Can a robot be taught to laugh at the right time? Scientists think so
Humanoid robot Erica learns to laugh.
Imagine: you say a funny sentence to a robot, and it bursts out laughing. This scenario could become reality thanks to a research team from Kyoto University. More specifically, these scientists are trying to teach an artificial intelligence (AI) to laugh in the right way and at the right time.
If laughter may seem trivial, it is nevertheless a very complex action. You can giggle politely, howl in hilarity or even laugh sarcastically.
All these nuances of humor, if taught to an artificial intelligence system, would allow more natural conversations between humans and robots.
According to Dr. Koji Inoue, lead author of the study, one of the important functions of conversational AI is empathy. This is why the team bet on laughter, which, once reciprocated, is one of the ways for a robot to show empathy [to its interlocutors].
Scientists work with Erica, a laughing robot equipped with an AI system. This collected data from some 80 speed-dating dialogues between university students and the machine, operated remotely by amateur actresses.
Each laugh was analyzed in order to be labeled as lonely, social (polite, when humor is not involved), embarrassed or hilarious.
Our biggest challenge in this The job has been to identify real cases of common laughter, which is not easy because […] most laughter is not shared at all, Mr. Inoue said in the newspaper. scientist Frontiers in Robotics and AI.
We had to carefully categorize the laughs we could use for our analysis and not just assume that any laugh could be answered, he added.
The audio files and their associated data (including laugh tags, in particular) were then transmitted to a machine learning system.
To test Erica, scientists gathered the algorithm's information to feed into four two- to three-minute dialogues – routed to chat software – that the robot could have with human beings. The research team compared them to times when Erica was not laughing, or only giving a polite laugh at each detected chuckle.
Excerpts from these conversations were presented to more than a hundred volunteers, who rated the performance of the laughter algorithm. Result: A more favorable score was given to the AI for empathy, naturalness of conversation, understanding and human-likeness.
According to the research team, the robots could have their own character and show it through their conversational behaviors, with laughter, gaze, gestures and speaking style.
The Dr. Koji Inoue moderates expectations, however: it will only be in more than 20 years that it will be possible to have an informal conversation with a robot, as if it were a friend.
With information from The  ;Guardian, and The Independent