Monday, September 19, 2022

To Bond With Humans, Robots Are Learning to Laugh at the Right Time

Leslie Katz -

Anyone who's shared a laugh with a friend knows how deeply bonding humor can be, so it stands to reason our future robot companions have a better chance of winning our trust and affection if they can laugh with us.


Teaching Erica, the robot, to laugh is serious business. Kyoto University/Hiroshi Ishiguro Lab© Provided by CNET

But just because a robot tells jokes doesn't mean it can respond to them appropriately. Did a comment warrant a polite robot giggle or an all-out bot belly laugh? The right response could mean the difference between an approachable android and a metallic boor.

That's why Japanese researchers are attempting to teach humorless robot nerds to laugh at the right time and in the right way. Turns out training an AI to laugh isn't as simple as teaching it to respond to a desperate phone tree plea to cancel a subscription. "Systems that try to emulate everyday conversation still struggle with the notion of when to laugh," reads a study published Thursday in the journal Frontiers in Robotics and AI.

Erica the humanoid robot is in the lab getting a sense of humor. 
Osaka University, ATR© Provided by CNET

The study details the team's research into developing an AI conversational system focused on shared laughter to make chatter between humans and robots more natural. They envision it being integrated into existing conversational software for robots and agents, which are already learning to detect emotions and deal with open-ended complexity like vague human commands.

"We think one of the important functions of conversational AI is empathy," Koji Inoue, an assistant professor of informatics at Japan's Kyoto University and the study's co-author, said in a statement. "Conversation is, of course, multimodal, not just responding correctly. So we decided that one way a robot can empathize with users is to share their laughter."

The key is that the system not only recognizes laughter, it also decides whether to laugh in response and then chooses the right type of laughter for the occasion. "The most significant result of this paper is that we have shown how we can combine all three of these tasks into one robot," Inoue said. "We believe that this type of combined system is necessary for proper laughing behavior, not simply just detecting a laugh and responding to it."

To gather training data on the frequency and types of shared laughter, the team tapped Erica, an advanced humanoid robot designed by Japanese scientists Hiroshi Ishiguro and Kohei Ogawa, as a platform for studying human-robot interaction. Erica can understand natural language, has a synthesized human voice and can blink and move her eyes when listening to humans go on about their people problems.

The researchers recorded dialogue between male Kyoto University students who took turns chatting face-to-face with Erica as amateur actresses in another room teleoperated the bot via microphone. The scientists chose that setup knowing there'd naturally be differences between how humans talk with each other and how they talk with robots, even those controlled by another human.

"We wanted, as much as possible, to have the laughter model trained under the similar conditions to a real human-robot interaction," Kyoto University researcher Divesh Lala, another co-author of the study, told me.


At left, a human talks to Erica the robot, who's being controlled from a separate room by an actress
. Kyoto University© Provided by CNET

Based on the interactions, the researchers created four short audio dialogues between humans and Erica, who was programmed to respond to conversations with varying levels of laughter, from none at all to frequent chuckles in response to her human conversational buddies. Volunteers then rated those interludes on empathy, naturalness, likeness to humans, and understanding. The shared laughter scenarios performed better than the ones where Erica never laughs or laughs every time she detects a human laugh without using the other two subsystems to filter context and response.

The Kyoto University researchers have already programmed their shared laughter system into robots besides Erica, though they say the humanoid howls could still be more natural sounding. Indeed, even as robots becoming increasingly lifelike, sometimes unsettlingly so, roboticists concede that infusing them with their own distinct humanlike traits poses challenges that go beyond coding.

"It may well take more than 10 to 20 years before we can finally have a casual chat with a robot like we would with a friend," Inoue said

Erica, needless to say, isn't ready for the stand-up circuit yet. But it's intriguing to think there may soon come a day when it truly feels like she gets your jokes.

No comments: