Google’s Astro Teller. (Rob Pegoraro/Yahoo Tech)
AUSTIN — Google’s first “Project Loon” wireless Internet balloons were designed to explode. And Google’s own employees convinced designers of its self-driving car that it would be better if the car didn’t have a steering wheel for backup control by humans.
Astro Teller, the “Captain of Moonshots” at the five-year-old Google X research lab, shared those tidbits during an hourlong “Moonshots and Reality” SXSW keynote address that focused on the “bumps and scrapes” those high-risk, high-reward projects endured, and what they taught the engineers involved.
“The longer you work on something, the more you don’t really want to know what the world is going to tell you,” Teller said. The only way to defeat that impulse is to get your product into the real world — “running at all of the hardest parts of the problem first and as fast as we possibly can.”
The Internet in the sky
Take Project Loon, Google’s scheme to provide wireless Internet access over vast stretches of the developing world with a flock of balloons in the stratosphere. At this point, Google can get the balloons to stay up six months at a time, but at first it couldn’t ensure they would stay in the right countries.
The answer, Teller said, was to make them out of brittle latex that would explode if the balloon was sent up to about 100,000 feet. Then the company would have to pick up the remnants: “Going out to the Arctic Circle in a helicopter to stuff the balloon in the back of the helicopter … was not how we wanted to be spending our time,” he said.
A Project Loon balloon in 2013. (iLighter/Flickr)
Later on, Google X engineers had to deal with persistent leaks in the polyethylene of the balloons. “We were so desperate that we ended up doing a detailed study about the fluffiness of the socks of the techs who did the balloons,” Teller said, because they had to walk on the balloons when they were assembling them. And it mattered; when techs wore fluffier socks, they had fewer balloon leaks.
Don’t trust humans
Teller — born Eric Teller, his “Astro” moniker dates to when a flattop haircut earned him the nickname “Astroturf” — also related lessons learned from Google’s self-driving car venture. Key among them: Don’t trust the humans.
Google had grown confident enough in its self-driving car software that it invited Google employees outside the project to use it for automated highway commutes, with only surface-street driving left to them. They had to promise to pay attention to the road at all times, though. But that didn’t work.
“People do really stupid stuff when they’re driving,” Teller said. “The assumption that humans can be a reliable backup for the system is a total fallacy.” That led to the design it revealed last May, without a steering wheel or brake pedals.
Google’s self-driving car prototype. (Google)
Google’s still racking up test miles in the hope of picking up additional real-world possibilities. One example: What should the car do if there’s a woman in an electric wheelchair in its path who is shooing a duck across a suburban side street? In this unplanned, real-world case, the car stopped before the human “safety driver” had to intervene.
Know when to fold them
With the former, the problem was spending too much time on a flawed design for the vehicle. Teller said, “80 percent of the team, including me, knew it was a failure after a year and a half, and we still hadn’t killed it.” Now that an arbitrary five-month deadline from Google co-founder Sergei Brin pushed the team to get a first round of flight tests done, it’s working on a new design Teller didn’t describe.
With the kitelike Makani flying wind turbines, the failure has actually been the lack of crashes by small prototype vehicles. “We failed to fail,” Teller said. Tests of the full-size design, which sends 600 kilowatts of electricity down an 84-foot tether to a ground station, start next month.
Google X’s most public venture so far has been its Google Glass eyewear. Teller didn’t skip over that, but his answer may not satisfy critics of the technology.
A mustachioed man wearing Google Glass. (Getty)
“The thing that we did not do well and that was sometimes a failure was that we allowed and encouraged too much attention to the program,” he said. But how exactly were Glass “Explorers” ever going to wear computers and cameras on their faces with any subtlety?
Teller didn’t say what the next generation of Glass will be like. But if it exists at all, you’ll see it soon enough in public, where Google can hear from its potential customers.
“Let them tell you what’s wrong with it,” he said. “You’re never going to get the right answer sitting in a conference room.”