by Jean-Louis Gassée

I want one! You probably do, too. So do millions of readers of John Markoff’s October 21st NY Times piece on Google’s revolutionary cars that drive themselves—in traffic! (More drooling on this ABC News video here.)

Is this really real, or is it a demo, another carefully choreographed PR exercise designed to enhance Google’s image as a nurturing friend of the people and their planet?

Autonomous, self-driving, driverless… The concept and implementations have been around for a while, in Europe, in Japan—the land of robotics—and, of course, in the US. In 2004, DARPA, the forefather of the Internet, launched the DARPA Grand Challenge, a driverless car competition. No entrant finished the first race, but in 2005, Stanley, Stanford’s vehicle, took the prize.

In September of this year, we saw Shelley, a driverless Audi, ascend Pikes Peak (14,115 ft/4,302 m). The car performed well although the event was marked by tragedy when helicopter filming the ascent crashed on the mountainside. The project, a joint venture between Stanford and Volkswagen’s Electronic Research Lab in Palo Alto, has since been suspended. (On a personal note, about 15 years ago I drove to the very top of Pikes Peak—slowly but easily—in a rental car. You can do it and enjoy yet another movingly beautiful part of Big Sky Country.)

How real is Google’s driverless car? Based on the DARPA and Stanford heritage, I’d say pretty real. Google was, of course, founded by two Stanford alumni, now with unlimited money and computer power. Moreover, they’ve advanced the art. Stanley and Shelley were solo drives on closed roads. Google’s car drove itself in traffic. From the Markoff article:
“…seven test cars have driven 1,000 miles without human intervention and more than 140,000 miles with only occasional human control… The only accident, engineers said, was when one Google car was rear-ended while stopped at a traffic light.”

Now that we know about the project, sighting stories surface. Robert Scoble, a noted blogger, has one. In retrospect, I now recall two mornings when I found myself driving next to the self-driving Prius on the often-packed Route 101 somewhere between Palo Alto and San Francisco. I mistook the vehicle for one of the Google Street View cars…

Henry Blodget, Business Insider’s boss, wonders if self-driving car research is a good use of Google shareholders’ assets. Perhaps the company should spin the project off to an independent venture and give the engineers a piece of the action.

Technically, the Google car is no mean feat. Take another look at how it avoids an unexpected pedestrian in the ABC video. The car uses cameras—one perched high above the roof, oodles of computing power both on-board and in Google’s Cloud, servo motors and, of course, bleeding-edge software. The mechanisms may be a bit cumbersome for now but, on paper, this is nothing that can’t be “Moore’s Lawed”, shrunk in size and cost, and debugged as the experiment moves into the millions of miles.

In practice, one has to worry about the infamous last 10%, the part of any project that consumes 90% of the effort, time, and $$. And that’s when things go well. Otherwise, it’s 110%…

In many programs, exception handling, dealing with “things that shouldn’t happen but will happen anyway”, takes most of the code and even more than most of the development and testing time. We’re not dealing with Flash Player crashes, here: Human lives are at stake. This is why Google is so careful when naming a date for the commercial availability of such a car. “Eight years or more” is the guarded answer.

The hardware and the software will fail, no question. The real riddle is determining the socially acceptable failure rate. Today, there are about 40,000 car fatalities per year. Note the euphemistic “car fatalities” or “car accidents”, as if the drivers weren’t to blame. You can imagine the news headlines when the first self-driving car fatality happens: Killer Robot! Killer Software! (A literal killer app?). Isaac Asimov, the author of the Three Laws of Robotics will spin in his grave.

Then, we have a scalability problem: for the system to work, the cars will need to continuously report their positions and events to their Mother in the Cloud. And Mother will continuously feed back indications, clues, warnings about traffic, weather and construction hazards, congestion. She’ll suggest alternate roads in real time so the entire school of fish doesn’t swim down the same primary route.

Google just established feasibility, a proof of concept—for one car at a time. What happens when robotic vehicles number in the millions? Where is the wireless bandwidth? And, perhaps, the computing power?

The answer might lie in a progressive, two-stage deployment. At first, a few cars on a few itineraries, such as well-trodden work commutes. I don’t have statistics but I’m sure most (90%?) of our drives are repeats. And, thinking of saving lives, what about the drive home after a solid—or overly liquid—dinner?

As I wrote at the beginning, I want one. It might take a long time to get to my driveway, but that’s fine, it might arrive just in time, when, in a distant future, the DMV decides I’m too old to drive. This is a politically sensitive problem. The number of aging drivers is growing, they vote, they’re often called The Grey Panthers. Quoting from a USA Today story:
“Fatality rates for drivers begin to climb after age 65, according to a recent study by Carnegie Mellon University in Pittsburgh and the AAA Foundation for Traffic Safety, based on data from 1999-2004. From ages 75 to 84, the rate of about three deaths per 100 million miles driven is equal to the death rate of teenage drivers. For drivers 85 and older, the fatality rate skyrockets to nearly four times higher than that for teens.”

But back to today: Is the “GooCar” a clever Google PR stunt? Probably…nothing wrong with that. They’re doing amazing engineering and they owe it to themselves and to their shareholders to use their feats to burnish their reputation. Not everyone agrees. Henry Blodget, Business Insider’s boss, questions the use of shareholder assets and company resources in pursuing self-driving car research. Shouldn’t the company spin the project off to an independent venture and give the engineers a piece of the action?

This got a few people riled up, such as Michael Arrington, the TechCrunch founder (see his retort here). Arrington makes two points. First, this is Silicon Valley: Let the crazies be crazies. Many of the Valley’s biggest companies/products/innovations came from the brains and loins of people who didn’t cling to the corporate handbook. Second, critics such as Dan Lyons who complain that the Valley no longer does hard science are wrong. Google, with this project and all their other hypergalactic-scale computer systems breakthroughs, goes where Man hasn’t gone before.

I’ll add my own view of Google’s quest: The company needs a business model that goes beyond their current search/advertising gambit. They know it, too. That’s why they try “everything” from Google Docs to Picasa and then watch what sticks. So far, as Blodget points out, they’re still a One Trick Pony: 95% of their profits come from search and advertising.

Pursuing the Everything for Everyone line of thought, my guess is that Google wants to become a Cloud Computing utility. Imagine a modest monthly subscription per car or per mile. Add this to other services such as communications and entertainment and the numbers can get interesting. Today, there are about 254 millions cars on US roads alone.

One way or the other, Google could get a piece of the savings in human lives, time, pollution and general aggravation, and make gigantic amounts of money as a result.

How much would you pay—monthly or per commute…or even per after-dinner drive—to be chauffeured by Sergey and Larry and their robotic avatars?