Yesterday, March 19, the first known pedestrian fatality involving an autonomous vehicle was registered in Tempe, Arizona. A woman, 49-year old Elaine Herzberg, crossing a busy street around 10 PM was struck by a driverless vehicle, and died of her injuries. The vehicle was an AV pilot car, A Volvo sports utility vehicle owned by Uber, and there was an emergency human driver in the car at the time of the accident. The accident in Tempe was terrible, and has led to an understandable outcry of concern in news media across the country. Uber announced this morning an immediate suspension of its autonomous vehicle pilots across the US.
Yet AVs are coming. Waymo and Uber, the industry leaders, say their AVs have covered over 8 million miles between them. As with many technologies, relative costs will be central to their adoption: Bloomberg New Energy Finance and McKinsey estimate that the cost of a driverless, shared-use car will be less than 10% of the average cost of taxi services. It is not as though everyone has to make an individual decision to buy an AV: people will just switch from owning cars to using an app on their phone to order transportation, as is already happening with Uber and similar services. Corporate vehicle fleets are likely to adopt AVs earlier than retail users. Convenience, much as with the cell phone, can be expected to be another big driver (pun intended). Greater efficiency may translate into much shorter urban driving times (as AVs can move much more efficiently through intersections and maintain shorter following distances – unless of course there are so many of them that no one can move anyway…), and climate benefits (less idling and fewer jack-rabbit starts). So we can expect a collision (pun again intended) between the attractions of this new technology, and its risks – including, as we tragically saw yesterday, loss of life.
What lessons can we draw from the accident in Arizona?
1. Autonomous vehicles are not perfect, and never will be. This is hardly a revelation, and is true of essentially all technologies. AV technology is in fact still at an early stage, and surely has a long way to go in further reducing the risks of injuries and fatalities. Ye the strain of thought, understandably part of today’s reactions, that somehow AVs are inherently more dangerous than vehicles with human drivers and always will be, does not make sense. With 37,000 road deaths in 2016 just in the US, human drivers are no model of safety themselves. That’s 100 road deaths per day. David Leonhart’s Op-Ed column in the New York Times on the topic points out that, as underlined by research, we as humans tend to mistrust technology and its ability to learn from mistakes when something has gone wrong, but then to assume we ourselves will instead correct our mistakes. That AV technology can be developed to the point that AVs cause proportionately lower deaths than human drivers seems to me highly likely, notwithstanding yesterday’s fatality. The potential for greater safety is certainly being touted by AV manufacturers, and human error accounts for 94% of vehicle crashes. Whether the technology is such that they are safer today is hard to tell, but what is obvious is that the technology will get better, much better, while human driver risk is unlikely to improve much. Assuming, of course, that AVs don’t learn to consume alcohol before they get on the road.
2. Regulation matters. Again, this sounds self-evident. But it matters – a lot. The quality of regulation will take on far greater importance as new infrastructure technologies emerge and get deployed. This point is already becoming clear with new power generation technologies – wind and solar generation, and especially Distributed Generation. How jurisdictions set regulations is having a very large effect on the speed of deployment of new technologies. And in power generation, as the costs of buying wind and solar power continue to plunge, how jurisdictions set regulations is having a real impact on the cost of electricity. With lower cost energy storage technologies on the horizon, this will be more and more the case. With new mobility technologies – Electric Vehicles and Autonomous Vehicles (not to mention drones) – the same dynamic will hold true. Good regulation will help cities, states and countries achieve better outcomes. That yesterday’s accident happened in Arizona is not a surprise – the no-regulation or minimal-regulation stance of Arizona’s state government has made it an attractive place for early testing of AV technology. State officials declared Arizona “a regulation-free zone” in 2015 to attract technology companies with operations in neighboring California. Arizona is one of the first place Uber has gone to test its AV cars. Waymo is also active there, using AVs without humans in the driver’s seat to pick up and drop off passengers. Encouraging new technology is good, but doing so with no rules will invariably entail higher risk. To date 21 states have introduced legislation regulating AVs: you can bet many more will do so shortly.
3. Regulating infrastructure well is getting more difficult. While Arizona was chosen by Uber as a place to test its AVs presumably at least in part due to lack of regulatory obstacles, it’s not obvious yet how best to regulate AVs. Los Angeles and a number of other cities are struggling with the issue, and a set of regulatory principles are beginning to emerge, but still have a ways to go (like the technology itself). This does not make regulation less necessary, just more complex and challenging to design and execute (the current anti-regulatory mood in the US congress, which may produce a prohibition on states’ abilities to regulate AVs, is another challenge). Many cities believe they are paying the costs of failing to understand the implications of shared-services models, such as Uber and Lyft (which while technically a business model change rather than a technology itself, has similar effects to technology change and is facilitated by the advent of new communications and big-data technologies) – and failing to regulate these models well. Regulating badly is easy to define: either attempting to forbid TNCs (Transportation Network Companies) from operating, as New York did with Uber, or not having any rules whatsoever, which is the most general case. Attempting to block TNCs is a bit like trying to put a genie back in a bottle, and effectively prevents consumers from gaining the cost and convenience advantages that have been obvious with TNCs. Doing nothing means having absolutely no influence on how TNC models play out in a city, including potential negative effects on congestion (as London and other cities have visibly experienced), on the financial sustainability of mass transit (which appears to be an effect in Washington, DC, along with simultaneous poor performance of the metro system there), on municipal revenues from taxi services, among other effects. Regulating well is less easy to define. Yet like for many other things, planning and experiment are building a set of experiences and lessons that can create a body of effective regulatory practices. This dynamic, around mobility technologies, again mirrors the dynamic which is playing out around new energy technologies, and which played out previously around new communications technologies. For AVs, policy questions include whether to require AVs to be able to communicate with each other (which makes a big safety and efficiency difference), where they can or cannot drive, where they can be charged, where they are when not driving someone (parked, or driving around), and what to do with the data they generate.
4. Learning is critical for cities. Unlike energy and communications technologies, whose effects cover wide geographies, the impact of the new mobility technologies like AVs are heavily concentrated in cities. The need for regulating well, and the attendant challenges of learning how to regulate well, will fall disproportionately on cities. Networks of cities, and other forms of learning, will be increasingly valuable to mayors and city administrations around the world. Yesterday’s accident will not be helpful to Tempe Mayor Mark Mitchell’s political fortunes. No mayors will want to be blamed for doing nothing if and when the first AV fatality happens in their city.
Robots, in any case, will surely cause bigger problems in other ways than driving our cars.