The team behind the smiling robot, which is called Erica, says the system could improve natural conversations between humans and AI systems. “We believe that one of the important functions of conversational AI is empathy,” said Dr. Koji Inoue, of Kyoto University, the lead author of the research, published in Frontiers in Robotics and AI. “So we decided that one way a robot can empathize with users is to share their laughter.” Inoue and his colleagues have set out to teach their AI system the art of conversational laughter. They collected training data from more than 80 speed-dating dialogues between male students and the robot, which was initially remote-controlled by four amateur actors. Dialogue data were annotated for solo laughs, social laughs (where no humor is involved, such as polite or awkward laughter), and gleeful laughter. This data was then used to train a machine learning system to decide whether to laugh and choose the appropriate type. It may feel socially awkward to mime a small laugh, but empathetic to join in with a hearty laugh. Based on the audio files, the algorithm learned the key characteristics of social laughter, which tends to be more subdued, and happy laughter, aiming to mirror them in appropriate situations. “The biggest challenge in this work was to identify the actual instances of shared laughter, which is not easy because, as you know, most laughter is not actually shared at all,” Inoue said. “We had to carefully categorize which laughs we could use for our analysis and not just assume that any laugh could be responded to.” The team tested Erica’s “sense of humor” by creating four short dialogues to share with one person, embedding the new shared laugh algorithm into existing chat software. These were compared to scenarios where Erica did not laugh at all or gave a social laugh whenever she detected laughter. The clips were played to 130 volunteers, who rated the shared laughter algorithm favorably in terms of empathy, naturalness, human-likeness and understanding. The team said laughter could help create robots with their own distinct character. “We think they can show this through their conversational behavior, such as laughter, eye contact, gestures and speaking style,” Inoue said, though he added that it may take more than 20 years to become possible. a “casual chat with a bot like we would with a friend”. Professor Sandra Wachter, of the Oxford Internet Institute at the University of Oxford, said: “One of the things I will keep in mind is that a robot or an algorithm will never be able to understand you. He doesn’t know you, he doesn’t understand you and he doesn’t understand the meaning of laughter. “They’re not sensitive, but they can be very good at making you think they understand what’s going on,” he added.
