“Data Driven”: In a Car, Is the Driver a Bug… or a Feature?

“Data Driven”: In a Car, Is the Driver a Bug… or a Feature?

Driver: Bug or Feature?

"The fact that you're still driving is a bug," [Google's Anthony] Levandowski says, "not a feature."

I'm sure there'll be the inevitable responses about computer crashes vs. auto crashes, but Levandowski has a point. Read the sentence again, not as "the driver is a bug" but as "the fact that you still have to drive is a bug."

Once upon a time, driving was the reason for getting in a car. Sometimes it still is, but most of the time driving is simply something we have to do in order to get the car containing us from Point A to Point B. When I worked at Microsoft, I often faced a 90 minute commute coming home... all of 13.9 miles. (Yes, that's a decimal point, not a smudge on your monitor.) Driving was certainly no fun on that commute; in fact, I eventually gave up my stick-shift car for an automatic because I didn't know which would wear out first: my left leg, the clutch, or my patience.

Think about how much we do at work that we really would like to be able to turn over to an intelligent, semi-autonomous agent. Yes, some of us have secretaries and assistants and the like; still, a certain amount of what we do is not, to put it mildly, intellectually challenging to us. Apple Siri, to the extent it works (which is to say, not very well at all in real life), is a start on this process for a few small areas. But if we think differently about the problem, we can conceive of far better solutions.

Email, for example, was one such solution 20 years ago. Internet (and Lexis-Nexis) search is another. There will be more.

Just as there will be robot-driven, semi-autonomous cars.

Will they crash? Undoubtedly less than the humans they replace.

Of course, people, being people, will notice the occasional crash far more than the steady successes... and will have forgotten how much more often they used to crash.

What Data Driven Means

I got so tired in the corporate world of hearing, "I'm data driven." I rarely told such people that they were no such thing, preferring not to make needless enemies.

But they weren't.

They were just subjective in different ways... and self-deluded to boot about their blind spots.

Here's what data-driven really can mean:

The road is just another data set to be mined. So Google isn't teaching its computers how to drive. It's collecting data-its cars have driven 200,000 miles in total, recording everything they see-and letting its algorithms figure out the rules on their own. "If you read the DMV handbook on four-way stop signs, it's easy," [Google'ss Chris] Urmson says. "Whoever gets there first gets to go. If there are simultaneous arrivals, priority goes to the vehicle on the right." But it rarely works that way. "People optimize stop signs," he says.... "[A robot]  needs to learn how people really drive. "This is the data-driven viewpoint," says Sebastian Thrun, the Stanford roboticist who heads the self-driving project. "The data can make better rules"

It's not making selective decisions based on interpreting the data. Rather, it's truly letting the data define the rules.1

That will play out in ever more interesting ways.


1I have way, way oversimplified here. I worked in artificial intelligence research in the mid-1980s and can go into this in far more detail than is necessary; please accept the oversimplification as an attempt to cut to the chase.