A team of scientists from Kyoto University are working on making robots laugh.
A team from the University of Kyoto is working on an ambitious project, it consists of making robots laugh when they can do it, properly and when it is appropriate to do so. To achieve this, they have taught an artificial intelligence (AI) system known as: “the art of laughter” during conversations with humans, a far from easy task for which they collected data from more than 80 talks in which students participated and the own AI, a system that Japanese scientists have baptized “Erica”.
The experience served to accumulate information about solitary laughs – that funny joke that you remember suddenly, silently – and those that arise when interacting with other people. Among the latter, they differentiated between laughter of joy and others caused by less obvious reasons, such as embarrassment, nerves or a simple matter of courtesy. With such background, the Japanese team dedicated themselves to training Erica, basically teaching her two keys: when to laugh and how to do it.
“Our biggest challenge was identifying the real cases of shared laughter. It’s not easy because, as you know, most laughs aren’t really shared at all,” Dr Koji Inoue of Kyoto University told The Guardian. The task seems anything but fun: Inoue and his colleagues had to categorize the laughs, choosing which ones were worth their study.
With Erica’s already primed sense of humor more than ready, the scientists decided to put her to the test. They designed four short dialogues, ran the new laugh-share algorithm, and then showed the results to about 130 volunteers who evaluated them on questions such as how natural they were, whether they reflected understanding or resembled a human reaction.
The key, the researchers detail, who have collected the entire process and their conclusions in Frontiers in Robotics and IA, is always the same: identify well the laughs of the interlocutor to which we can respond and select the best way to do it. “The prediction and selection showed higher scores than a random model”, explain its authors, who also detail that the level of detection was also high. His model, in short, conveyed a greater sense of empathy than others in which, quite simply, you always respond with social laughter.
The objective of this project is to make the AI act more naturally and also with a greater capacity for interaction. “We believe that one of the important functions of conversational AI is empathy,” says Dr. Koji Inoue, one of those responsible for the study, to the British newspaper, “so we decided that one way a robot can empathize with users is sharing his laugh.
“Dialogue systems must be able to express empathy to achieve natural interaction with humans. However, generating laughs requires a high level of dialogue understanding. Implementing laughter in existing systems, such as conversational robots, has been a challenge.” The Japanese team believes that laughter could be used to design robots with a peculiar character. Of course, Inoue assumes that it will be decades before we can chat with a robot as we would with a friend.