SCOTT DETROW, HOST:
Imagine having a robot around your house to do things like laundry or make your bed. Some scientists say they have made a key breakthrough that would allow robots to figure out these tasks on their own. But this raises big questions about how much risk comes with letting robots be in charge of their own learning. NPR's Katia Riddle has more.
KATIA RIDDLE, BYLINE: For years, robotic scientist Sthithpragya Gupta has been dreaming up the perfect robot.
STHITHPRAGYA GUPTA: I personally want the robot to make me a coffee.
RIDDLE: With his engineering and science colleagues, says Gupta, people stay at the lab all hours. They drink a lot of coffee. He'd like to be able to give his robot specific directions.
GUPTA: Little bit of sugar, a bit more creamer, stuff like that, then oh, that would be a dream come true.
RIDDLE: It's not just his dream. There's a slew of startups full of engineers trying to develop robots that could do things like unload a dishwasher or sort packages at a warehouse. Humans have been able to teach robots to do all kinds of specific tasks, but it's hard to teach them how to learn things. Gupta likes to explain this difference with a tennis metaphor.
GUPTA: Imagine that, you know, like, there's a tennis academy and there's the instructor with some people who are there to learn tennis.
RIDDLE: The instructor demonstrates how to hit a backhand. Most people can learn it eventually. Robots could also potentially learn to hit a backhand, but then something changes. The opponent moves. The light shifts. A person can adjust their backhand to accommodate these changes. The robots, not so much - they can't adapt like humans.
GUPTA: It's very difficult to transfer this behavior from human to all the robots.
RIDDLE: That is until now, says Gupta. He and his colleagues have just published a paper in the journal Science Robotics about a new way of teaching robots. They've used machine learning, which is a kind of AI, to help robots adjust their movements based on their own physical abilities and limitations. They demonstrate their work in this video. It shows robots with a single arm attached to a base.
(SOUNDBITE OF ARCHIVED RECORDING)
UNIDENTIFIED NARRATOR: The way forward is robots learning from humans and transferring those skills across different designs.
RIDDLE: In the video, an instructor picks up a ball and throws it into a small container. Robots watch, then they pick up the ball and throw the ball into the container as well, adjusting for their own position and very nonhuman bodies.
(SOUNDBITE OF ARCHIVED RECORDING)
UNIDENTIFIED NARRATOR: This opens the door to flexible, easily upgradable robot fleets.
RIDDLE: Gupta says we're likely years away from our own personal baristas, but this development introduces some thorny questions like, if this robot can adjust and improve upon its own behavior, does that make it self-aware or even conscious?
SUSAN SCHNEIDER: It looks like this robot is capable of doing some very impressive feats of learning, but that doesn't mean something has full-blown consciousness or inner awareness or is a self in the sense that biological beings have it.
RIDDLE: Susan Schneider studies AI at Florida Atlantic University. She points out that this robot doesn't feel things. That's a critical distinction between humans and robots.
SCHNEIDER: Consciousness is the felt quality of experience. So when you sip your morning espresso shot, when you see the rich hues of the sunset, when you have a headache, it feels like something from the inside to be you.
RIDDLE: The robot is feeling none of these things, even if it knows how to teach itself to improve. But this lack of feeling introduces a new issue. If there's no consciousness, is there morality? What's to keep someone from teaching the robot how to hurt someone?
SCHNEIDER: It immediately raises alarm bells in, you know, any AI safety researcher's mind.
RIDDLE: The researchers on this project are including safety protocols to ensure that robots can't hurt people. But even they acknowledge it's a risk. Susan Schneider says this moment is an inflection point in how robots will evolve with AI.
SCHNEIDER: It's a very exciting time, and we just don't know where it's headed.
RIDDLE: The robots that emerge in the next five to 10 years, says Susan Schneider, could be life-changing for good or not. Katia Riddle, NPR News. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.