“People are super comfortable riding in elevators”: Can we shape people’s attitude to driverless cars?
Udacity’s head of driverless cars talks timescales, safety and getting people to feel comfortable without a human in charge
by Alan Martin
On Monday, Uber – a company not known for effective PR damage control – entered uncharted waters. One of its autonomous cars had hit and killed a woman in Arizona and gave news outlets around the world an unwanted focal point: the first pedestrian casualty caused by a driverless car.
The fact that this in itself was newsworthy is worth analysing. Accidents caused by human-driven cars are so frequent that they seldom get acknowledged outside of official statistics. Driving is risky; humans are fallible – fatalities are, therefore, an inevitability of shooting round the world 15 times faster than we’re designed to go. Autonomous cars don’t get the same free pass.
As a non-driving journalist who sometimes covers autonomous vehicles, I’m torn between longing for the day I can buy one, and a slightly Luddite sense that this just shouldn’t work. Logically I can see that one casualty in hundreds of millions of miles of testing sounds far better than the 1,792 reported UK road deaths recorded in 2016. And yet…
Two weeks before the Uber incident, I’d sat down for a drink with David Silver: a former driverless car engineer at Ford, who now heads up Udacity’s driverless car educational courses. We’d spent a lot of our time together talking about safety, public unease and theoreticals.
“I think what we as an industry need to do a better job communicating to people is the number of lives saved by putting driverless cars out on the road and removing humans from the equation,” he said. “Particularly in the United States, drunk driving is a huge cause of automotive fatalities and cars don’t get drunk, and that makes self-driving cars in many ways a safer alternative to human drivers.”
Not only that, but driverless cars have another big advantage over you or I: autonomous vehicles never forget. “Every single person has to learn those things individually as they drive, and the wonderful thing about self-driving vehicles is that every time one vehicle learns something, all of that goes up into the cloud,” Silver explains. From there, it can be distributed to every vehicle so each car learns from others’ mistakes.
Dealing with feelings
This could be genuinely transformational, assuming manufacturers are prepared to share their data with each other. But logic and expert opinion doesn’t tackle feelings. So how can the industry get over that sense of unease, especially when one fatality can take up all of the headlines for a week?
Some things are simple: transparency on the part of the companies (covering up accidents would not be a good look) and of the vehicles themselves – if the car is telling you what it can see and how it’s going to react, that could well be reassuring to some. Mainly though, it’s down to time: you can’t speed up an attitudinal sea change, but in a decade’s time, this article will almost certainly appear quaintly archaic. “The idea of a hailing a stranger’s car and just getting in with a total stranger would seem wild ten years ago, and now it’s super common,” Silver points out.
Perhaps a better comparison is the first truly driverless vehicle, already over 100 years old. “People are now super comfortable riding in elevators,” Silver says. “It didn’t take that long – if there were lots of elevator accidents, people would be concerned but the elevator companies are very, very responsible about their product and people are fine with it.”
It’s a good analogy, but not a perfect one. Elevators are on a set track, and they don’t have to interact with human-driven pods in the same shaft. For safety to no longer be a concern, wouldn’t it be easier if humans just voluntary handed in their car keys?
Yes, but it’s a moot point: “It isn’t realistic to imagine that we’re going to remove humans from the road in the foreseeable future,” Silver says. “Maybe in the medium range future, five to ten years, small parts of urban areas may be specifically dedicated to driverless cars. But we are still continuing to sell cars today, and the average car has a lifespan of a dozen years.
“My vision for this is that driving will become a little bit like biking is now. Bicycling is a real form of transportation for people in urban centres, but for most people in America it’s more of a recreational activity. And I can see driving transitioning to that over the next ten years.”
Pranks, pedestrians and death by natural causes
For those keeping track, that’s the third time the magical timeframe of “ten years” has been mentioned in this article. 2028 may feel like a lifetime away, but some critics are already considering unusual scenarios. A piece by John Adams in London Essays from 2015 brought up an interesting consequence of overhauling our roads: “Such a programmed response [cars which are programmed to stop automatically] will have invented an exciting new game for children: throw the ball and watch the car stop.” Without the disincentive of likely injury, what is to stop new behaviours like these, or just crossing motorways wherever pedestrians feel like it, becoming the new normal?
“That’s a really good question and that’s something that self driving car engineers are already having to grapple with,” Silver says. “On the highway, human drivers have sometimes identified self-driving cars as being easy vehicles to cut off because you know they won’t hit you.”
But will it get worse than that when driverless cars are no longer the minority? “There are things that humans could do now that would cause massive chaos and traffic,” Silver argues. “They could change around stop signs, they could cut down traffic lights. In general they don’t because: 1) Most people are good people and don’t want to cause chaos and 2) Because they know that there are punishments if they do these things.
“We could see a point in the future where society views jumping in front of a self-driving car for fun as a much more serious offence than we might see it now. It could cause a lot of chaos.”
You may think that catching a prankster jumping in front of a car would be difficult (even if they feel inclined to put the footage on YouTube), but interestingly the very technology of the medium could potentially make convictions a lot easier.
To function properly, autonomous vehicles are fitted with a whole array of sensors, including cameras to capture the road ahead and behind. Could that footage be used to catch a new batch of as yet unlegislated crimes? Silver doesn’t rule it out, comparing it to current instances where government or law enforcement subpoenas internet companies for criminal data. “Could a car company look at that data and determine who committed a crime? I think that’s a real issue that could in the not too distant future be something that as a society we need to grapple with.”
Okay, let’s have another one. Ever since I saw this tweet, I’ve found it hard to dislodge the possibility from my mind:
It may seem macabre and darkly comic, but surely this is actually just a statistical inevitability? Someone, somewhere will die of natural causes in a driverless car, and then be delivered to their planned destination. How much do car makers ponder these theoreticals? “You have tremendous amounts of driving [tests] to catch up to these really rare cases,” Silver replies. “Sebastian Thrun – my boss – has lots of amazing videos from his time working with self-driving cars, with chairs falling out of the sky right in the middle of the highway, or cars driving down the wrong side of an interstate. Chris Urmson who later led the self-driving car project famously tells the story of a woman in a wheelchair who was herding ducks in the middle of the street!”
So if strange behaviour can be predicted, we could respond to these problems before they happen. What’s to stop us putting something to detect heart rate into a seatbelt to automatically redirect you to a hospital if things look dicey? “Exactly. You don’t even have to imagine: some companies are already building the sensors in the car for things like gesture recognition.”
Is it worth it?
With all these hurdles to overcome, you do occasionally have to focus on the prizes at stake: fewer deaths by dangerous driving, reduced carbon emissions, less demand for parking spaces as driverless cars just pootle around on their own, waiting for someone to hail them. Most interestingly to Silver is the potential increase in productivity: the WHO has estimated that some countries lose 3% of their GDP every year through road accidents, traffic and other car-related problems.
So can anything stop a future of autonomous cars? “I think the transportation curve will bend towards an anonymous future,” Silver says, adding that it’s in the interests of vehicle developers to be responsible and cautious in the rollout of autonomous cars as “a few big accidents could really set back the industry in terms of public acceptance in regulation.” The long-term response to the autonomous Uber casualty – which occurred after this interview took place – will put that to the test.
And while the autonomous space is an area that all kinds of automotive, hardware and software manufacturers want to get into, at the moment finding qualified talent is a real bottleneck. “Acquiring self-driving car engineers is a really difficult feat right now,” Silver concedes. “There are very few people who are trained to work on this, and that’s what we’re trying to do at Udacity.” The company’s nanodegree program has had 10,000 students, many of whom now work in the industry, according to Silver. And while there’s no such thing as a job for life anymore, Silver is confident that those taking the course will be ready for the future of transport: “I think in the long run autonomous vehicles seem inevitable at this point.
“It’s just a question of how long.”