Generally, artificial intelligence (AI) is an automation of a thing that a human being can do, or the simulation of intelligent human behavior by a machine† In other words, AI performs what a human can but with vastly more data and processing of incoming information. Unfortunately, claiming AI in adherence to its typical definition is akin to asking for a Section 101 subject matter eligibility rejection in the United States. Europe and China have already updated their patent examination procedures for AI. If the United States sustains its current examination procedure of machine intelligence in accordance with the abstract idea doctrine under the Alice and mayo framework established by the Supreme Court, will we be leaving this industry behind?
AI is an umbrella term that encompasses four main categories: reactive AI, limited memory AI, theory of mind AI and self-aware AI. Reactive AI includes machines that operate solely based on the present data inputted; its decisions take into account only the current situation. Reactive AI don’t make inferences based on the data inputted. Examples of reactive AI include spam filters, Netflix show recommendations and computer chess players.
Limited memory AI is capable of making decisions based on data from the recent past and is capable of improving its decision-making processes over time. This is the category in which the vast majority of research and development and patenting is taking place. Examples of limited memory AI include autonomous vehicles capable of interpreting data received from the environment and making automatic adjustments to behavior when necessary.
The machines grow more prescient in the next two categories – these include AI that can understand human emotions and make decisions based on that understanding in theory of mind AI. Even more futuristic are self-aware AI–these machines are capable of processing the mental states of others and emotions, as well as having their own. Think about the robots in Wall-E or, more darkly, in I Robot.
When the original patent laws were drafted, lawmakers did not anticipate that one day we might have machines with decision-making capabilities that would mirror that of humans. As a result, the United States Patent and Trademark Office (USPTO), the European Patent Office (EPO) and the China National Intellectual Property Administration (CNIPA) all have subject matter eligibility restrictions with respect to mental processes and patenting tasks that a human can perform, particularly what a human mind can perform, which includes processing information and data and making decisions based on said information and data. The idea is to prevent the patent system from being abused in this way. But, in view of the new technologies emerging in the field of AI, each of these offices have attempted to update their examination procedure to at best try to capture some of this subject matter.
The CNIPA prohibits patenting methods for mental activities. Recently, the CNIPA issued Draft Examination Guidelines on examining inventions related to the improvement of algorithms for artificial intelligence (such as deep learning, classification and clustering and big data processing). When looking for a “technical solution” that can render machine intelligence patentable, the CNIPA proposes looking at improvements to algorithms and big data processing, whether the algorithms have a specific technical relationship with the internal structure of the computer system, and/or improvements to hardware computing efficiency or execution effect. The CNIPA considers improvements to data storage size, data transmission rate and hardware processing speed as evidence of a technical solution required for patentability.
In March of this year, the EPAs 2022 Guidelines for Examination came into effect, which state explicitly, “[a] mathematical method may contribute to the technical character of an invention, ie contribute to producing a technical effect that serves a technical purpose, by its application to a field of technology and/or by being adapted to a specific technical implementation.” The EPO is going so far as to explicitly state that mathematical formulas can be patentable if used in specific technical implementation. Specific examples of improvements to technical effect include efficient use of computer storage capacity or network bandwidth. EPO has published a series of examples of mathematical formulas that contribute to a showing of technical effect.
The USPTO issued its latest Guidance on examination back in 2019. The Guidance heavily emphasized technical improvements to a machine or functioning of a machine for overcoming a subject matter ineligible directed to abstract idea rejection. Notably, technical improvements in the US basically exclude end user benefits, which is different from the new CNIPA and EPO practice which allows user benefits to be a consideration of technical effect. Also unique to the United States is our Supreme Court, which occasionally intervenes on patent matters, particularly with the Alice and mayo decisions, which supersede any type of USPTO guidance. The USPTO Guidance was constructed within the confines of the abstract idea/law of nature framework of Alice and mayo, so it was unable to go as far as what the CNIPA and the EPO Guidelines have done in designating mathematical formulas to be patentable when implemented by a machine, and designating big data processing and improvements to hardware processing speed to be patentable. So, in terms of examination procedure for machine processes and machine intelligence, we are unfortunately a bit behind.
To learn more, watch the latest installment of IP Practice Vlogs, available here†