Beyond the traditional practice of litigation, opinion work and proceeding before the U.S. Patent and Trademark Office, attorneys are more frequently being engaged by clients confronting the impact of artificial intelligence technology on businesses. Unlike many intellectual property issues that percolate up from the lab bench to the C-suite, interest in developing a strategy for addressing AI is being driven at the highest corporate levels. A panel at the 2017 World Economic Forum in Davos-Klosters, Switzerland, for instance, focused on AI as a disruptive technology that will drive productivity as it continues to make its way into enterprise systems and computing platforms.1

AI in technology in one form or another is an increasingly relied upon tool for conducting business. According to Forbes, the AI market will grow from $8 billion in 2016 to more than $47 billion in 2020. Reports put current AI penetration in businesses at 38 percent, and its adoption is predicted to grow to 62 percent by 2018. This is precipitated by "a greater than 300 percent increase in investment in artificial intelligence in 2017 compared with 2016."2

AI IP issues typically stem from two business objectives: maintaining a "freedom to operate," or FTO, without violating third-party IP rights, and protecting investments in AI research and development. Businesses that incorporate AI technology as part of product or service offerings should ascertain the scope of the IP landscape to respect the boundaries of potential third-party claims that could place them at risk. In addition, businesses are increasingly interested in protecting their investments in the development of AI IP. Due to low cost, high-capacity storage and computing power, and the ubiquity of sensors that capture data of all types, companies are adding AI features to existing products and creating entirely new product offerings based in AI. The world of "big data" has created both the availability of robust training sets used to develop AI technology and a need for technology that can process and filter large volumes of data for business applications. Recognizing the need to protect the value of their investment in AI, companies are increasingly securing IP protection. The PTO, for example, has seen a 500 percent increase in the past five years in the number of patents issuing to class 706, a classification exclusively designated for AI data processing systems.

At their root, both an FTO analysis and IP protection strategy invoke the same exercise: Determining the proper scope of what has been protected by others or what products stemming from a company's innovative efforts can be protected. At this level of abstraction, there is no discernible difference between this analysis in the AI context and the analysis applicable to traditional technologies. Beneath the surface, however, there are several significant and evolving issues implicating patent, trade secret and copyright law.

Inventorship Issues Arising from AI Technology

Under U.S. law, inventorship is the first point of analysis for determining ownership of IP. Identifying what contributed to the development of an AI-related patent for the purposes of determining whether someone was an "inventor" will probably happen more frequently. Although drawing the inventorship line may be complicated, the legal analysis substantially follows the legal touchpoints currently applied to other complex technologies. As AI develops, however, the patent bar may be confronted with another type of inventorship analysis that may be outside of the scope of current U.S. law. Currently, inventors are individuals. But what if an AI-enabled machine invents something? What if an AI algorithm—without any human intervention—develops a new drug, a method of recognizing diseases in medical images, or a new blade shape for a turbine? Section 100(f) of the Patent Act, 35 U.S.C.A. § 100(f) defines "inventor." The legislative history of that section indicates that Congress intended statutory subject matter to "include anything under the sun that is made by man," according to the U.S. Supreme Court.3 Accordingly, perhaps Congress, and not the courts, may have to make changes to existing patent law to address potentially patentable subject matter developed autonomously by AI.

Patent Disclosure Issues Relating to AI

When it comes to seeking patent protection for AI-based inventions, satisfying disclosure requirements can present challenges. Underlying the patent laws is a quid pro quo. In exchange for a limited monopoly via a grant to exclude others from practicing the claimed invention, an inventor must disclose to the public enough information about the invention to enable one of ordinary skill in the art to practice what is claimed.

Given the nature of some AI inventions, meeting this requirement can be challenging. For example, when seeking protection for rule-based AI systems, a research team may have developed rule sets that are effective for a specific application. Patent claims directed to a broader scope of application may not be enabled by the rules developed. Disclosing only those specific rules may not satisfy the disclosure obligations of Section 112 of the Patent Act, 35 U.S.C.A. § 112. Similarly, the performance of AI embodied in artificial neural networks can depend on network topology, which can include the number and types of layers, the number of neurons per layer, neuron properties, training algorithms and training data sets. The scope of the claims will depend on what the limited set of topologies disclosed in the patent teaches one skilled in the art to practice. In both the rule-based and network-based systems described above, where the systems have been developed heuristically, there may be questions regarding whether the patent discloses generalizations necessary to support the desired claim scope. There could be millions of permutations of the network architecture or rules adaptable for various applications. Disclosing only a few and trying to define a broad claim scope may introduce risks. Providing a comprehensive disclosure laying out many embodiments may reduce some risk. But practically,how many can and should be disclosed? This is an area where guidance may come from the pharmaceutical arts, which may aid in an understanding of the bounds of patent disclosure and written description requirements.

AI Claiming Strategies to Address Disclosure Requirements

Certain types of AI technology can be defined by identifying building block functions. For example, elements may be claimed as "means for classifying" or "means for responding to backpropagation learning." In additional to including in a patent application claim sets that recite specific structural details of an AI invention, alternative claim sets that define claim boundaries by the functions that the elements perform could be beneficial. Section 112(f) authorizes this type of functional claiming.

While functional claiming may have some strategic advantages, it does not entitle an inventor to claim elements functionally with the expectation of including all structure for performing the functions claimed.4 An inventor is entitled only to the structure that is disclosed for performing the claimed function and equivalents to what is disclosed. Also beware of expecting to satisfy disclosure requirements with "black box" schematics in a patent application with the expectation that one skilled in the art would be able to fill in the blanks. Recent court decisions have raised the bar by requiring more details in some situations. Courts often describe functions other than those commonly known in the art as requiring "special programming" for a general purpose computer. These require disclosure of the algorithm for performing the claimed function.5 Functions known by those with ordinary skill in the art as being commonly performed by a general purpose computer or computer component such as a "means for storing data," do not require disclosure of additional supporting structures.

Accordingly, the ability to claim functionally does not obviate the need to disclose embodiments in the AI context.

Patent-Eligible Subject Matter and IA

For companies developing and seeking to protect their investments in AI innovation through IP, the current status of the law presents several hurdles. One of the fundamental challenges with respect to protecting AI technology with patents involves claiming subject matter that is patent eligible.

Under Section 101 of the Patent Act, 35 U.S.C.A. § 101, the subject matter of a patent claim must be directed to a "process, machine, manufacture or composition of matter." However, the U.S. Supreme Court held in Diamond v. Diehr, 450 U.S. 175 (1981), that claims directed to nothing more than an abstract idea, such as a mathematical algorithm, or to natural phenomena or a law of nature are not eligible for patent protection. The technology underlying AI is generally based on computer programming or hardware implementing mathematical models, deep learning algorithms or a neural network. An improperly drafted patent application directed to AI may fall within this judicially recognized exception to patent-eligible subject matter.

In Alice Corp. v. CLS Bank International, 134 S. Ct. 2347 (2014), the Supreme Court provided the framework for determining "whether the claims at issue are directed to a patent-ineligible concept." If the claims are, then the elements of all claims must be examined "to determine whether [they contain] an 'inventive concept' sufficient to 'transform' the claimed abstract idea into a patent-eligible application."

The PTO expressly recognizes that AI can be patentable through the express designation of class 706, a section of the agency's patent application classification system.6 In addition, two PTO "examining art units" for reviewing prior art are specifically devoted to reviewing applications directed toward AI algorithms.7 While the number of AI patents and patent applications filed in the U.S. over the past several years has grown exponentially, obtaining such a patent presents a unique set of challenges in view of Alice. Patent examiners have rejected claims directed to AI algorithms under Section 101 on the basis that the concept claimed is a certain method of human activity and is similar to claims that courts have deemed an abstract idea. Because the goal of AI is often to replicate human activity, the challenge practitioners face is rooted in how to claim AI to make it patent eligible.

Strategies for Claiming AI as Patent-Eligible Subject Matter

A key component to protecting AI investments with patents is to claim the AI in a manner that transforms the abstract idea into patent-eligible subject matter. Several recent cases provide guidance regarding how to claim patent-eligible subject matter.

One strategy is to claim the application or the use of data, not just the generation of data. For example, in Thales Visionix Inc. v. United States, 850 F.3d 1343 (Fed. Cir. 2017), the patent at issue claimed a technique for positioning inertial sensors in a particular configuration and using the raw data from the sensors to more efficiently and accurately calculate the position and orientation of an object moving on a platform. Because the patentee sought protection of the application of physics and the novel configuration of the sensors rather than the mathematical equations used to make the calculations, the U.S. Court of Appeals for the Federal Circuit found these claims contained patent-eligible subject matter.

In contrast, the Federal Circuit in Vehicle Intelligence and  Safety LLC v. Mercedes-Benz USA LLC, 636 Fed. App. 914 (Fed. Cir. 2015), reached a different result. In that case, the patent contained claims reciting the use of an undefined expert system without providing a particular use or application of the system or any details as to how the system produced faster, more accurate and reliable results. The Federal Circuit found an inventive concept was lacking.

Another approach practitioners can use is to provide details in the claims. U.S. Patent No. 9,569,726, which lists Microsoft as the exclusive assignee, is titled a "server computing device for recommending meeting a friend at a service location." This device is listed as a class 706 patent and is an example of when an inventor provided details in a claim. That claim recites in part:

A server computing device for providing recommendations to a user computing device, the server computing device comprising:... receive friend activity of a friend using a friend computing device, the friend activity including a detected current location of and direction of travel of the friend computing device and calendar activity of the friend, and to receive from the user computing device a request for a recommendation for a target product or service.

Although providing details in the claim can help avoid abstraction, the Federal Circuit noted in Bascom Global Internet Services Inc. v. AT&T Mobility LLC, 827 F.3d 1341 (Fed. Cir. 2016), that doing so can severely narrow the scope of protection. Therefore, when considering this approach, an analysis of whether to pursue patent protection should be conducted. If the claim scope is drawn too narrowly, it may be more suitable to seek protection through other forms of IP, bykeeping the AI a trade secret, for instance.

Because the core of AI technology often consists of some level of computer programming, claiming AI using computer-readable medium claims is another potential claim-drafting strategy. CRM claims offer the benefit of having properties of both an apparatus and method claim. This is because they take the form of computer-readable medium storing instructions that, when executed by a computer, cause it to perform a specified method. While there was concern about whether CRM claims would survive Alice, the PTO expressly "endorsed CRM claims as patent-eligible by listing CRM claims in its post-Alice  Section 101 guidelines on patent-eligible subject matter."8

Trade Secret Protection for AI

Not all patent applications result in a granted patent. Furthermore, a patent application that focuses on reducing the risk of inadequately disclosing an invention may unnecessarily disclose valuable trade secrets, even if that patent is granted. When a patent application is published, it discloses to the public—including competitors—all of the previously proprietary details contained in the application. If everything disclosed is not protected by claims that are ultimately granted, or if no claims are ever granted, the inventors will have disclosed to competitors potentially valuable research in return for nothing. Even if the PTO ultimately grants a patent that covers some or all of the disclosed technology, during the post-publication, pre-grant period the information may reach competitors with no recourse available to the inventor. Moreover, this period can last for a few years. Keep in mind also that a patent grant provides a right against patented activity only in the jurisdiction granting the patent. For example, a U.S.-issued patent does not stop a Russian competitor from taking the information disclosed in the patent and practicing that invention in Russia. This shortcoming highlights the potential advantages of trade secret protection over patent protection for AI inventions.

The practice of protecting AI inventions as trade secrets offers the advantage of avoiding a need for disclosure. In fact, proprietary technology remains a trade secret only as long as it is not publicly disclosed. As a consequence, trade secret protection can last longer than patent protection, which generally has a 20-year term. Trade secrets do not require governmental approval, and there is no application or examination process—and consequently no prosecution costs or application fees. The only cause of action trade secret status provides is misappropriation. Unlike patent protection, there is no cause of action against a competitor that independently develops technology that is a protected trade secret or ascertains it by reverse engineering it from products in the public domain.9

Trade secret protection may be particularly well-suited for rapidly developing and changing AI inventions, where refinements and improvements are fluid. When inventors rely on trade secrets to protect their AI inventions, they do not need to determine when an invention is ready for patenting or deal with a continuous evaluation of what has been submitted to the patent office. Nor do they need to concern themselves with whether the newest evolution is covered by existing filings or whether additional applications or claims should be filed.

Copyright Protection for AI

Copyrights can be used as another form of protecting AI, because AI software can be copyrightable. In Synopsys Inc. v. ATopTech Inc., No. 13-cv-2965, 2013 WL 5770542 (N.D. Cal. Oct. 24, 2013), Synopsys had patents directed to static timing analysis but instead relied exclusively on its copyrights of the software to secure a jury award of over $30 million based on ATopTech's alleged infringement of Synopsys' copyright.

Whether AI that is capable of generating copyrightable material can obtain a copyright is a different matter. A district court recently found that a monkey had no rights to his selfie because the current copyright statute as interpreted affords rights to humans, not animals.10 This case demonstrates that future legislation would likely be required to allow animals, or AI for that matter, to obtain copyright protection.

AI and the IP issues it presents are continuing to evolve, creating a new frontier for businesses. Companies will need to consider changes in the law to employ the appropriate legal strategies to guide them as they deploy and protect AI-based innovations.

Footnotes

1 World Economic Forum, World Economic Forum Annual Meeting 2017 System Initiatives Programme (2017), www3.weforum.org/docs/Media/AM17/AM17_System_Initiatives.pdf.

2 Gil Press, Top 10 Hot Artificial Intelligence (AI) Technologies, Forbes (Jan. 23, 2017, 9:09 AM), www.forbes.com/sites/gilpress/2017/01/23/top-10-hot-artificial-intelligence-ai-technologies/#63abf3c91928.

3 Diamond v. Chakrabarty, 447 U.S. 303 (1980).

4 Eric P. Raciti, Means Plus Function Claiming: What Does It Mean to Be a Means, When Are Means Means, and Other Meaningful Questions, Finnegan (March/April 2016), www.finnegan.com/resources/articles/articlesdetail.aspx?news=83016cea-2d7f-4406-a247-6ea4a925a7f3.

5 Dev Batta, Fed. Circ. Guidance for Means-Plus-Function Software Claims, Law360 (May 22, 2015, 10:33 AM), www.law360.com/articles/657462/fed-circ-guidance-for-means-plus-function-software-claims.

6 Class 706: Data Processing- Artificial Intelligence, U.S. Patent & Trademark Office (June 30, 2000), www.uspto.gov/web/offices/ac/ido/oeip/taf/def/706.htm.

7 Classes Arranged by Art Unit: Art Units 1764-2691, U.S. Patent & Trademark Off ice (Nov. 15, 2012), www.uspto.gov/patents-application-process/classes-arranged-art-unit-art-units-1764-2691 (showing that art units 2121 and 2129 involve artificial intelligence).

8 Jason E. Stach & James D. Stein, The Computer-Readable Medium Claim: The Best of the Apparatus and Method Worlds, Finnegan (Sept.-Oct. 2015), www.finnegan.com/resources/articles/articlesdetail.aspx?news=7183d349-efe8-4e41-85fc-c401913e8d7f.

9 Trade Secret Policy, U.S. Patent & Trademark Off ice (May 12, 2017) www.uspto.gov/patents-getting-started/international-protection/trade-secret-policy.

10 Naruto v. Slater, No. 15-cv-4324, 2016 WL 362231 (N.D. Cal. Jan. 28, 2016).

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.