The future is now, and once-hypothetical questions about the ability of artificial intelligence to produce a patentable invention are no longer so.
Back in April, computer scientist Stephen Thaler’s creation, DABUS, was named as the inventor on a patent application which the USPTO rejected for lack of a human inventor. Earlier last month, he sued the USPTO to challenge this rejection, claiming that the USPTO’s rejection “is anti-intellectual property and anti-business, and it puts American businesses at an international disadvantage.”
I don’t think that Thaler has a leg to stand on. The Patent Act clearly defines “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.” (Oddly enough, the term “author” does not have such a clear definition in the Copyright Act, though the Copyright Office’s policy is to only register works created by a human being).
But let’s leave the law aside to ask a more serious question: why shouldn’t the USPTO issue a patent to something invented by an artificial intelligence?
This is undesirable for two reasons. First, since the AI is doing all of the “work” associated with the creation of new inventions but (unless we’re talking about Skynet) won’t be bothered one way or another with others free-riding off of his invention, the market failure has been addressed by not existing. If, as Thaler’s lawyers state “[a]llowing patents for AI-generated inventions will incentivize the development of AI capable of producing AI-generated inventions” then it makes much more sense to just give patents to the inventor of that AI itself.
I’m even skeptical that we need to give patents for such AI. If an AI can consistently generate patentable inventions, the inventor need only capture a vanishingly small share of the returns (as innovators generally do) to repay their investment, which could easily be done without legal exclusivity.
Second, an AI owner’s ability to hold the patent to the former’s invention has serious implications for inequality and competition. With all the promise of AI-generated inventions comes the risk that the first mover into the market will become so dominant that it will outpace teams of human inventors. What happens when the AI builds its own patent thicket? Or what happens when it patents other patentable-invention-generating AIs? These hypotheticals seem outlandish, but so did questions about the ability of AI to generate a patentable invention only a few years ago.
Of course, this inequality death spiral becomes impossible if, say, the government were to exercise eminent domain to purchase an inventor AI–or build one of its own–and have all the inventions dedicated to the public domain. Suddenly, the corporatist nightmare looks more like a techno-utopia.