Whether running down a track, doing backflips, dancing to music or kickboxing, more and more videos are showing humanoid robots doing increasingly impressive things.
However, Fortune magazine speakers Artificial Intelligence Brainstorming Session Tuesday warned against getting too dazzled by acrobatic displays. A robot doing a backflip – a difficult thing for a human to do – looks impressive. But getting a robot to perform seemingly simple tasks, like climbing stairs or grabbing a glass of water, is still difficult for many of today’s robots.
“Things that look hard are actually easy, but things that look easy are actually hard,” explains Stephanie Zhan, a partner at Sequoia Capital, citing an observation by computer scientist Hans Moravec. In the late 1980s, Moravec and other computer scientists pointed out that computers were more likely to perform well on intelligence tests but not perform well on tasks that even young children could perform.
Deepak Pathak, CEO of robotics startup Skild AI, explained that robots, and computers in general, excel at performing complex tasks when run in a controlled environment. “The robot doesn’t interact with anything except the ground,” Pathak noted while showing a video of the Skild robot jumping on a sidewalk.
However, Pathak explained that for tasks like picking up a bottle or walking up stairs, a person is using vision to “constantly correct” what he or she is doing. “This interaction is the root cause of general human intelligence, and you don’t appreciate it because almost everyone has it.”
Zhan explained that the viral video of the humanoid robot did not show how the product was trained or whether it could operate in an uncontrolled environment. “The challenge you face as a consumer of all these videos is to really discern which ones are real and which ones are fake,” she said.
What’s next for robots
Still, both speakers were optimistic that advances in general intelligence will soon lead to more advanced, more flexible robots.
“In the past, robots were driven more by human intelligence. A super-intelligent human would observe (a task) and then…pre-program the robot mathematically to complete it,” Pathak said.
But now, the field of robotics is moving from “programming” to “learning from experience,” he explains. This allows new robots to handle more complex tasks in less controlled environments and to be easily adapted for other tasks without having to reprogram and re-equip them.
Stuart Isett Fortune Magazine
Zhan believes that today’s robotics companies are “still constrained by making robots only for specific things.” Robotic platforms with more general intelligence could open up “possibilities beyond our reach,” including tasks that are currently dangerous for human workers.
Consumers benefit too. “You see all these home robots, but they can only do one thing,” Zhan said. “But if we succeed in building general intelligent robots, you will eventually have consumer robots that can handle all the household tasks you have now.” A similar point was made earlier at Brainstorm AI Qualcomm CEO Rene HaasHe said the general adaptability of humanoid robots would make them better suited for factory work than the robot arms used today.
The boom in robotics will have a social impact, replacing jobs that currently require humans. However, Pathak is optimistic about the social benefits of spreading automation. One is safety, because in the long run, robots eliminate the need for humans to perform dangerous or unhealthy tasks. Another benefit is the huge labor shortage to fill blue-collar and manufacturing jobs. (This gap has been an obstacle to U.S. efforts to reshor much of its advanced manufacturing from Asian economies.)
Yet Pathak also envisions a future in which robots free humans from the drudgery of daily work, although he acknowledges that society needs to figure out how to spread the benefits of automation. “There’s a scene, a good scene, where everyone is doing what they love,” Pathak said. “The jobs are more selective and they’re doing something they love.”