Artificial intelligence (AI) encompasses technologies that have become increasingly relevant — and strategically important — to companies of all sizes across a wide range of fields. For example, AI technologies such as computer vision are critical to the operation of autonomous vehicles and pick-and-place robots. Natural-language processing, speech recognition, and context awareness are core capabilities of any product or automated service that communicates by voice with end users. As more players crowd the AI space, aggressively and efficiently developing AI patent portfolios is an important consideration for start-ups and multinational corporations alike.
In the U.S., most AI technologies can be protected by a patent. But certain AI technologies can face increased scrutiny at the U.S. Patent and Trademark Office (USPTO) with respect to whether the invention is directed to patent-eligible subject matter. The USPTO follows a two-step analysis to determine whether a patent claim is patent-eligible. First, the USPTO determines whether the patent claim is directed to a patent-eligible concept. Certain areas are not patent-eligible: abstract ideas, laws of nature, and natural phenomena. If a patent claim is directed to one of these concepts, the USPTO will next examine whether the claim as a whole amounts to “significantly more” than the aforementioned concepts, in which case the claim may still be patent-eligible.
An AI-based invention related to autonomous vehicles or robots that aims to control or manipulate a tangible object (e.g., a vehicle, a package) generally faces relatively minimal scrutiny. Such technology is usually considered patent-eligible because it is considered to produce a tangible result and thus is not abstract.
In contrast, an AI-based invention that is not directed to controlling tangible objects, such as a software algorithm, may face heightened scrutiny as to whether it is directed to an abstract idea. Nevertheless, many technical aspects of such an AI-based invention may still be patent-eligible. While the USPTO has not provided an explicit definition of an “abstract idea,” a number of court decisions have been illustrative. For example, the Federal Circuit has held that patent claims directed to a specific, self-referential arrangement of data is patent-eligible, stating in particular that “the self-referential table recited in the claims on appeal is a specific type of data structure designed to improve the way a computer stores and retrieves data in memory.” Thus far, recitations of specific data structures, specific rules, specific combinations of steps, or specific hardware configurations that result in improvement of the functioning of a computer have been found to cover eligible subject matter, while recitations of mere high-level user scenarios or use cases implemented using a general purpose computer are often found to be ineligible subject matter.
Thus, when identifying subject matter suitable for patent protection, a company developing technology in the AI space should look beyond the high-level user scenarios enabled by its AI technologies (e.g., a method of using conventional AI technologies to solve a general problem) and identify the unique technical features of its AI pipeline that improve the functioning of a computer. These technical features can include:
The pre-processing of training data
(e.g., data structure/taxonomy)
The training process
(e.g., topology of a neural network, configuration of parameters, termination conditions)
The use of trained classifiers or solutions
(e.g., sequence in which classifiers are used, use of solution space of a genetic algorithm)
The end-to-end workflow
(e.g., user interfaces, control logic)
(e.g., integration of AI algorithms with hardware components, hardware acceleration)
In summary, when seeking patent protection, a company in the AI space needs to focus on the aspects of its technology that produce tangible results or concretely improve the functioning of a computer system. Once such technical features have been identified, the key considerations shift toward determining how these features diverge from conventional technologies and to what extent these features can be generalized.
The above approach often requires collaboration among engineers, product managers, and in-house and outside patent counsel. Based on our experience, this interdisciplinary approach is essential for establishing strong and valuable patent portfolios.
 See Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016).
 See, e.g., McRO Inc. v. Bandai Namco Games Am. Inc., 837 F.3d 1299 (Fed. Cir. 2016); DDR Holdings, LLC v. Hotels.com, L.P., 773 F.3d 1245 (Fed. Cir. 2014); Thales Visionix, Inc. v. United States, 850 F.3d 1343 (Fed. Cir. 2017). See also FairWarning IP, LLC v. Iatric Systems, Inc., 839 F.3d 1089 (Fed. Cir. 2016). In FairWarning IP, LLC v. Iatric Systems, Inc., the Federal Circuit held that a claimed method of detecting improper access of a patient’s information is directed to an abstract idea despite the recitation of a “rule.” In particular, the court stated that “[the] claims merely implement an old practice in a new environment” and that the recited rules ask “the same questions (though perhaps phrased with different words) that humans in analogous situations detecting fraud have asked for decades, if not centuries.”