If you have seen the video of Google’s humanoid military robot “Atlas,” you can be forgiven for feeling terrified. The video shows it trudging tirelessly through the snowy woods, impervious to the cold or the slippery ground below its feet. Even when the thing falls over it just gets back up as if nothing happened.
The video does not say that if Atlas comes after you, it will find you and kill you. Your own paranoid mind made that bit up for itself. But Courtney Hohne, a PR person at the corporate parent of Boston Dynamics, which made Atlas, wrote an internal email that described it as “terrifying, ready to take humans’ jobs“:
We’re not going to comment on this video because there’s really not a lot we can add, and we don’t want to answer most of the Qs it triggers.
Here’s a snatch of the video she was talking about:
Google is now planning to sell Boston Dynamics, the company under Google-parent Alphabet that makes military robots, because they aren’t commercial enough.
“We don’t want to answer most of the Qs it triggers” could mean a lot of different things — most likely that it is difficult to get across the benefit of a product that can do a bunch of things humans can do when the media atmosphere is filled with humans worrying about machines taking their jobs (robots don’t steal jobs, by the way).
Still, is not a very reassuring statement from a company that makes robots for the military. At the same time, Google also demonstrated that its AlphaGo DeepMind computer was smarter than a human at the game of Go, which until recently most people felt required human intuition to win.
So these things are smarter than you, and they can chase you. Like this BigDog robot, also made by Boston Dynamics, which can jump across holes:
But you probably shouldn’t be very scared of these robots.
First, experts in artificial intelligence actually believe they are incredibly dumb. AlphaGo is great at Go, but it is terrible at everything else. “It is easier to simulate a grand master chess player with a machine than it is to simulate a 2-year-old child,” says UC Berkeley professor Alison Gopnik, who studies AI.
The second issue is logistics. Atlas might be able to chase you through a forest, but it will run out of juice in about an hour, according to the people who test Atlas. In other words, Atlas is a bit like your phone — great in the morning but it needs to plug itself back in before lunchtime.
There is a similar problem with BigDog. The robot is powered by a go-kart engine that has a 10-pound tank of fuel. The more fuel you give it, the heavier it gets, and the more fuel it uses up. It is not clear just how far BigDog can march in one go, although 20 miles seems to be the industry benchmark.
Both robots have limited range, in other words. You can read a bit more about the extent of that handicap here, where DARPA program manager Dr. Gill Pratt explains that Atlas will largely need to be tethered to a truck full of battery power for it to get anywhere. In other words, Atlas remains very, very dependent on the humans looking after it.
That might be one reason why Alphabet believes that Boston Dynamics isn’t likely to produce a marketable product anytime soon. It’s still easier to get humans to do the work.