You Are a Better Driver Than a Tesla

What Autopilot is really like

My commute is a little more than an hour each way. I’ve been eagerly tracking the progress of the Google Car. I can’t think of any reasonably feasible technology that would improve my life more than a car that would let me read, work, or take a nap on the way to work. And then there are all the benefits to society: fewer accidents, better fuel efficiency, less traffic congestion, etc.

I also happen to own a Tesla. About four weeks ago, I got in my car one morning and it said it could drive itself. Enough friends have asked me what this is like that I decided to describe the experience here.

A couple of caveats. First, I’m not a car person. I see the car as a way to get around, and my first three cars were a Saturn SL1, a Chevrolet Prizm, and a Subaru Impreza. I bought the Impreza because it had the highest fuel efficiency of any AWD car at the time, and I bought the Tesla for the same reason. Second, Autopilot is just adaptive cruise control (which many cars have), plus the ability to keep you in your lane (which some cars have), plus the ability to shift lanes by activating the turn signal (which is really just a parlor trick since, as described below, you still have to check behind you). Elon Musk isn’t claiming that it’s truly self-driving, and the manufacturer recommends that you only use it on the highway, with your hands on the steering wheel. But it still teaches a potentially important lesson about true autonomous driving and when it will really be available.

First, let it be said:

Tesla Autopilot is a technological marvel.

Turn it on and watch the steering wheel turn by itself. It can handle sweeping turns at high speeds, and I think it’s smart enough to slow down if it thinks it can’t handle a turn at the speed you have set. (I’m not certain about this, but I think I observed it once.) The cruise control has gotten better about slowing down when someone cuts in front of you. (Before, I had the distinct feeling it wouldn’t stop in time.) The side and corner sensors notice cars or barriers that are close to you on either side, and the car adjusts by steering you slightly to the other side of the lane. If you activate the turn signal, it will watch for a gap in the adjacent lane before switching into that lane smoothly. I’ve driven from New Haven to Amherst—about 80 highway miles—with Autopilot on about 95% of the way.

OK, here’s the catch:

What Autopilot does is the easiest part of driving.

It’s the things that you can already do instinctively, without thinking, and while carrying on a complicated conversation—on autopilot, as it were.

By contrast, here are some things you can do better than a Tesla:

• Anticipating possible collisions: If I’m in the left lane of a three-lane highway and I want to switch into the middle lane, I look all the way over to the right lane to see if there’s a car that might want to switch into the same position. It doesn’t happen often, but two cars simultaneously switching into the same apparently empty space is a predictable cause of a collision. I’m pretty certain the Tesla doesn’t look two lanes over, since the sensors on the sides of the car are only good for about fifteen feet.
• Keeping the optimal position in your lane: Sometimes I’ll be in the left lane of a highway with a barrier or guardrail on my left and big trucks on my right. The Tesla’s sensors are better at picking up the barrier (because it’s constant) than the trucks, so the car drifts toward the right—away from the barrier and toward the trucks. This is more dangerous, however, because the barrier isn’t moving and the trucks are under human control. A human being would drive closer to the barrier than the trucks, which is safer.
• Avoiding road debris: I don’t know how big a piece of debris (say, a piece of plywood with nails sticking out of it) has to be for the Tesla to pick it up, and if the car senses debris in front of it, I believe it will just stop, which is not particularly safe on a highway. A human being is better at seeing the debris and driving around it, which is safer.
• Driving without lane markers: Obviously, Autopilot needs lane markers, and it won’t activate if it can’t see at least one line. But if the lanes go away unexpectedly—because of roadwork, or glare from water on the road, or because Kramer painted them over—it’s lost. By contrast, you can adapt.
• Most importantly, watching for dangerous drivers: If you are switching into the left lane and a car behind you is zooming up in that lane at 90 miles per hour, you will see it and not change lanes. The Tesla won’t see it. At least, I don’t think it will. I base this on two pieces of evidence. One is that my Autopilot has tried to change lanes into a space that was rapidly being occupied by another car. The other is that the Tesla’s side and rear sensors don’t have a long range, and I don’t have any reason to think the Autopilot is using the rear camera. Maybe it is, and maybe the Tesla would spot that speeding car. But I’m not going to test it in real world conditions, and unless I’m certain Autopilot can avoid that collision, I’m going to look behind me, and at that point Autopilot isn’t doing much for me.

The reasons you’re a better driver than a Tesla are pretty simple. You can see farther and in more directions, since the Tesla only has two cameras, one of which probably isn’t used for Autopilot anyway, and a set of short-range ultrasonic sensors. And you have a lot more experience with real world driver behavior, so you can anticipate potential problems that the computer can’t—at least not yet.

That said, Autopilot is better than a distracted driver. If you’re going to be texting, or eating a big sandwich that requires two hands, or picking up a toy your kid dropped in the back seat, or using Google Maps on your phone because Tesla’s built-in navigation is prone to random acts of stupidity, or taking a picture out the window, or anything else you shouldn’t be doing, then you’re definitely better off turning on Autopilot. That way you won’t swerve into another lane and hit another car.

But the problem is that, as currently designed, Autopilot requires you to watch what’s going on whenever you’re in relatively crowded traffic. It’s not just that it might malfunction, as some YouTube videos purport to show. It’s that when you have heavy traffic, aggressive drivers, and poor road conditions, you are a much better driver than Autopilot, assuming that you’re paying attention to your surroundings.

Yes, duh, but Autopilot will get better, you say.

Of course it will get better. But it has to get a lot better before it will significantly improve my quality of life. It isn’t enough for it to be technologically better.

I have to trust that it can drive better than I can.

This may sound preposterous. Of course a computer can drive better than a human being, right? Let’s take the example of a car cutting in front of you on a crowded highway. The Tesla has gotten better at responding quickly, and obviously its reflexes are faster than min. But a human being can see the twitchy driver one lane over before he actually cuts into your lane and take preventive action. Right now, I don’t know just how good Autopilot is at avoiding these collisions, which means I have to continue watching the road—and as long as I have to continue watching the road, my driving experience is pretty much the same as it always has been. Autopilot will get better and better at tracking cars in adjacent lanes and predicting dangerous behavior, and at some point it will be better than I am. But how will I know when that point has arrived?

It isn’t just a matter of getting used to the technology. Even after driving, say, 50,000 highway miles with the Tesla, that won’t prove that it’s a particularly good driver. I’ve driven a lot more than that in my lifetime and never gotten into a highway accident (when moving; once I got rear-ended while sitting still in a traffic jam).

Probably, at some point in the next five years or so, there will be enough data to prove that Teslas on Autopilot get in fewer accidents per mile than human-controlled Teslas. That still won’t be particularly convincing, since most of those human-caused accidents will be due to easily avoidable driver error: texting, being drunk, surfing the web (why isn’t the browser disabled when the car is moving?), etc. Measuring the accident risk of an alert, safety-conscious driver is not an easy proposition.

The Google Car apparently has gone hundreds of thousands of miles without getting in an accident—that was its fault. I used to be amazed by that statistic. Now, after driving my Tesla on Autopilot, I think: so what? I don’t want to get in any accidents, whether my car’s fault, someone else’s fault, or no one’s fault. If a car cuts dangerously close in front of me and Autopilot can’t stop in time, that’s not a good outcome, even if it’s the other driver’s fault. What matters is the absolute accident rate and, again, you have to compare it to good drivers, not all drivers. (If I have to pick up a toy on the floor of the back seat, I turn on Autopilot first. In that way, Autopilot definitely increases my safety—but, again, it doesn’t change the basic experience of driving.)

So what’s the lesson of Autopilot? There is a big difference between a car that can do the easy parts of driving—staying in a lane, not hitting the car in front of you—and one that can drive as well as a human being. At some point, cars will driver better than humans. But for a reasonably long time, the safest combination will be a car using autopilot (which will eliminate the accidents caused by driver distraction) combined with a human being watching out for other stupid, distracted, or drunk drivers (who will minimize the accidents caused by other bad drivers). That’s great, but as long as the human being needs to be looking out the window, autonomous driving will be a useful safety technology—not a revolution in the commuting experience.

And I won’t be able to take that nap.