USPTO, EPO, and Precedent
As some readers may recall, the increasing efficacy and ubiquity of artificial intelligence has instigated philosophical and legal debate concerning whether an artificial intelligence may itself be considered the inventor of an innovation it has generated. Granting a machine the status of an inventor would deviate from the traditional view that only a human being may be an inventor. The United States Patent and Trademark Office (USPTO) requested comment on various issues concerning patents of “AI inventions,” whether directed to or created using artificial intelligence. Among the questions the USPTO solicited was:
4. Should an entity or entities other than a natural person, or company to which a natural person assigns an invention, be able to own a patent on the AI invention?
At least one person appears to have resolved this question to his own satisfaction. Dr. Stephen Thaler has been filing patent applications in various patent offices around the world, including the USPTO and the European Patent Office (EPO), listing his artificial intelligence program “DABUS” as the inventor. In the latter two jurisdictions, Dr. Thaler’s electronic protégé has not yet met with success. The USPTO upheld an examiner’s requirement that Dr. Thaler name a “natural person,” rather than DABUS. The EPO has similarly required thus far that an inventor have “legal capacity.” Neither matter has been finally resolved, as the USPTO case has been advanced to the District Court for the Eastern District of Virginia, while the latter is set for oral arguments in December of 2021.
However, DABUS and Dr. Thaler have enjoyed one small victory in another state. On July 29, 2021, South Africa granted a patent naming DABUS as an inventor. It is worth noting that South Africa does not have a formal examination process for patents, which somewhat tempers the scope of this victory.
It is interesting to note how Dr. Thaler characterized DABUS to the USPTO; he referred to the machine as a “creativity machine” constructed of a number of neural networks. These networks were trained in a particular domain of knowledge, but not with training data directed to any specific problem within that domain. [Id.] Furthermore, the solution that the neural networks produced was not prompted by any specific problem that was presented to them; instead, the neural networks themselves are purported to have “recognized the novelty and salience of the instant invention.” If true, this suggests that DABUS is a somewhat intermediate category of artificial intelligence: many AI systems today are trained and designed to solve specific and well-defined problems, such as tuning the parameters of a control system, performing facial recognition through image classification, and other endeavors where the programmer intends a particular outcome and designs the system to achieve it. While the dream of a truly sentient “generalized” AI remains firmly in the realm of science fiction, a system like DABUS represents a step in that direction, as a program designed to look at a knowledge domain or problem space and identify both problems and solutions therein based on patterns the system has detected.
Problem-specific AI may be easy to analogize to other tools, such as netlist optimizers in hardware design. A hardware engineer may provide “register-level” hardware descriptions to a program that generates detailed circuit diagrams and placement thereof on an integrated circuit or a field programmable gate array; while the hardware engineer may not have conceived the particular details of the circuitry, few would argue the program was the inventor or the hardware engineer was not. On the other hand, difficult philosophical problems would arise if a machine that could be verified as conscious identified social needs for new technologies, chose domains of technology suitable for addressing such needs, and then designed a solution to address them. DABUS appears to be some step in that direction: while presumably not aware of a purpose for its innovation, it may be able to identify patterns representing technical challenges and solutions thereto. How much closer to generalized intelligence must a device like DABUS become before we must consider changing the inventorship doctrine? The ultimate rulings concerning the DABUS applications may begin to answer that question.