AI and machine-learning patentability: a strategic perspective

Companies innovating in AI and machine learning have a vested interest in the limitations and requirements for patenting their technology. This has led the EPO and the USPTO to issue guidelines clarifying what constitutes a patentable invention in this field.

AI is distinct from software patenting, despite often being seen as an extension of it. This is because it often forms a core part of an invention in its own right; perhaps more so than has ever been the case for algorithms in general.

EPO requirements: general software patents

Probably the most important terminology to consider is the technical character of an invention. This is repeatedly referenced in EPO guidelines and was the focus of a recent clarification that presented examples of mathematical methods or software, which contribute to the technical means by which a problem is solved, including:

• controlling a specific technical system or process;
• determining the required number of passes in a compaction machine to achieve a desired material density; and
• separating sources in speech signals and speech recognition.

An algorithm may therefore be eligible if it is used as part of an invention that performs a technical function.

AI implications

Consequently, the EPO seems more likely to grant patents to AI and machine-learning inventions that interact with images, machines or other external factors as opposed to algorithms that simply sort or tag data. The problem solved must also be technical.

In May 2018 the EPO reaffirmed the question that it asks of AI and machine-learning inventions: “Do(es) the AI and machine-learning method (steps) contribute to the technical character of the invention?” It is recommended that applicants include as much information about technical effects as possible in their application.

US requirements: eligibility criteria

The USPTO has issued no explicit guidance in these areas, although some direction can be attained from examining explicit exceptions to laws of nature, physical phenomena and abstract ideas.

An example of the abstract ideas exception is PurePredictive, Inc v H20 AI, Inc, where the defendant’s pure-data application of machine learning was deemed to be too abstract for patent protection. The judgment stated that the invention makes “use of computers only as tools, rather than provide a specific improvement on a computer-related technology”. Since in this case (as in many others) the question of software patentability rests heavily on the “abstract ideas” exception, the USPTO released guidelines in January 2019 in order to clarify this exception with regard to software. These provide guidance on abstract ideas following Alice Corp v CLS Bank, dividing them into three categories:

• mathematical concepts;
• methods of organising human activity; and
• mental processes.

The mental processes exception is a main focus. One example invention included a ranking process and noted that there was no need for a computer to do the ranking – a human could conceivably do it mentally. Thus, the invention was deemed patent ineligible.

However, if the invention provides an output (ie, anything that means that the process could not be carried out mentally) it is eligible.

The involvement of a mental process does not automatically preclude patentability – examiners are directed to consider whether the mental process is “integrated into a practical application”. An example where this does not hold was when an invention comprised a process for solving a problem (which could be framed as a mental process) but it happened to be the case that the problem was solved by a computer. It was therefore not enough to claim that a mental process was part of a practical application by just saying “it is performed by a computer”.

Actions to consider

From a strategic perspective, it is vital to keep abreast of updates and new information from the EPO and USPTO in this area. Small and medium-sized entities can expect ongoing interest from big tech companies and should be mindful of their freedom to operate given the large portfolios of some of these (especially as they approach significant funding rounds, trade sales or an initial public offering). Prior art searches are vital, especially as AI and machine learning are not always explicitly mentioned in patent claims but may fall under generalised terms such as methods involving analysing, inferring or classifying. The patent landscape can provide invaluable information. First, it highlights as-yet untapped areas of applications for AI and machine learning that could be exploited in future filings. Second, it shines a light on patents that are being granted. Examining successful patents may help with the construction and scope of future applications. The results highlight important distinctions in different jurisdictions and clarify the difference between a narrow, low-value patent and a high-value asset that protects a company’s core technology.