The Cat's Meow
Kindred Systems wants to design artificial intelligence that learns like people do. No pressure
At the recent Bctech Summit, visitors couldn't resist the Kindred Systems Inc. booth. No wonder: the company had a corral of furry robotic cats on display. On hand to supervise these strangely fetching felines were co-founder and CEO Geordie Rose, along with cofounder and chief scientific officer Suzanne Gildert.
The cats are just one of a variety of robots built by Kindred, whose goal is to create machines with humanlike artificial intelligence. With AI, there are two key parts to a robot body, Rose explains: cameras and other sensors that perceive what's going on around it, and the physical actions it can take. “The core technology that we're developing is a software layer that sits between the perception and the action,” he says.
Unlike humans and other animals, most machines lack general intelligence; their actions are predetermined, so they can't adapt to changes in their environment. To create a so-called strong AI that will work on any machine, Kindred's software uses reinforcement learning. “You specify what you want as an outcome but not the process to get there,” Rose
explains. The robot then experiments,
and the things that work stick: “It continually gets better and better and better at the tasks that you've given it, even complex tasks.”
If that sounds daunting, don't bet against Rose, who holds a PHD in theoretical physics from UBC. Before launching Kindred in Vancouver in 2014, he co-founded Burnaby-based D-wave Systems Inc., developer of the first commercial quantum computer. Kindred, which is backed by Google Ventures, has since moved its head office to San Francisco, where most of the 40-strong team work. The company's eight Vancouver staff handle the bulk of AI research.
Kindred makes humanoids as well as quadrupeds, but its biggest robot family consists of one-armed, onehanded machines suitable for dextrous manipulation and grasping. In March, the company was preparing to launch a product based on that body type. From supermarkets to factories, grasping objects and moving them is a core function for much of the workforce, Rose notes, but such tasks are tough to automate. “Those are the ones that we're focused on as a business, to go after the things that are physically easy, that require thinking and judgment and care, delicacy when it's needed, but strength and power and speed when it's needed.”
Kindred wants to start by deploying its software platform in a specific vertical industry. Once that product gains traction, the plan is to do the same in other verticals. “There are hundreds or thousands of opportunities to build machines with general intelligence,” Rose says. “Each of these is a multibilliondollar opportunity.”
Although he concedes that predicting the future is difficult, Rose doesn't think robots will make humans redundant anytime soon. “The type of work that is going to be automated in the short term opens up all sorts of other possibilities,” he argues. Automating certain tasks at a warehouse will boost profits, Rose says, so the owner can hire more people to do things that machines can't. “There's going to be a net job gain in the manufacturing and distribution centre businesses.”