Autonomous cars are driving the reinvention of IP protection
Darryl M. Woo
Lucas Dahlin is an associate in Goodwin’s Intellectual Property group. He focuses his practice on complex intellectual property matters, with a particular experience in patent and trade-secret litigation.
Julius Jefferson is an associate in Goodwin’s IP Litigation group. Prior to joining Goodwin, he clerked for judges of the District of Delaware and of the Eastern District of Texas. Before attending law school, Julius was a Research Fellow at Wyeth Pharmaceuticals (now Pfizer), where he worked on finding cures for Alzheimer’s disease.
Darryl M. Woo
Darryl Woo is a partner in Goodwin’s IP Litigation group. He is a veteran trial lawyer, concentrating his practice on patent litigation and other complex technology litigation.
Early 1900s society struggled to transition from horse-drawn buggies to automobiles. It may seem odd today, but there was a time when there was no concept of “right of way,” speed limits or traffic signs and signals. Those rules had to be invented so that cars encountering carriages would not frighten horses into runaways — or run each other off the traveled dirt paths that sufficed for “roads” — and require help from their four-legged counterparts to free them.
Fast-forward a hundred years. Quo Vadis is Latin for “whither goest thou,” an apt phrase for the free-ranging lifestyle enabled by the modern automobile. Come and go when you want, where you want and with whom you want. Take a Forrest Gump-inspired drive across the country in your dinosaur-fueled car or zap across town in your EV for a midnight snack. We have become a truly mobile society.
But crowded streets and highways full of distracted or inept drivers have turned “autopia” into “disautopia.” Enter the promise of the autonomous vehicle, enabled by artificial intelligence. All of the benefits of driving, without having to drive. Cars guide themselves in perfect harmony, easing congestion, promoting traffic flow and optimizing road utilization. Even better, you can sit back and let the car guide itself while you catch a TED talk and sip your latte.
Well, not quite. The state of technology at present can only account for, at best, a high percentage of driving situations, but not the marginal. The edge case/remaining 5-10 percent still depends on human intervention. Such cases may be benign, like the car not knowing what to do when it confronts an unplanned pothole repair crew. Or, a more Machiavellian example could involve human-operated vehicles approaching a stopped autonomous vehicle from 90 degrees at a four-way stop. Realizing that the AI vehicle would not endanger its occupants, the human drivers could serially execute “rolling stops,” effectively freezing the AI vehicle in place while the slow-rolling train of human-controlled cars continued unending during rush hour.
In either case, a human (or remote operator) may easily take over control and follow the flag person’s hand gestures, or out-bluff the human-controlled cars at the intersection. But what if the need for human intervention arose on the freeway, perhaps when dirt dropped by the truck ahead obscures the lane markings? The autonomous car might well begin to steer itself into a concrete barrier and the only hope would be for a human driver to instantly take over — if they were even to look up from watching their video in time before everything went dark amid the thunder of crunching metal and glass…
A major concern is that current patent laws are inadequate to protect AI systems.
Artificial intelligence is a glamorous term that suggests human-like thinking. It conjures images of popular movie characters who can speak and interact as would a person. But AI is more accurately described as “machine learning.” Present-day machine learning attempts to duplicate human intelligence by interacting with the world and receiving corrective input. Almost like teaching a child right from wrong through praise and scolding, present-day AI machines learn based on similar binary corrections. Go 35, not the posted 30 mph, and feedback that this is wrong behavior is processed until the machine “knows” not to exceed 30 mph down Maple street, and knows that use of all three lanes in the westbound direction is permissible at the present time because Maple is one-way westbound during rush hour.
Hundreds if not thousands of test vehicles are sharing the road with people-guided cars and trucks, logging the millions of miles needed to teach the machines until they can replicate or at least approximate the knowledge of an experienced driver so that they can be relied on to obey the road repair crewperson, or learn to deal with the rolling-stop human drivers. Unlike technology taught in patents, which allow a newcomer to catch up (provided they are willing to pay a license fee or can design around), present-day machine learning takes time and experience that can’t be avoided. Or can it?
A major concern is that current patent laws are inadequate to protect AI systems. For example, patents cannot be used to protect data compilations, such as AI training sets, or a programmer’s particular expression of source code. Further, given the machine learning process and iterative/incremental evolution of the underlying algorithms, it can be difficult to accurately, and narrowly, describe the methods or functions of an AI system as required for a patent grant.
And what of inventorship? The self-learning process of AI means that subject matter could be developed autonomously by the AI itself. Even if the result is patentable, would HAL be named as the inventor? This is an unsettled area of law, likely requiring action by Congress to resolve. A practical problem also is that the field may be evolving so rapidly that a patent, which can take years to fully prosecute from application to issuance, may be useless or antiquated before it issues, if allowed at all.
Given these concerns, most autonomous vehicle (and AI) developers rely on trade secret laws to protect their intellectual property. But this creates a significant technological hurdle for new companies entering the field. Rather than being able to rely on the publicly disclosed information that would be required, for example, if the AI developers were patenting their technology, new competitors would essentially have to start from scratch — a severe disadvantage. With fewer competitors able to enter the market, consumers will no doubt be limited to fewer choices.
There are numerous drawbacks, as well, to developers relying on trade-secret protection, not the least of which is the temptation of competitors to shortcut the multi-million-mile learning process by hiring away valued employees and encouraging theft of the learned data. To be protectable, and thus actionable in the event of a breach, the trade secrets must be kept confidential, which can translate into cumbersome security measures and “need to know” protocols and limitations.
All of this is somewhat reminiscent of how rules of the road, signage and traffic signals had to be developed to facilitate transition from our horse-drawn age. In the same way, our concepts of intellectual property protection will need to evolve if not require invention of new concepts to enable transition to autonomous vehicles. Stay tuned — and buckle up for a fascinating ride.