AI-generated inventions do not exist, at least not in the eyes of the law. Earlier this year, the U.S. Patent and Trademark Office (USPTO) rightly rejected two patent applications that named an AI system as inventor making clear its position to limit inventorship only to human beings, even when it is an AI system that is functionally inventing.
The AI inventor at the heart of this case is DABUS, an AI system that mimics the neural patterns of the human mind to independently combine basic concepts into a complex idea that it can self-identify as original. Stephen Thaler, who created DABUS, listed the AI system as the inventor of two ideas; interlocking food containers that are easy for robots to grasp and a warning light that flashes in a rhythm that is hard to ignore.
Thaler argues that the law should recognize DABUS, rather than him, as the inventor of these ideas because crediting a human with work they did not wholly invent devalues traditional human inventorship. But Thaler is wrong. The role of the patent system is to protect an inventor’s economic rights, not their moral rights. If anything, Thaler’s proposed changes do little to promote innovation—the primary goal of patent law—and instead introduce a legally unpunishable inventor that threatens, not protects, human inventors.
The main point Thaler misses is that the value of protecting new work is for a patent owner rather than an inventor, meaning it makes no difference who creates this value. To see this, consider Alice who works for a company Acme Labs. Alice’s contract, like most employee contracts, dictates that all intellectual property she invents is the property of her company. Suppose now, that Alice creates and trains an AI system that independently invents ideas. As ITIF explains in its comments to the USPTO, AI-generated works that add value to society should certainly receive the same patent protections as human inventions. But Acme’s economic return on their investment in Alice is the same whether she invents something herself or creates an AI inventor. This means an employee like Alice should get the same salary, the same bonus, and the same promotion in either case.
The question then becomes: If it makes no difference to the patent system, why not list DABUS as inventor? The reason is, if USPTO recognizes AI as an inventor, then AI is being treated as a legal person which is problematic in the context of patent law. The government cannot effectively hold AI systems, unlike other legal persons such as individuals or corporations, directly accountable if they illegally infringe on the IP rights of others. How can the government subject an AI system to a fair trial or fine it? The government could of course hold the owners and operators of AI systems accountable, but if patent law begins treating AI systems as separate legal persons, then the owners and operators of these systems could plausibly argue the government should not hold them responsible for illegal acts committed by these systems since they aren’t responsible for these systems when they engage in legal activity (i.e. invention).
Treating an AI system like DABUS as a natural person is an absurd concept. Even if USPTO recognized DABUS as an inventor, AI systems cannot independently decide to license their inventions to others, execute contracts, or collect payments.
It is a well-judged move for USPTO, as well as the EU and U.K. patent offices, to reject Thaler’s application on the basis that it does not name a human inventor, because permitting an entity incapable of functioning as a legal person in the system only serves to inhibit the system from doing its job: Protecting economic rights and promoting growth.
What is more, Thaler also argues it is unfair to place human creators of inventive AI systems on par with traditional human inventors because only the latter are true inventors. He suggests that placing them on equal footing would be so emotionally discouraging to traditional human inventors that they would lose the will to invent. But even if patent law were concerned with the emotional impact, rather than the economic impact, on human invention, there is little evidence to support Thaler’s view. In his book, The Right to Employee Inventions in Patent Law, Kazuhide Odaki explores the main motivators to invent using data from a study by the European Commission that surveyed more than 10,000 inventors from 2008 to 2011 in Europe, the United States, and Japan. It revealed that employee inventors are most intrinsically driven by the desire to solve practical problems, to be intellectually challenged, and to create value for their firms. These motivations are completely independent of other patent arrangements and who is or is not listed on them. Not only does it make little sense to change the patent system—an economic tool—to credit AI inventors for “moral” reasons, there is little grounding that the proposed changes would be effective on an individual or societal level.
Proposals to treat AI systems like natural persons simply reflects the dystopian fantasy that AI systems are developing autonomous intelligence or even consciousness. They are not. AI is closer to being a spreadsheet than it is to Star Trek’s android Data. AI is still just code—governments do not issue patents to electron microscopes because they help an inventor, and they should not give patents to other tools, regardless of how advanced they are.
The patent system is not perfect; patent scope, breadth, and length should certainly be reexamined as inventive AI brings new speeds to the traditionally slow-moving inventive process. But on the whole, the system works fine as it is to stimulate economic growth and protect human inventors. Until that changes the system is best left as it is.