AILSA CHANG, HOST:
Artificial intelligence can find you a recipe or generate a picture, but it cannot hang that picture on a wall or cook you dinner, at least, not yet. Some researchers are now trying to get AI to work in the real world by putting it into robots. NPR's Geoff Brumfiel has more.
GEOFF BRUMFIEL, BYLINE: In a laboratory at Stanford University, graduate student Moo Jin Kim is working on a new kind of robot powered by AI, similar to the AI used in chatbots.
MOO JIN KIM: It's one step in the direction of, like, ChatGPT for robotics, but still a lot of work to do.
BRUMFIEL: OK. All right. Well, you want to show me how it - show me what it can do?
KIM: Yeah, for sure. So...
BRUMFIEL: The robot itself is pretty unremarkable - just a pair of mechanical arms with pincers. What makes it different is what's on the inside. A regular robot must be carefully programmed. An engineer has to write it detailed instructions for every task. But this robot is powered by a teachable AI neural network. Kim has trained it how to do a bunch of different tasks simply by showing it.
KIM: So basically, like, whatever task you want to do, you just keep doing it over and over, maybe like, 50 times or 100 times.
BRUMFIEL: The robot's AI neural network learns, and then it does the task by itself. Kim brings out a tray of different kinds of trail mix, and I type in what I want it to do.
OK, so scoop some green ones with the nuts into the bowl.
KIM: All right.
BRUMFIEL: See what happens.
This idea that you could just tell a robot to do something on the fly is the dream of Chelsea Finn. She leads the laboratory at Stanford.
CHELSEA FINN: So in the long term, we want to develop software that would allow the robots to operate intelligently in any situation.
BRUMFIEL: By intelligently, she means the robot could understand a simple command, like scoop some green ones into a bowl, and execute in the real world.
FINN: Even just to do very basic things, like being able to make a sandwich or being able to clean a kitchen or being able to restock grocery store shelves.
BRUMFIEL: Simple, tedious jobs many humans would love to get a hand with from a machine. Finn is also cofounder of a startup called Physical Intelligence. It recently demonstrated a mobile robot that could take laundry out of a dryer and fold it. Again, the robot was taught by humans training its powerful AI program.
FINN: In that case, we actually had a workstation that was in the apartment that was computing the actions and then sending it over the network to the robot.
BRUMFIEL: So we don't have to worry about the robots taking over outside of Wi-Fi range then.
FINN: (Laughter) I think that, like - I don't think we have to worry about them taking over at all. We can just throw some water on them, and in most cases, they don't work anyway so...
BRUMFIEL: These AI robots still get confused, misunderstand, make mistakes and just get stuck.
KEN GOLDBERG: Robots are not going to suddenly become the science fiction dream overnight.
BRUMFIEL: Ken Goldberg is a professor at the University of California at Berkeley who studies robotics. It's true that AI text-writing has improved massively over the past couple of years, but that's because chatbots have a huge amount of data to learn from. They've basically taken the entire internet to train themselves how to write sentences and draw pictures. But, says Goldberg...
GOLDBERG: For robotics, there's nothing. We don't have anything to start with, right? There's no examples online of robot commands being generated in response to robot inputs.
BRUMFIEL: And if robots need as much data as their virtual chatbot friends, then having humans teach them one task at a time is going to take a while.
GOLDBERG: You know, at this current rate, we're going to take 100,000 years to get that much data.
BRUMFIEL: One solution might be to let the AI learn inside a computer simulation. Researchers in Switzerland recently trained a drone how to race by putting its AI-powered brain into a simulator and running it through a preset course over and over again.
(SOUNDBITE OF DRONE FLYING)
BRUMFIEL: When it got into the real world, it was able to regularly fly an indoor course faster and better than a skilled human opponent.
(SOUNDBITE OF DRONE FLYING)
BRUMFIEL: But anything that wasn't simulated - wind, rain, sunshine - could make the drone crash, and flying around is fairly simple. Trying to interact with stuff, like folding laundry, that gets really hard to simulate. So real-world training is too slow. Simulation leaves out some important details. And some researchers think there are even deeper problems of trying to put AI into robots. One of them is Matthew Johnson-Roberson at Carnegie Mellon University in Pittsburgh.
MATTHEW JOHNSON-ROBERSON: In my mind, the question is not, do we have enough data? It is more, what is the framing of the problem?
BRUMFIEL: Getting back to AI chatbots for a minute, Johnson-Roberson says, for all their incredible skills, the task we're asking them to do is relatively simple - look at what a human types and then try to predict the next words that user wants to see. Robots will have to do so much more than compose the sentence.
JOHNSON-ROBERSON: Next best word prediction works really well, and it's a very simple problem 'cause you're just predicting the next word. And it is not clear right now - I can take 20 hours of GoPro footage and then produce anything sensible with respect to how a robot moves around in the world.
BRUMFIEL: In other words, the tasks we want our sci-fi robots to do are so complicated compared to sentence-writing, no amount of data may be enough unless researchers can find better ways to teach the robots.
Back at Chelsea Finn's lab, graduate student Moo Jin Kim and I watch as the robot tries to scoop the trail mix I asked for. On a video feed, it places an X on the right bin, to Kim's relief.
KIM: Usually, that spot right there, where it identifies the object and goes to it, that's the part where we hold our breath (laughter).
BRUMFIEL: Then very slowly and hesitantly, it reaches with its claw and picks up the scoop.
(SOUNDBITE OF TRAIL MIX DROPPING)
BRUMFIEL: Moo Jin, did I just program a robot?
KIM: (Laughter) You did. Looks like it's working.
BRUMFIEL: It's a very small scoop, but a scoop in the right direction.
Geoff Brumfiel, NPR News.
(SOUNDBITE OF MUSIC) 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.