an AI engine, the individuals that review the results or outputs of an AI engine, the individuals that select the ML algorithms used to train the AI model and tune the modeling parameters, and the individuals that write the source code to implement an AI engine, among others. If the IP rights are protectable by copyright (e.g., source code), then the work for hire doctrine may result in the IP rights automatically vesting in an employer. But the same is not true for independent contractors, where absent an agreement saying otherwise, copyright rights vest with the contractor. Moreover, the default rule for patent rights is that they vest with the inventor, regardless of status as an employee or contractor (again, absent an agreement to the contrary). Life sciences companies should be prudent about getting IP assignment agreements, ideally with present-tense assignment clauses, signed by all individuals, employees or contractors, that in any way interact with or are involved with designing or developing an AI technology.

Tip 3: Be careful when using open source software

One of the most convenient, but also most dangerous, parts about software development is there is a vast universe of software code made readily available on the internet, much of it for free under so-called “open source” licenses. Using this pre-existing and available software code is quite common and can significantly increase the efficiency of a software development project by avoiding the need to reinvent the wheel for every component and feature of a particular product, especially those that are conventional. But what is engrained in the DNA of most software companies that may not be readily apparent to life sciences companies entering the software development space for the first time is that a company must be careful about the terms on which it uses third-party software. Open source software is made available to the user under a license agreement, but unlike the more formal written and heavily negotiated license agreements a life sciences company may be familiar with (e.g., when out-licensing technology from a university), these licenses are less conspicuous and often assented to simply by making use of the software. Though they come in all shapes and flavors, open source licenses can generally be characterized into two groups: (1) permissive open source licenses, and (2) copyleft open source licenses. A permissive open source license (e.g., the MIT license) makes software code available for free to a user, but does not place significant restrictions on how the code must be used. Importantly, this means the user of code under a permissive open source license can combine the code with its own proprietary code and be under no obligation to disclose or license the combined code. Conversely, copyleft licenses (e.g., the General Public License (GPL)) also make software code available for free, but require that any modified code be licensed under the same terms. Therefore, if the copyleft licensed code is combined with proprietary code, the user may be required to make its proprietary code publicly available for free as well. Obviously, this is not a good outcome for a company desiring to keep its AI software secret. To avoid this negative outcome, companies should incorporate good hygiene around their use of open source software and implement policies and procedures to ensure that no source code is used that could jeopardize the secrecy of the company’s proprietary code.

Tip 4: Be thoughtful about the type of legal protection you want for your technology

There are several legal tools available to protect the IP associated with AI technology, but each tool has benefits and drawbacks, and some tools are better or worse suited depending on the technology and the company’s business strategy. A thoughtful analysis and identification of the technology to be protected, followed by selection of the appropriate tool(s) for legal protection, is critical to maximize protection and value of AI technology. In some cases, a company may select different strategies or tools for different aspects of their technology.

Generally, there are two legal IP protection tools that a company should consider using to protect its AI inventions from external competitors: patents and trade secrets (note that a third type of IP protection, copyright, will also protect against any direct copying of the source code implementing any AI technology, but will not protect the ideas underlying the source code). Patents and trade secrets each provide different benefits and limitations. Patents require that you prepare and file a patent application for examination by government patent offices. The application is typically published (and always published if it is granted as a patent) and must describe your technology in sufficient detail to enable others to be able to practice the invention. If you are granted a patent, you have the exclusive right to exclude others from practicing the invention, even if they invented the invention completely independently from you.[9] Trade secrets, on the other hand, do not require an examination process and do not require that you publicly disclose your invention. However, trade secrets provide no protection against an independent inventor or someone that reverse engineers your invention.[10]

Life sciences companies should consider the following factors when deciding between patent and trade secret protection:

  • Likelihood of independent invention. If it is likely a competitor will independently develop your AI invention, a patent is the best line of defense. If such independent development is unlikely, keeping the invention as a trade secret may be preferred.
  • Detectability of the invention. Even if you obtain a patent, in order to enforce it you need to be able to identify who among your competitors is using your patented invention. Given that many software inventions are implemented within non-public servers, this can be a significant consideration for AI inventions. However, often potential infringement can be detected based on the features, performance, etc. of a publicly available product. If competitor use of your technology can be detected, patents may be a good option; otherwise keeping the invention as a trade secret may be preferred.
  • Speed of innovation. The patent application process takes time. In some cases, it can be two to three years or more before the application is examined and granted (though this can sometimes be reduced to approximately one year if you pay for accelerated examination). On the other hand, once a patent is granted, it comes with exclusive rights for 20 years from the application’s filing date. But for some types of AI technologies, the technology is evolving so rapidly that by the time a patent application makes it way through examination, let alone by the end of a 20-year patent term, it is already obsolete. If you think an invention may become obsolete quickly, trade secret protection may be preferred.

Tip 5: If you choose patent protection, employ strategies to maximize chances of success

In the United States, inventions can be patented if they are (1) new and non-obvious over the prior art, and (2) directed to patent-eligible subject matter.[11] Even when AI-based inventions improve upon the state of the art in the life sciences and arise from substantial investments in cutting-edge R&D, their patent-eligibility is very carefully scrutinized by the USPTO. This is because AI-based inventions often intersect with subject matter that is not eligible for patenting, such as laws of nature, natural phenomena, or abstract ideas (collectively, “judicial exceptions”).[12] Inventions that are directed to these exceptions are eligible for patenting only if they amount to “significantly more” than the exceptions themselves.[13] While there are many ways to argue that an invention amounts to significantly more than a judicial exception,[14] one of the best approaches with AI-based inventions is to describe, in the patent application, the AI model’s performance and the improvement(s) over the performance of conventional techniques. Ideally, model performance and comparisons to conventional techniques can be shown using statistical data such as ROC curves, measures of positive predictive value (PPV) or negative predictive value (NPV), confusion matrices, F1 scores, and other similar data. The presence of such data in the application generally goes a long way toward showing that the invention is a patent-eligible improvement over the prior art rather than an ineligible attempt to monopolize a judicial exception. Employing these techniques greatly increases the likelihood of the USPTO granting a patent for an AI-based invention.


[1] Liana B. Baker, Tech moguls declare era of artificial intelligence, REUTERS (June 2, 2016, 9:06 PM), https://www.reuters.com/article/us-tech-ai-conference/tech-moguls-declare-era-of-artifical-intelligence-iduskcn0yp035.
[2] Research Topics – AI: The next generation of intelligence, IDC, https://www.idc.com/itexecutive/research/topics/ai (last visited Feb. 1, 2021). 
[3] USPTO, INVENTING AI: TRACING THE DIFFUSION OF ARTIFICIAL INTELLIGENCE WITH U.S. PATENTS 2 (2020).
[4] Aditya Kudumala et al., Scaling up AI Across the Life Sciences Value Chain: Enhancing R&D, Creating Efficiencies, and Increasing Impact, DELOITTE (Nov. 4, 2020), https://www2.deloitte.com/us/en/insights/industry/life-sciences/ai-and-pharma.html. 
[5] Lincoln Tsang et al., The Impact of Artificial Intelligence on Medical Innovation in the European Union and United States, 29 INTELL. PROP. & TECH. L.J., no. 8, Aug. 2017, at 3, 4.
[6] Id.
[7] David W. Opderbeck, Artificial Intelligence in Pharmaceuticals, Biologics, and Medical Devices: Present and Future Regulatory Models, 88 FORDHAM L. REV. 553, 566 (2019).
[8] Susan Y. Tull, Patenting the Future of Medicine: The Intersection of Patent Law and Artificial Intelligence in Medicine, 10 LANDSLIDE, no. 3, Jan./Feb. 2018, at 40, 41.
[9] See 35 U.S.C. § 271.
[10] Restatement (Third) of Unfair Competition § 43 (Am. Law. Inst. 1995).
[11] 35 U.S.C. § 101 et seq
[12] See Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208 (2014); Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66 (2012).
[13] See Alice Corp. Pty. Ltd. v. CLS Bank Int’l, 573 U.S. 208 (2014); Mayo Collaborative Servs. v. Prometheus Labs., Inc., 566 U.S. 66 (2012).
[14] See 2019 Revised Patent Subject Matter Eligibility Guidance, Federal Register vol. 84, no. 4 (Jan. 7, 2019).

Contributor Author
Kurt Hoppmann