Technology challenges: robots do somersaults, but can’t climb stairs

Whether running on a track, doing a backflip, dancing to music or kickboxing, there are more and more videos of humanoid robots performing increasingly impressive feats.

Still, speakers at the Fortune Brainstorm AI conference warned against getting too dazzled by the stunts. A robot doing a backflip — something difficult for a person — looks impressive. But ask a robot to perform seemingly simple tasks, like climbing stairs or fetching a glass of water, and many of today’s robots still struggle.

See also: Chinese delivers “army of robots” that will work 24 hours a day

“What seems difficult is easy, but what seems easy is really difficult,” explained Stephanie Zhan, partner at Sequoia Capital, paraphrasing an observation by computer scientist Hans Moravec.

In the late 1980s, Moravec and other computer scientists observed that it was easier for computers to perform well on intelligence tests, but they failed on tasks that even young children could do.

Deepak Pathak, CEO of robotics startup Skild AI, explained that robots — and computers in general — are good at performing complex tasks when operating in a controlled environment.

Showing a video of a Skild robot jumping across the sidewalk, Pathak noted that “other than the ground, the robot is not interacting with anything else.”

For tasks like picking up a bottle or climbing stairs, a person uses vision to “continuously correct” what they are doing, Pathak explained. “This interaction is the root of human general intelligence, something you take for granted because virtually every human has it.”

Zhan explained that viral videos of humanoid robots do not show how the product was trained or whether it can operate in an uncontrolled environment. “The challenge for you as a consumer of all these videos is to really discern what is real and what is not,” she said.

The next step for robots

Still, both speakers expressed optimism that advances in general intelligence will soon lead to more advanced and flexible robots.

“Before, robots were much more guided by human intelligence. Someone extremely intelligent would analyze (a task) and … pre-program the robot mathematically to perform it,” Pathak said.

Now, however, the field of robotics is moving from “programming something” to “learning from experience,” he explained. This enables new robots capable of handling more complex tasks in less controlled environments and which can be easily adapted to other roles without the cost of reprogramming and reconfiguration.

Today, robotics companies are still “limited to having robots built only for specific things,” Zhan argued. A robotic platform with more general intelligence could open up “possibilities that we otherwise wouldn’t be able to achieve,” including tasks that are now dangerous for human workers.

Consumers can also benefit. “You see all these home robots, but they can only do one thing,” Zhan said. “But if we can build robots with general intelligence, you will finally have consumer robots capable of taking over the entire range of household tasks that fall to you today.”

A similar point was made previously on Brainstorm AI by Arm CEO Rene Haas, who stated that the general adaptability of humanoid robots will make them much better suited to factory work than the robotic arms currently used.

There are social repercussions in a robotics boom, with the displacement of jobs that, until now, still needed to be done by humans. Still, Pathak was calm about the social benefits of spreading automation.

One of them is safety, as robots eliminate the need for humans to perform work that is dangerous or harmful to health in the long term.

Another benefit is filling the huge labor shortage in operational and industrial jobs. (This shortage has been a barrier to U.S. efforts to bring back some of its advanced manufacturing, now concentrated in Asian economies.)

Pathak also envisioned a future in which robots free humans from the drudgery of everyday work, although he admitted that societies need to figure out how to distribute the gains from automation.

“There is a scenario, a good scenario, where everyone is doing things they enjoy,” Pathak said. “Work is more optional, and people are doing things that give them pleasure.”