(ORDO NEWS) — Japanese scientists have developed an artificial intelligence system for the android Erika, which allows her to naturally respond to the laughter of the interlocutor.
Now Erica can not only recognize a person’s laughter, but also determine whether she should laugh back and what type of laughter to use for this.
Scientists from Kyoto University (Japan) have developed an artificial intelligence system that allowed the Japanese android Erika to appropriately respond to the laughter of the interlocutor.
At the same time, scientists did not seek to make Erica detect laughter or respond to a joke, but tried to endow artificial intelligence with a human sense of humor.
According to researchers, empathy is one of the most important functions of conversational artificial intelligence (AI).
The authors decided that one of the mechanisms of empathy could be shared laughter, and robots with a sense of humor could improve the quality of conversations with a person.
Erika is an android created by Japanese scientists who looks like a girl with natural lively facial expressions and silicone skin. Erica knows how to conduct a dialogue, identify people‘s emotions and respond to them.
In the collaborative laughter model developed by the authors, the human laughs first, and then the AI responds.
This approach required the development of three subsystems: one detecting laughter, the second deciding whether to laugh back, and the third choosing the appropriate type of laughter.
Scientists have recorded more than 80 dialogues during Kyoto University students’ speed dates with Erica, who was remotely controlled by several actresses.
That is, these were conversations between two living people.
The biggest challenge was identifying actual cases of shared laughter: the scientists had to carefully classify exactly which laughter they could use for further analysis, since not every laughter was worth responding to for Erica.
The type of laughter also mattered, as in some cases a polite chuckle might be more appropriate than a loud laugh.
The trained AI was tested by the authors in four short two- to three-minute dialogues between a human and Erica.
In the first scenario, she responded with only polite laughter, in the second and third, she burst out laughing merrily, and in the last, she used both types of laughter.
The scientists also created two other sets of the same dialogues, but in one of them Erika did not laugh at all, and in the other she laughed every time the interlocutor laughed, regardless of the context.
After that, 130 people listened to each dialogue in three scenarios and rated how empathic and human Erica was, as well as how natural and appropriate her reaction was.
The system of joint laughter proved to be the best. Thus, the robot can simultaneously perform several tasks of processing human laughter and choose a specific answer, which in the future will make conversations with robots more natural and comfortable for users.
Contact us: [email protected]