How long does it take for a robot to wash your dishes?
Imagine the biggest market for a physical product that you can. Thinking about cell phones? Cars? Ownership? These are all massive markets, but in the coming decades a new product will be launched to eclipse these giants, says Geordie Rose, CEO of Sanctuary AI. The Vancouver-based firm is developing a humanoid robot called Phoenix that, when finished, will understand what we want, understand how the world works, and have the ability to carry out our commands.
"The long-term total addressable market is the largest it's ever been in the history of business and technology — which is the labor market. Those are all the things we want to do," he says. Before we get too ahead of ourselves, he clarifies this statement: "There's still a long way to go from where we are today."
Mr. Rose isn't willing to put a time frame on when a robot could be in your house, doing your laundry or cleaning your bathroom. But others I've spoken to in the sector say it could be within a decade. Dozens of other companies around the world are working on the technology.
Perhaps the most famous company on the market is Tesla, Elon Musk's electric car company. We'll see if that's the case. We can now say that leaps forward in the field of artificial intelligence mean the acceleration of the development of humanoid robots.
"Ten years at the rate technology is moving now is an eternity. You know, every month there's a new development in the world of artificial intelligence that's like a sea change," says Mr Rose, who has a background in theoretical physics and previously founded a quantum computing company.
Mainstream interest in AI exploded late last year when a powerful version of was released. Its ability to generate all kinds of useful text and images has spawned rivals and a wave of investment in AI technology. But developing AI to enable a robot to perform useful tasks is a different and more difficult task. Unlike their rivals, humanoid robots must navigate the physical world and need to understand how objects in that world are related.
Tasks that seem easy to many people are the main achievements of humanoid robots.
.jpg "which Mr Musk says could be on sale to the public in a few year's time.")
For example, in a test project, Sanctuary's Phoenix robot packed clothes into plastic bags in the back room of a Canadian store. “This is a problem that affects a lot of different complex problems in an AI-driven robotic system because the bags are pliable, they're transparent... there's a place where they open.
"Usually after you open a bag by hand, you have to free one hand and then go and put something in the bag," says Mr Rose. "Handling bags is actually very, very difficult for robots," he adds — a line that makes today's humanoid robots seem far less terrifying than some of their Hollywood counterparts.
Sanctuary has a system for training Phoenix for specific tasks such as packing bags. In cooperation with the company, he films the specific task being performed and then digitizes the entire event. This data is used to create a virtual environment that simulates physics including gravity and drag in addition to all objects.
The artificial intelligence can then practice the task in this virtual environment. It can have a million tries, and when the developers think the AI has mastered an event in the virtual world, it will be allowed to try it in the physical world. In this way Phoenix was trained to do about 20 different roles.
Mr Rose sees this as the way forward for humanoid robots – mastering specific tasks that will be useful for business. A robot that can handle housework is a long way off. One of the biggest challenges is giving the robot a sense of touch so it knows how much pressure to apply to the object. "We have devices with these types of tasks that come from an evolutionary heritage that's about a billion years long ... they're very demanding on machines," says Mr. Rose. There is still a huge amount of work to be done to build a robot that can handle all the events that could occur in a home or busy workplace.
"You can't put a robot in an unstructured environment and then ask it to move without basically destroying things. There's too much on the technology at the moment," says Prof. Alireza Mohammadi, who founded the Robotic Motion Intelligence Lab. University of Michigan-Dearborn.
He points out that you can subject an AI to millions of training scenarios, but in the real world there's always the chance that it will encounter something it's never seen before and react in an unpredictable and possibly dangerous way. Part of the problem, he says, is that people intuitively understand connections and consequences. For example, we can guess that an overexcited dog may jump in front of us and count on that.
Building it into a robot is extremely difficult.
"Within ten years, we could have robots that are able to walk with some guidance, but not in completely unstructured environments," says Professor Mohammadi. But if these challenges can be overcome, could humanoid robots begin to take over the jobs currently performed by humans?
Mr. Rose points out that there is a shortage of workers in many countries, and his robots could one day fill those positions. Stewart Miller is chief executive of the National Robotarium, a partnership between Heriot-Watt University and the University of Edinburgh that focuses on AI and robotics.
"Inevitably, robots will be doing jobs that were done by human beings in the past...the question then is, what does that mean?" he says. "We're going to go through some growing pains. But if we think about it, we can start to emphasize and focus on what human beings do best — free up that capacity to do that and not have to spend time doing what machines do best."
0 Comments