Geordie Rose has a tough act to follow – himself.
His last startup, Burnaby, B.C.-based D-Wave Systems Inc., created the first quantum computer, a machine expected to some day solve problems well beyond the capacity of today's fastest computers. Investors include Amazon.com Inc.'s Jeff Bezos and the U.S. Central Intelligence Agency's venture capital arm, and Google Inc. and Lockheed Martin Corp. are customers. Time called it "the Infinity Machine" in a 2014 cover story. Dr. Rose, who co-founded D-Wave, once said his quantum computer "feels like an altar to an alien God."
That was just a warm-up act. For the past two years, the 44-year-old Dr. Rose – a visionary theoretical physicist, two-time national champion wrestler and Canada's answer to Elon Musk – and a small team spread among Vancouver, Toronto and San Francisco have been working on a secretive startup, Kindred Systems Inc., built around another groundbreaking technology whose impact will be even more transformative: Robots that can think for themselves.
Little has been revealed about Kindred – until this week. In a wide-ranging interview, Dr. Rose and his team outlined what Kindred – backed by $15-million (U.S.) from some of California's best-known venture capital firms (including Google Ventures and early Uber financier Data Collective) and employing some of the top names in the emerging field of "deep machine learning" – intends to do: build robots with "human-like intelligence" that can "understand and participate in our world, with the ultimate goal of a future where intelligent machines work together with people to create abundance shared by all," according to its mission statement.
"The business plan I invested in was one page long," says Canadian entrepreneur and angel investor Dan Debow, who backed Kindred's $3.7-million (Canadian) seed-funding round. "It said, 'What we are trying to build is Data.'" Kindred wasn't just talking about "Big Data," a key piece of the puzzle, but Data – the humanoid character from Star Trek: The Next Generation.
KINDRED ROBOTS IN ACTION
Kindred's founder and chief technology officer, Suzanne Gildert, who invited Dr. Rose to become chief executive officer after working for him at D-Wave, said the 34-person company is "going after the original goal of artificial intelligence … of having humanoid robots that behave just like we do but live amongst us. Examples might be Data from Star Trek … or C-3PO. We want to build that thing and we want to think about how to actually solve the engineering problems you need to solve to get to that vision."
But in the near term, Dr. Rose says, the goal is to build "a profitable business as soon as possible" by deploying its technology to do things commercial robots haven't been able to do before: autonomously perform tasks in factories and warehouses beyond the typical rigid, highly specific, repetitive tasks they now do. Kindred robots will be able to autonomously sort, manipulate and move items, work still largely done by humans. He declined to say whether any potential customers are testing the technology or what industries Kindred is targeting.
"One of the things I've learned through this process is how commercially valuable moving things from one bin to another is," Dr. Rose said. "It's an absolutely enormous market. … We will look for applications of this where robotic arms are currently being used or human arms are being used to perform a task which is, essentially, recognize the object, pick it up and move it."
Kindred is guided by a belief that robots can only learn intelligence by physically interacting with the world, led by humans who will train and continue to interact with them.
"We have a hypothesis that intelligence requires a body," Dr. Rose said. The team has also been enabled by the increasing quality and falling price of technology – including more human-like robots – the availability of massive amounts of data and the ability to store it on cloud computers, Dr. Gildert said. The company is working with about 50 commercially available robots and off-the-shelf virtual reality hardware in its test labs.
Kindred's approach is different from that of other companies and researchers in the artificial intelligence field, said Matt Ocko, managing general partner with Data Collective, a Silicon Valley venture capital firm that has backed the startup. "We saw no other company or academic group advancing in the same way," he said.
Kindred's system involves human operators remotely controlling robots to perform tasks – sometimes with a mouse, and sometimes by wearing a robot suit, whose motions and actions are mimicked remotely by the robots.
Mr. Debow said the "immersive telepresence" experience of wearing an early version of the "exosuit" two years ago convinced him to invest and that Kindred would change the world. As he manipulated the robot through his motions, he could hear what the robot was hearing, see what it was seeing – and when he turned its head, found himself staring across the room at himself, through the robot's field of vision.
"It was like my consciousness had transferred" to the robot, he said. "My brain exploded. The phenomenon of [almost teleporting] to the robot so quickly becomes so immersive it made me understand how it would become such a massive thing."
As these interactions are happening, Kindred's "machine learning" algorithms – built to self-write programs – capture everything about what the operator is doing, and how the robot is responding, allowing it to infer goals, automate the behaviour "and come up with a way to do it without the human in the loop at all," Dr. Rose said.
Humans in action, it turns out, make for excellent sources of "quality data" to teach the machines how to work like them. "It just speeds up the whole process of learning," said Dr. Gildert, who holds a PhD in experimental physics from the University of Birmingham.
If the thought is terrifying in an era when smarter machines are already recommending what books to read, powering driverless cars, beating Jeopardy! champions – and poised to replace millions of workers in the coming years – Dr. Rose brushes off any dystopian fears, saying Kindred's technology will work in concert with humans, augmenting what they do, not replacing them outright.
"The idea that human labour would be replaced by a machine doesn't seem right to me," he said. "What I see is that the same number of workers can accomplish more with this sort of technology. … These types of jobs, the ones we're interested in, are bottle-necked by the ability of people to perform the task."
Kindred's other investors include Bay area venture capital firms First Round Capital and Bloomberg Beta, XPrize founder Peter Diamandis's Bold Capital Partners from Los Angeles, and early Google engineer Jeff Dean. Mr. Diamandis is also on Kindred's advisory board, along with Russ Salakhutdinov, Apple's director of artificial intelligence research, and Yoshua Bengio, an artificial intelligence pioneer who has launched an incubator in Montreal, Element AI, to develop artificial intelligence companies from the University of Montreal, where he teaches.
"Geordie is fantastic and he thinks very big," said Shivon Zilis, a Canadian-born partner at Bloomberg Beta who participated in Kindred's $13.2-million (U.S.) series A venture round last year. "We need more Geordies in Canada."