If you believe the optimists, self-driving cars will dominate our roads within a couple of decades. This will reduce or even eliminate such human-created ills as traffic jams and fatal accidents.
I’m not so sure.
Granted, the rise of autonomous vehicles does seem imminent, with companies such as Apple, Google, Ford and General Motors already testing small fleets in realistic conditions. Some think the US could have 10m by 2020. Experts predict that the shift away from car ownership could save 1.2m lives a year, and cut the number of cars on the road by 80%.
That said, obstacles abound. The technology still has trouble recognizing objects, making good decisions and failing safely when necessary, all of which will require extensive testing to overcome. By some estimates, level-five vehicles — able to operate without any human assistance on essentially all typical road types, and in all driving conditions — might not even reach the public market for a decade or more.
Even if self-driving cars do become ubiquitous, the benefits aren’t certain. Consider traffic — as an overview from the Victoria Transport Policy Institute points out. Freed from driving, people might demand more luxurious vehicles, in which they can work or even sleep.
The added amenities will take up more space on the road, while the comfort will encourage people to take bigger trips and endure longer commutes. They might even prefer to move more slowly, to avoid unexpected acceleration that could wake them up or tip over a wine glass. Empty cars might circle endlessly to avoid parking charges. The potential result: more kilometers driven, more congestion and increased emissions.
Even if self-driving cars do become ubiquitous, the benefits aren’t certain.
What about safety? True, driver error currently causes 90% of traffic accidents. But autonomy presents other dangers. Lulled into a sense of security, riders might not wear their seat belts, or be too busy texting to take over if the technology needs help.
Unless human drivers are banned from the roads, they can create added dangers by driving aggressively to take advantage of the comparatively cautious self-driving cars, or by trying to insinuate themselves at high speed into sleek, synchronized groups of autonomous vehicles.
Worse, the vehicles’ algorithms and network connections could create new vulnerabilities. A hacked car could become a potent, remotely piloted weapon. It’s even plausible that terrorists could multiply their destructive power by manipulating the coordinated flow of hundreds of vehicles — something that wouldn’t be possible if each were operated by a hard-to-control human.
The history of automotive technology suggests that innovations can take a long time to become practically useful. As the Victoria report points out, automatic transmissions were first developed in the 1930s, yet became reliable and affordable only in the 1980s. Airbags were introduced in 1973 but weren’t cheap and safe enough to become standard equipment until 1988 — and weren’t a regulatory requirement until 1998.
Like these technologies, self-driving cars will likely start out expensive, imperfect, and difficult to operate, and then gain wider adoption over time as they get cheaper and work better. We might see some benefits in the next couple of decades, including better mobility for wealthy non-drivers. But things like uncongested roads, better safety, greater energy efficiency, and reduced pollution probably won’t arrive until later — if they ever do.
— Column by Mark Buchanan, (c) Bloomberg LP