In early 2016, Google DeepMind unveiled the AlphaGo project, a gaming algorithm based on machine learning and dedicated to the last remaining game where humans held superior against computers—Go, the ancient Chinese board game.

The AlphaGo algorithm was “fed” a database of over 30 million moves, and was subject to reinforcement learning. This allowed the algorithm to optimize the search space of actions, reducing the required calculations to optimize AlphaGo’s next move. This is somewhat analogous to human intuition—the algorithm identifies the type of moves that doesn’t “seem” worthy of further consideration. The algorithm could thus devote computational resources towards the consideration of other “worthwhile” moves. Through such advancements in machine learning, AlphaGo defeated our human champion, the 9-dan Go professional Lee Sedol, by taking four out of five games. It was one small win for game algorithms, and one giant leap forward in the advancement of artificial intelligence.

Computers capable of mimicking human intuition with the capability to execute vast computational commands opens endless possibilities. Applications include complex medical diagnosis, weather pattern analysis, and even accurate election predictions. The question is no longer whether IT companies should invest in this technology—it’s how they will protect their innovations against competitors.

The Supreme Court’s 2014 ruling in Alice Corp. v. CLS Bank International has been considered a huge setback for the patentability of software inventions. Certain types of inventions are categorically excluded from the grant of patent rights – inventions directed to an abstract idea, natural phenomenon, or law of nature. In applying the Mayo two-step test on patentable subject matter, the Court ruled that Alice Corporation’s computerized method of exchanging financial obligations are not patent eligible under Section 101. In the view of the Court, such patents were directed to an abstract idea in finance, merely executed by a computer device. The biggest misfortune is the lack of guidance in the Alice decision on the type of software patents that would pass the patentable subject matter test. Software patents took a hit as subsequent district court decisions followed Alice down the rabbit hole.

The recent line of cases in the Federal Circuit provides the patent community with much needed clarification on the rules that govern patentability of software inventions. Enfish, LLC v. Microsoft Corporation, decided on March 2016, involved “model of data for a computer database explaining how the various elements of information are related to one another” for computer databases. On June 2016, the Federal Circuit decided another case on subject matter of software inventions. Bascom Global Internet Services, Inc. v. AT&T Mobility LLC was on a patent claiming an internet content filtering system located on a remote ISP server. Shortly after Bascom, the Federal Circuit decided McRO Inc. v. Bandai Namco Games America Inc. on September 2016. The case involved an automated 3D animation algorithm that renders graphics between two target facial expressions.

The Alice test follows two steps: (1) is the invention directed to an ineligible concept such as abstract idea, natural phenomenon, or law of nature, and if so, (2) does the elements of each claim, both individually and combined, transform this invention into a patent-eligible application. Enfish discusses what constitutes an abstract idea at the first step of the Alice inquiry. Judge Hughes instructs that one must look at whether the claims are directed towards a specific improvement or at abstract ideas. If the former, the patent is deemed to have solved an existing problem by a specific, non-generic improvement to computer functionality. The court then found that such inventions were patent-eligible.

McRO also ruled that the animation algorithm was not an abstract concept. Here, the court again emphasized that a patent may pass step 1 of Alice if claims “focus on a specific means or method that improves the relevant technology.” Additionally, the McRO court mentions preemption concerns—that the improper monopolization of “the basic tools of scientific and technological work” is a reason why there are carve outs against granting patents on abstract ideas.

Bascom provides guidelines on what would fail step 1—if the patent regards a conventional, well-known method, then the invention would be considered abstract. Cases that fail step 1 could then look to the “inventive concept test,” the second step of the Mayo/Alice test. In essence, this suggests that a patent application is not an abstract idea if (1) the invention addresses an existing problem by specific improvements rather than by conventional, well-known methods and (2) the claim does not raise preemption concerns.

In Bascom, the court decided in favor of allowing the patent under the second step of the Alice test. In doing so, the court ruled that even if elements of a claim are separately known in prior art, an inventive concept can be found in the non-conventional and non-generic arrangement of known, conventional pieces. This is a lenient standard compared to the obviousness inquiry, and hence it is unclear if we need this step for invalidating or rejecting a patent. Nonetheless, the court found that merely showing that the fact that all elements of a claim were already disclosed in prior art was not sufficient reason to make an invention patent ineligible.

While recent cases have provided insight regarding subject matter, we will have to wait for subsequent cases to firmly draw the line on what constitutes patent eligible subject matter. There are many intriguing questions that may require more guidance from the courts. For instance, can deep learning methods be considered specific improvements when such improvements are extracted from conventional data? Can the concept of machine learning be too broad to avoid preemption concerns? Until we have more cases, we can only continue to try our best to climb out of the Alice rabbit hole.

Hyunjong Ryan Jin is a J.D. candidate, 2018, at NYU School of Law.