Smiling is one of the most important parts of communication and bonding for humans, and therefore, having our future robot friends laughing with us will increase their chances of gaining our trust and love. But just because a robot can joke doesn’t mean it can respond appropriately to jokes. Being able to accurately decide whether a phrasing requires a smile or a laugh in response could mean the difference between a connectable android and a programmed robot.
That’s why Japanese researchers are trying to teach humorless robots to laugh at the right time and in the right way. Training an AI to laugh is not as simple as teaching it to give a presentation on a particular topic. “Systems that try to mimic everyday speech still struggle with the idea of when to laugh,” says a study published in the journal Frontiers in Robotics and AI.
The study details the team’s research to develop an AI speech system that focuses on shared laughter to make conversations between humans and robots more natural.
“We think one of the important functions of speech AI is empathy,” says Koji Inoue, assistant professor of informatics at Kyoto University in Japan and co-author of the study. We decided that one way to empathize was to share their laughter.”
The important point here is that the system not only recognizes the laughter, but also decides whether to laugh in response, and then chooses the type of laugh that suits the situation. “The most important result of this paper is that we have demonstrated how we can combine these three tasks into a single robot,” Inoue said. “
The team used Erica, an advanced humanoid robot designed by Japanese scientists Hiroshi Ishiguro and Kohei Ogawa, as a platform to collect educational data on the frequency and types of shared laughs and to study human-robot interaction. Erica can understand natural spoken language, has a synthesized human voice, and can blink and move her eyes while listening to people’s problems.
The researchers recorded the dialogue between male Kyoto University students who took turns chatting face-to-face with Erica, while amateur actresses in another room teleoperated the bot through the microphone. The scientists state that they chose this order knowing that there will naturally be differences between how people talk to each other and how they talk to robots, even those controlled by another human.
“We wanted the laugh model to be trained under conditions as similar to a real human-robot interaction as possible,” says Kyoto University researcher Divesh Lala, another co-author of the study.
Based on these interactions, the researchers created four short, audio dialogues between humans and Erica, who are programmed to respond with varying levels of laughter in response to human conversational companions, from no laughter to frequent chuckles. Volunteers then rated these breaks for empathy, spontaneity, similarity to people, and understanding. Shared laugh scenarios outperformed scenarios where Erica always laughed or never laughed when she detected a human laugh, without using the other two subsystems to filter context and response.
Kyoto University researchers have also programmed their common laugh system into robots other than Erica, although they say humanoid voices may sound more natural. While robots are becoming more and more realistic, sometimes to a disturbing degree, robotics experts point out that infusing robots with their own distinct humanoid characteristics poses challenges that go beyond coding.
“It may take more than 10 to 20 years before we can finally have a casual conversation with a robot like we do with a friend,” Inoue says.
25703
