Artificial Intelligence Is Better Than None

The carmaker Tesla has been in the news lately, and not in a good way. I’m not referring to business news about production holdups of the sort that occur in all manufacturing from time to time (recently in the Ford F150 line, for example), but to accidents involving the Tesla Autopilot. By the numbers, this self-driving feature is safer than a human driver. Much safer. Among human-driven cars in the U.S. there is 1 fatality for every 86,000,000 miles driven. For Autopiloted Teslas there is 1 per 320,000,000 miles – a 270% improvement. Yet, accidents do happen. In the past couple of months they include non-fatal collisions with a parked police car and a fire truck as well as a fatal collision with a dividing barrier: all on Autopilot.

Tesla X interior

Self-driving capability is a form of Artificial Intelligence, and accidents by AIs tend to be of a different character than those by people. People get distracted. They misjudge distance, speed, acceleration, risks, and time. They are careless. They deliberately take chances because of impatience or just for the sport of it. AI is great at speed/time/distance judgments, it doesn't get distracted, and it doesn’t know how to be careless. AI is not very good, though, at comprehending and responding to novel situations. People can tell the difference between a road and a road hazard in pretty much all circumstances, no matter how unusual. AIs might have trouble recognizing the difference in unfamiliar configurations. People often are foolish drivers, but they generally know when they are being foolish and taking risks. AIs don’t have a clue. This means that the AI accidents that do happen probably wouldn’t have happened had a human been driving. That more lives and property are saved by AI drivers than are lost to them provides no comfort to the victims. Still, we shouldn’t lose sight of the numbers, and the self-driving systems get better and less accident-prone each year.

The tech has come a long way since the early 2000s. In 2004 DARPA offered a $1 million prize to the winner of the Grand Challenge, a 150 mile course in the Mojave Desert to be driven by totally autonomous self-driving vehicles. Not a single one of the 15 vehicles that entered the race finished it. Three never made it past the starting line -- one of those three flipped over. In the 2005 race five out of 23 finished, though one of those exceeded the maximum 10-hour time limit. By 2007 the tech had advanced so much that the desert no longer was deemed sufficiently challenging, so the competition became the Urban Challenge complete with traffic lights and 4-way-stops. All 11 teams completed the course.

AIs, however, do not think like people and probably never will. The limitations of AI at making “common sense” judgments are what concern many folks not just with regard to civilian cars but with regard to the growing number of autonomous military robots. Said UN investigator Christof Heyns, “a decision to allow machines to be deployed to kill human beings deserves a collective pause worldwide." Yet, the military situation is analogous to self-driving cars. Presumably war machines won’t be unleashed except in combat situations, and in those circumstances they are less prone than people to friendly fire or to misidentifying targets – and they never act in anger or from fear. They, as The Economist noted, "have the potential to act more humanely than people. Stress does not affect a robot's judgment in the way it affects a soldier's." Robots, in short, are kinder: in effect, that is. They don’t understand kindness or cruelty as such. For that they would need consciousness – the meta-state of not only knowing but knowing that one knows – and outside of science fiction they are far from having that.

(Whether robots ever could be better than people at love as well as at war is a discussion I’ll leave to others, e.g. roboticist David Levy who authored Love and Sex with Robots: The Evolution of Human-Robot Relationships.)

As for cars, I’m not willing to surrender the steering of mine just yet. I don’t even like to use Cruise Control even though every vehicle I’ve bought for the past 25 years has had it. Nonetheless, it suits me just fine if all the other cars on the road are self-driving. I would feel happier and safer.

Radiohead – Killer Cars