Navigating the transition from human-controlled cars to a better driverless future
In the six-plus years that Google has been developing self-driving cars, its test fleet has achieved an impressive safety record: Nearly one million miles of real-world autonomous driving, eleven minor accidents, no injuries, with none of the accidents caused by the self-driving cars themselves.
That impressive track record has people projecting when the first commercially available autonomous vehicles will hit the streets. But commercial availability is just a waypoint on the truly transformative journey that lies ahead of us. And that means there’s a bigger question we should be considering now. Namely, how soon will it be illegal to operate human-driven cars on public streets?
At a conference in March, my friend and colleague Elon Musk raised this proposition, exclaiming that “people may outlaw driving cars because it’s too dangerous.”
In a follow-up tweet, Elon clarified that “Tesla is strongly in favor of people being allowed to drive their cars and always will be.”
That perspective makes sense. Although Tesla is developing autonomous technologies, its high-performance cars are a blast to operate in a traditional hands-on manner. Many Tesla customers buy the company’s cars precisely for this reason, and as a smart and dedicated CEO, Elon will continue to serve his customers well.
But even though I’m a happy Ford Expedition owner myself, I also recognize that self-driving cars have the potential to radically reshape the world in ways that will benefit everyone. In the future, getting from point A to point B is going to be safer, faster, more energy-efficient, more economical, and more fun.
In addition, autonomous vehicles will also be able to share information with each other better than human drivers can, in both real-time situations and over time. Every car on the road will benefit from what every other car has learned. Driving will be a networked activity, with tighter feedback loops and a much greater ability to aggregate, analyze, and redistribute knowledge.
Today, as individual drivers compete for space, they often work against each other’s interests, sometimes obliviously, sometimes deliberately. In a world of networked driverless cars, driving retains the individualized flexibility that has always made automobility so attractive. But it also becomes a highly collaborative endeavor, with greater cooperation leading to greater efficiency. It’s not just steering wheels and rear-view mirrors that driverless cars render obsolete. You won’t need horns either. Or middle fingers.
Already, the car as network node is what drives apps like Waze, which uses smartphone GPS capabilities to crowd-source real-time traffic levels, road conditions, and even gas prices. But Waze still depends on humans to apprehend the information it generates. Autonomous vehicles, in contrast, will be able to generate, analyze, and act on information without human bottlenecks. And when thousands and then even millions of cars are connected in this way, new capabilities are going to emerge. The rate of innovation will accelerate – just as it did when we made the shift from standalone PCs to networked PCs.
So we as a society should be doing everything we can to reach this better future sooner rather than later, in ways that make the transition as smooth as possible. And that includes prohibiting human-driven cars in many contexts. On this particular road trip, the journey is not the reward. The destination is.
The True Transformative Power of Autonomous Vehicles
Human-driven cars won’t disappear entirely. There will be designated areas where people can drive for pleasure – race tracks, larger areas that are similar to golf courses or game preserves, public roads in extremely remote areas, or areas where weather or terrain make it difficult to optimize infrastructure for autonomous driving.
But the benefits of self-driving cars are so significant that in time the public will demand prohibitions against old-fashioned legacy driving in most public spaces.
To date, most discourse on driverless cars has focused on their ability to reduce collisions – especially fatal ones. In reality, driving is already a fairly safe activity. According to Federal Highway Administration statistics, there are around 189 crashes involving injury or property damage for every 100 million vehicle miles traveled (VMT) in the U.S., and only 1.04 fatal collisions per 100
But even though collision rates are low, the fact that we do so much driving means that the real tolls that traditional driving exacts are significant. In the U.S. alone, around 33,000 people a year die in traffic accidents. Globally, there are more than 1.2 million traffic deaths per year, with China suffering around 275,000 and India 238,000.
And when collisions do occur, it’s usually the drivers – rather than mechanical failure – who are responsible. In fact, human error accounts for more than 90 percent of all motor vehicle accidents. Alcohol plays a role in more than 1 in 3 traffic-related fatalities.
Replacing the most fallible component in the driving experience – the human driver – will lead to significant reductions in the number of accidents and fatalities that automobility produces.
Autonomous vehicles equipped with lasers, infrared sensors, cameras, detailed 3D road-maps, and other technologies are able to assess their surroundings in ways that human eyes can’t. They can detect objects behind walls. They can accurately estimate distance at speed. They can brake and accelerate faster than humans can and change direction with more precision. They don’t drink and drive, text and drive, nod off six hours into a long trip, or experience road rage.
But it’s not just the capacity of autonomous vehicles to make a fairly safe activity even safer that makes them so transformative. It’s that they will substantially reduce fatalities and collisions while simultaneously increasing overall transportation efficiency and decreasing our need to pay attention while driving.
If thousands of human-controlled cars suddenly started traveling down highways at 90 MPH, with little space between their bumpers, there would certainly be a lot more than 189 collisions per 100 million VMT. If all those drivers also started texting and watching TV, those freeways would likely begin to resemble war zones.
But autonomous vehicles will be able to pull off such feats. Indeed, their ability to travel at speed with less distance between them can potentially increase freeway capacity by 6 or 8 times.
A world of networked cars will also mean less idling at traffic lights, no stop-and-go slogs during rush-hour, no endless circling for parking. Commutes will be shorter. Crashes will be rarer – so the traffic jams that crashes create will be rarer too. And because crashes will be rare, cars can be built smaller, from lighter materials. If they’re still using combustion engines, they’ll burn less gas. If they’re powered by electricity, they’ll travel farther on a single charge.
Any time spent in cars will also be far more productive. According to Harvard Medical School, the average American driver spends 101 minutes behind the wheel each day. There are 210 million licensed drivers in the U.S., which suggests that collectively we spend around 5.3 billion hours a year wondering why that minivan in front of us is going so slow and listening to soft-rock hits from the 1970s.
Traditional cars liberated us from geography and time. Autonomous cars will liberate us from driving. In the future, motorists will be able to work, sleep, you name it. Watching a game on TV will replace listening to the game on the radio. Enjoying a leisurely meal with a knife, fork, and a glass of wine will replace gobbling down a burger with one hand.
Drinking and driving won’t be a crime – it will be an entrepreneurial opportunity. Texting and driving will still be frowned on – but only because people will be wondering why you aren’t Meerkatting and driving. When you use the word “gridlock,” your kids will have no idea what you mean. But they’ll be able to look it up on Google, even if they’re speeding down the highway at 100 MPH.
Finally, if you find yourself missing the physical pleasures of driving, you’re in luck: Just slap on an Oculus Rift virtual reality headset and immerse yourself in a 3D version of Grand Theft Auto. Your daily commute to work will be more tactile and exciting than ever.
On a strictly quantitative level, at least, the man-hours that autonomous vehicles will save each year by making it possible to do other things while driving will far exceed the man-hours they save each year by eliminating fatal collisions. By fully embracing autonomous vehicles, humanity will free up billions of hours to spend in more flexible and potentially rewarding ways. The fact that driverless vehicles will significantly reduce driving-related deaths and the costs associated with them is an extremely valuable aspect of them. But what makes them truly disruptive is not only their capacity to save lives – it’s their capacity to save minutes and hours, and help billions of people live more productively year after year.
Moving Forward, Full Speed Ahead
So how fast can we get to the remarkable future that driverless cars promise? Some observers believe that we’re still at least decades away from a truly autonomous car that can function safely when you throw snow, rain, poor road surfaces and other factors into the mix.
Meanwhile, according to Elon’s estimates, there are more than 2 billion legacy cars on the road, globally. Currently, the car industry can only produce around 100 million new vehicles a year. Just from a manufacturing perspective, it could take 20 years to build a new fleet that approximates the one we have now.
And yet perhaps because announcements like these are coming at such a fast and furious pace, there is still a general consensus to proceed with caution. Two years ago, German automotive supplier Continental AG declared its intention to produce driverless cars that will function with a “zero-percent accident rate.” Others suggest that attaining a “four nines” safety level is necessary. That is, these vehicles must operate successfully – i.e. not crash – at least 99.99 percent of the time they’re in operation.
The emphasis on infallibility is misguided. Even the best artificial intelligence systems will be hard-pressed to completely protect against falling trees, mudslides, and other “acts of God.” In a world of mandatory driverless vehicles, accidents and even fatalities will continue to occur. To suggest otherwise only sets the stage for class-action lawsuits that could inhibit genuine progress.
In addition, the truth is that infallibility is not necessary to achieve significant progress. Imagine if we were able to make a complete switch to driverless cars in 2016, and 20,000 driving-related fatalities occurred over the course of the year. This outcome would likely be positioned as an unprecedented epidemic of robot car carnage. But a drop to just 20,000 fatalities in one year would actually represent the largest annual decline in U.S. traffic fatalities ever.
When we’re able to look past our fears about change and our biases toward the DeVille we know, driverless cars will make driving safer – just as higher levels of automation have made commercial aviation a remarkably safe form of travel. In fact, the Eno Center for Transportation, a Washington, DC think tank, estimates that attaining a 90 percent market penetration rate of autonomous vehicles will save 21,700 lives a year in the U.S. alone.
But why stop at 90 percent market penetration? Why not take it to 100 percent, as soon as possible?
The Great Merge
Long before software companies existed, the American car industry had perfected the art of the incremental upgrade, a new model each year with new bells and whistles. That template for innovation will be applied to autonomous
vehicles, especially by traditional manufactures.
In other words, they’ll gradually add autonomous features until eventually they’ll start selling vehicles that offer what Google’s first prototypes did – i.e. “limited self-driving automation,” or “Level 3” autonomy. These cars will be able to drive themselves for extended periods of time, but will also give drivers the option to assume control when necessary.
Such cars will share the streets with traditional cars, so the driving environment will be mixed on both micro and macro levels. Given how established traditional car culture is – it is literally entrenched throughout the world in millions of miles of roadway – this is inevitable. But it is also likely to prove somewhat challenging.
According to my friend and colleague Stefan Heck, a Stanford professor who is also founder of Nauto, a Silicon Valley startup developing autonomous vehicle technologies, research that Stanford has done shows that drivers resuming control from Level 3 vehicles functioning in autonomous mode take 10 seconds just to attain the level of ability that a drunk driver possesses. And to get back to full driving competence takes 60 seconds.
Because human drivers have little experience interacting with driverless cars, they also tend to act unpredictably when encountering them. That hasn’t been a significant issue to date, because there are so few autonomous vehicles on the road. But as autonomous vehicles become commercially available and their numbers increase, that may change. “There is a real potential for an increase in accidents when we first deploy autonomous cars,” says Stefan.
The most obvious detour around this dilemma is to minimize mixed environments. Of course, America is still a country where the right to drive is considered an essential facet of life, right up there with the right to vote and the right to own a firearm. While car sales dropped substantially in the recession, they’re rising again, even amongst millennials who had once been thought to be far more interested in car access than car ownership.
But the physical act of driving isn’t a natural or constitutionally protected right. It’s a licensable privilege. In the early days of automobility, we had to establish new infrastructure, new rules, and even new values to make them tenable – more police presence, stop signs, traffic signals, sidewalks, one-way streets, parking spaces, speed limits, crosswalks, etc. And it took decades to make the transition, as streets that had once been shared by multiple low-speed users, including pedestrians, bicyclists, horse-drawn carriages, and streetcars, were recast primarily as the domain of motor vehicles.
This time around, the new mode of transportation won’t just be far more powerful than the one it is disrupting. It will also be safer. But if we want to move forward, we will have to let go of the past. And just as we reshaped our world for the first generation of cars, we’ll need to create new rules and new infrastructure to accommodate autonomous vehicles as well.
Geographic Regions Must Begin to Innovate
In the early years of the automobile age, no federal or even state rules existed. Instead, individual municipalities had to figure out how best to manage this new form of traffic. A similar dynamic is already emerging with autonomous vehicles.
On a state level, California, Nevada, and Florida have led the way in passing legislation that explicitly asserts the legality of autonomous vehicles under certain conditions. Last summer, Iowa’s Johnson County approved a proclamation that encourages companies to use it “as a primary location for testing driverless car operations.”
In Ann Arbor, the University of Michigan is in the process of building a $6.5 million “city” on 30 acres of its campus to test autonomous cars in a simulated urban environment that will reportedly feature “complex intersections, confusing lane markings” and even “mechanical pedestrians” jumping from behind parked cars. By 2021, the university hopes to use the information it collects from this site to help Ann Arbor adopt a networked fleet of shared, self-driving cars.
In England, the New York Times reports, the British government has invested $61 million “for trials aimed at making Britain a global hub for testing rules for driverless cars and building the components for these vehicles.”
All of this recalls the early 20th century, when some cities, recognizing the transformative impact cars would have, made efforts to accommodate them sooner rather than later. Detroit, for example was the first city to adopt traffic signals, one-way streets, and many other conventions that helped usher in a new age of automobility.
The rewards for the places that lead the transition to a next stage of automobility will be substantial. Indeed, if San Francisco took this lead on this, much of the space it currently devotes to accommodating human-driven cars, including parking garages, parking lots, street parking spaces, private garages, and wide streets – could be freed up for new housing, which it desperately needs.
Still, I doubt major transitional initiatives will happen first in San Francisco or even Mountain View. Instead, they’ll happen in places that are eager to be the next San Francisco or Mountain View – places willing to experiment in largely uncharted ways.
Maybe that’s Johnson County, Iowa. Maybe it’s Singapore, which is also positioning itself as a place that is giving a green light to driverless car experimentation.
A welcoming attitude is a key first step – but there are certainly opportunities to go further than that. To date, most driverless car development has involved trying to create vehicles that can function well in the world that exists now. This is a smart approach that has allowed researchers and entrepreneurs to simply start experimenting, without trying to persuade cities and other institutions to adapt technologies, infrastructure, and regulations that would make it easier for driverless cars to function with maximum safety and efficiency.
But we’re now at a point where progress will happen faster if cities and other geographic regions embrace innovation as strongly as the institutions developing autonomous vehicle technologies.
When a Google car currently hits the road, it’s largely functioning as a solo actor and doing all the work to decipher the world it is navigating. But technologies that allow cars to talk to each other, through Wi-Fi-like networks that use dedicated short-range communications frequencies, exist too.
In these vehicle-to-vehicle (V2V) networks, cars share information with each other and other smart infrastructure elements – traffic signals, sensor-embedded roads, roadside cameras, eye-in-the-sky traffic drones, etc.
Adding such infrastructure elements is one way that cities could help accelerate the development of autonomous vehicles. Designating some streets or downtown areas as driverless only would be an even bolder and more useful move. Detroit, in other words, has a shot at becoming the Detroit of the 21st century. All it has to do is ban traditional cars from its streets.
New Rules for New Rides
Networked autonomous vehicles will bring new challenges along with new opportunities. Cybersecurity will be an issue. The first self-driving car that gets hacked in some cataclysmic way will generate just as many headlines as the first one that’s involved in a fatal collision (especially if it’s the same car).
Privacy is also a concern. Still, hacking is already an issue with current on-board systems like OnStar. And such systems are already able to collect information about your driving patterns, how fast you go, and more.
On a similar note, the Drug Enforcement Administration and many other law enforcement agencies use license-plate cameras to build increasingly comprehensive databases that keep tabs on millions of vehicles over time. Using a feature built into OnStar called Stolen Vehicle Slowdown, police can remotely decelerate a vehicle that has been reported stolen.
But OnStar and its analogs are optional. Law-enforcement license-plate surveillance relies on a relatively fixed number of data collection points. Autonomous vehicles will literally be able to record and store your every move. And unlike smartphones, you won’t be able to leave them behind or disable their GPS if you want to go off the grid for a while. Autonomous vehicles can’t function unless you tell them where you’re going.
At the very least, the companies developing these technologies will need to be extremely transparent about how they collect and use the data these vehicles generate. And ultimately consumers will likely demand regulations that let them control their privacy – especially if human-driven cars are prohibited.
The algorithms that govern facets of these cars’ behavior will also have to be transparent – and codified into law. Already, manufacturers are consulting philosophers to help guide how their algorithms make decisions in moments of emergency. For example, should an autonomous vehicle risk hitting a cyclist in its efforts to avoid a schoolbus that is heading straight for it, if choosing the former course of action will likely minimize overall deaths and injuries? If an autonomous ambulance is racing toward a hospital, should it take riskier actions as the patient it’s carrying is moving closer to death? And what sorts of preferences should emergency and law-enforcement vehicles be given in general?
Even in cases of non-emergency, a high degree of transparency is necessary. Every time a passenger indicates a desired destination, an autonomous vehicle must make choices about the optimal route. Presumably, it will do so based on current traffic conditions, as Waze does now. But it’s also possible that the companies designing these cars could choose routes for other reasons. For example, advertisers might pay companies to route passengers past their businesses. Passengers with preferred status could receive access to faster streets while others are routed to slower, higher-volume streets.
In some cases, passengers may accept these decisions. You might pay less or receive some other perk if you agree to take the slow route home, or pay more to take the fast one. On a similar note, we will probably see the introduction of literal “marketing vehicles,” i.e., cars that take you to your destination for free as long as you complete a survey or watch a promotional video of some kind.
Because the various algorithms that govern car behavior will encompass issues of liability, risk, and morality, no one company should be allowed to simply make up their own rules. Instead, we’ll need to establish uniform rules and standards through public processes. In the same way that we currently have regulations involving emissions standards, safety equipment, and other aspects of car manufacture, we’ll also have regulations that establish the parameters for how the necessary algorithms operate.
While government will clearly play a role in this regulatory process, we must also consider what limits to place on how government agencies utilize coming technologies. If you become a suspect in a crime, how extensively will police be allowed to examine your driving history? Will law enforcement agencies have the ability to tap into the system and control the routes of persons they want to detain? Public records detailing exactly how such agencies are using whatever capabilities are granted to them will be necessary.
Liability is another major that must be addressed. Because driver error is most often the reason for collisions involving traditional cars, insurance is currently oriented toward individuals. In driverless scenarios, however, a car’s occupants will be functioning as passengers, not drivers. Thus, when accidents happen, who will be held accountable?
Because most of the companies that produce driverless cars will be large and well-capitalized, the incentives to file lawsuits – and to award high-dollar damages – will be great. Unfortunately, if the threat of lawsuits is high enough, it could keep companies from offering cars that would ultimately reduce the overall number of traffic collisions and fatalities. Recognizing this, we must start thinking about how to ensure that liability laws fairly compensate victims without unnecessarily preventing the spread of technologies that will lead to better overall outcomes for everyone.
Throughout the 20th century, traditional cars functioned as engines of personal freedom and prosperity. The highly personalized mobility they provided gave people more access to a wider range of jobs, homes, friends, etc.
Autonomous vehicles, in turn, promise a new range of capabilities. But as we merge from our old way of driving to a new one, there will be objections. Some people will characterize self-driving cars as an infringement on personal liberty and an individual’s ability to fully control his own destiny. Others will raise concerns over privacy and the ways that driverless cars will collect and share data in their pursuit of greater safety and efficiency.
But autonomous vehicles won’t curtail personal freedom – they’ll amplify it. Autonomous vehicles will extend the convenience of individualized driving to people who aren’t currently able to obtain driving licenses –senior citizens, people with various disabilities, young people. They will let everyone pursue a greater range of activities while they’re in transit. They’ll speed up transit times and help people forsake transit altogether. (I.E., your car will run errands for you while you stay at home.) They’ll reduce the need to actually own a car, and thus release people from the economic obligations of that.
And privacy norms are not a fixed phenomenon. They change over time, in tandem with new technologies, new social standards, new expectations. As driverless technology evolves, we’ll develop features and policies that will allow people to maintain appropriate levels of privacy. But they will also have access to greater transportation capabilities than even George Jetson ever dreamed of.
An asphalt utopia is on the horizon. It promises streets that will be safer than they were when a tired horse was the fastest means between two points. It promises shorter, more productive commutes. It’s time to put the pedal to the metal in pursuit of this vision, and accelerate toward a world where self-driving cars are not just allowed but mandatory in the vast majority of spaces.