My three-year-old daughter takes her friend’s power wheels car for a first-time spin. I briefly show her how to turn the steering wheel and where the pedal is to move the car move forward. Within a few minutes, she has it down. She successfully navigates around sidewalk lamp posts, pauses for passers-by, and finds the horn. It strikes me: humans are good at driving.* This suggests that developing a successful fully automated vehicle is going to be really difficult, with ramifications for the public policies that support safer roads.
Let’s start with how humans are good at driving. Driving a vehicle is rather intuitive, like walking or running, only you control your speed and steering differently. Walking is a physical and social skill that has evolved in humans and other animals for millions of years. Our brains are designed to understand how to navigate, how to avoid or deal with unfamiliar objects in our paths, how to maneuver in crowds full of people. Driving is similar, and our brains are adept at picking it up. Give a teenager the keys to a 4,000-pound piece of machinery and most can learn to operate it in an afternoon.
Per the asterisk, (*) just because humans are naturally good at driving does not mean that it is safe. 39,000 people are killed in the United States by drivers ever year with many thousands more are injured. But this is because we put humans in situations that far exceed our brains’ capacities for making smart decisions. 70 mile per hour speed limits. 20 lane freeways with merging and exits. Bars located on two-lane rural roads. Hostile environments for walkers and bikers trying to get to where they need to go. Vehicles with limited visibility.
Safety is one of the most alluring potential attributes of automated driving. Theoretically, programming a car to follow the set rules of the road and anticipate future moves is something a computer is designed to do. Computers can have dozens of image sensors, can review the environment multiple times per second, do not get distracted or tired.
Computers are impressive to humans because we watch them do tasks that the human brain is not capable of doing. It takes hundreds of hours of practice and studying for a human to become an average chess player. But computers have been able to beat the world’s best human chess players since the 1990s and have grown powerful enough to easily win other games that are challenging to humans like Jeopardy or Go. We as humans are impressed by the capacity for a computer to think hundreds of moves ahead and calculate all possible outcomes.
But we far underestimate and undervalue the impressiveness of our own social and navigation skills. Take for example, a waiter in a crowded bar. You order a round of tall beers for you and your five best friends. The waiter fills the six glasses, balances them on a tray in one hand, and navigates a slippery floor in a dimly lit room while knowing how to avoid drunken groups of revelers, not split between two people who are taking, and bring the drinks to you without spilling anything. Society undervalues this skill in part by setting the wage floor below minimum wage, hoping they make up the difference in tips.
Ultimately driving is much more like carrying a tray of drinks than it is playing chess. We need to appreciate the fact that driving is based on social and navigational skills that humans are adept at after millions of years of evolution. We need to recognize that driving is not as simple as making the rules of the road into an algorithm and training artificial intelligence on edge cases.
What does this all mean for automated vehicles and road safety?
The current plateau of progress on vehicle automation is already resetting expectations. The fact that the way computers are wired means they simply cannot adopt the complex, socially oriented skills that are innate in humans. This means dim prospects for automated driving in the uncontrolled by common roadway environments that include weather events, pedestrians, bicyclists, and other human drivers who don’t always follow the rules. In the near and medium term, partial automation can help assist drivers avoid certain crashes and full automation might be best suited for highly controlled situations.
Years of more intensive testing might yet yield an automated driving system that can be trusted in the wild. I sincerely hope that full automation arrives in meaningful form one day, and we should certainly prepare our policies to accept it when it does.
In the meantime, we need to address the safety crisis in front of us by leveraging humans’ ability to drive while simultaneously simplifying the challenging roadway environments in which today’s drivers operate. Reduction in speed limits, roadway designs that discourage dangerous behavior, smart traffic enforcement, eliminating distractions, and better vehicle designs are straightforward changes that should be encouraged and implemented. And of course, it wouldn’t hurt to enact policies that make it easier to get out of the car altogether.
The views expressed above are those of the author and do not necessarily reflect the views of the Eno Center for Transportation.