No doubt you've heard that on Sunday, an Uber in Tempe, Arizona, operating in autonomous mode with a backup safety driver, or supervisor, in the car, killed a pedestrian.
What we know at this point is based on the Tempe police's preliminary findings after viewing the car's forward-facing camera and speaking to the Uber safety driver.
Whatever the ultimate findings of the NTSB, which is investigating the incident, and potential liability, if any, of Uber, or its driver, one thing is clear: When an autonomous vehicle is involved in an incident that results in injury or death, it's going to get our attention. And, even though I'm a big proponent of autonomous vehicles, I believe that it should.
They Are Not Driverless
It has always been the case that any sufficiently disruptive technology will evoke fear, resistance, and the wrath of naysayers who will always have ample evidence to prove that disruption is a very bad thing.
What happened in Tempe is tragic, as is every one of the 3,287 automobile deaths each day. None of them are acceptable to the people killed, their families, and friends. But that's exactly why I'm so keen on autonomous vehicles, because that carnage is something we no longer need to accept.
Tempe is one of countless incidents where autonomous machines will be scrutinized for their actions and behaviors. Yes, I said behaviors, because in many cases what these devices do and how they act will be far more akin to a behavior, in which decisions are not always apparent, versus a programmatic or machine-like response, which is predictable.
Notice that I'm also not calling them driverless, because they aren't. They have drivers, just not human ones. We need to drop the driverless label and start to understand what autonomous means if we're to accurately weigh the benefits against the risks.
This isn't about driverlessness. It's about an entirely new kind of driver; one that, although not infallible, is ultimately better and safer than the vast majority of human drivers in the vast majority of cases. We are creating a new species that we're going to co-habitate the world with. So, it's worth taking the time to understand what that means and, to whatever degree we can, put in place the cornerstones for its evolution.
To help reveal a bit about how AVs are going to evolve, and to provide a foundation for understanding them, I'm writing a three-part series of columns, each one focusing on a critical aspect of AVs. The first two will look at why we have such difficulty in seeing beyond the short-term risks of AVs.
Living in the Danger Zone
Of all the things we do each day, driving and sharing our world with vehicles is the riskiest. It's exactly because driving a car is so risky that we have a tough time imagining how an intelligent machine could ever do it as well as we humans. But there's just one problem with that thinking: We really don't do it all that well.
Motor vehicles account for 1.3 million deaths each year, placing them as the 10th leading cause of death globally and the only nondisease related cause of death in the top 10. Which means that if you adjust for the fact that there are only one billion vehicles globally, as opposed to the fact that all seven billion people risk acquiring any of the other nine diseases, you could make the case that vehicles are the leading cause of death for those who own or interact with an automobile.
Given that, you'd think we'd be thrilled at the prospect of having technology take over. We're not. In fact, we're mostly terrified by the prospect. And there are three good reasons why:
1. We measure new technology against an artificial goal of 100 percent accuracy and reliability when the current state of affairs isn't even close to that goal.
We measure the efficiency of a new technology against a standard of perfection rather than against the actual benefit it has over whatever the current risks are without the technology. Many of the arguments against AVs involve scenarios where the car might make a decision that results in injuries or fatalities. It's impossible to avoid every such incident. So, yes, autonomous vehicles will not eliminate all of the risks of motor vehicles. However, it is virtually certain that, as a whole, they will be orders of magnitude safer than human drivers. So much so that the ultimate tipping point for driverless cars will likely be economic because of the relatively high cost of insuring a human operating a vehicle over one driven by AI.
While doing research for my upcoming book, Revealing the Invisible, I spoke with a number of personal injury lawyers and insurance companies. Autonomous vehicles will have a dramatic impact on their business, perhaps putting them out of business. Surprisingly, the reactions I got to that were not what you might expect. Marc Lamber, a personal injury lawyer at Fennemore Craig PC, a 130-year-old Southwest law firm, voiced the predominant perspective:
Ultimately, self-driving cars are remarkable in the sense that they will put me out of business. I'm a personal injury lawyer and I'm glad [about that]. I don't think it's going to happen in the next year but as you look at the rapidity with which the technology is moving forward, it's an eventuality that personal injury lawyers, like myself, who represent victims of car crashes, will be out of business or our business will be reduced drastically--and it's such a wonderful thing if you think about that statistically.
One of the accepted statistics that I see thrown around is that at least 90 percent of traffic accidents are caused by human error. If we're in a stage 4 vehicle that's fully autonomous, 10 years down the road, we're going to eliminate almost all of those collisions. So, you're saving potentially a million lives a year and you're avoiding countless injury accidents.
Marc's comments echoed a recurring theme that I came across time and again. We've come to accept what I have no doubt we will soon look back on as totally unnecessary risk and human suffering.
2. We measure the known risks of a new technology against unknown benefits.
It is far easier to anticipate the risks (perceived and real) of a new technology than its future benefits. In an earlier column, I talked about how AVs would provide safe mobility for the elderly, disabled, and youth, radically changing their quality of life and the risks of currently available options. The societal value in doing that is nearly impossible to measure. Additionally, AVs will have an irrefutable record of every incident they are involved in. It will not longer be a guessing game of what happened based on the imperfect memory of those involved or of bystanders. Instead a precise account of all the factors involved will be preserved, not only for liability issues but more important for learning.
However, until we experience that new normal, the known risks far outweigh the unknown benefits.
3. The A.I. that drives autonomous vehicles learns from doing.
There's an old saying that goes something along the lines of "good judgment depends mostly on experience, and experience usually comes from poor judgment." While that may sound trite, few of us would argue that it's in the doing that we learn the most valuable lessons. The same is true of A.I. While A.I. can be trained using simulations and observation, it learns primarily by doing. And doing involves making mistakes and learning from them.
Some of those mistakes will terrify us. The difference, however, between how an A.I. driven car learns and how humans learn is dramatic, because once A.I. learns something the lesson is shared with every other vehicle using that same A.I. That means the rate at which learning positively impacts outcomes scales in a way that the proficiency of human drivers never could.
Tempe was tragic. Elaine Herzberg was one of the 3,287 people who lost their lives to automobiles on that day. Our objective should be to radically reduce that, not by retreating into the past but by moving boldly into the future, a future of setbacks and breakthroughs, tragedy and triumph. And, over time, a future in which we will share not just our roads with new drivers but our world with a new species.
Coming up in my next column about autonomous vehicles: This isn't the first time we've encountered the driverless dilemma. There's an example that's nearly 100 years old!