Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
It's funny. Laugh. AI Robotics

Scientists Try To Teach Robot To Laugh At the Right Time (theguardian.com) 34

Posted by BeauHD from the sense-of-humor dept.
Laughter comes in many forms, from a polite chuckle to a contagious howl of mirth. Scientists are now developing an AI system that aims to recreate these nuances of humor by laughing in the right way at the right time. The Guardian reports: The team behind the laughing robot, which is called Erica, say that the system could improve natural conversations between people and AI systems. "We think 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 gathered training data from more than 80 speed-dating dialogues between male university students and the robot, who was initially teleoperated by four female amateur actors. The dialogue data was annotated for solo laughs, social laughs (where humor isn't involved, such as in polite or embarrassed laughter) and laughter of mirth. This data was then used to train a machine learning system to decide whether to laugh, and to choose the appropriate type. It might feel socially awkward to mimic a small chuckle, but empathetic to join in with a hearty laugh. Based on the audio files, the algorithm learned the basic characteristics of social laughs, which tend to be more subdued, and mirthful laughs, with the aim of mirroring these in appropriate situations.

It might feel socially awkward to mimic a small chuckle, but empathetic to join in with a hearty laugh. Based on the audio files, the algorithm learned the basic characteristics of social laughs, which tend to be more subdued, and mirthful laughs, with the aim of mirroring these in appropriate situations. "Our biggest challenge in this work was identifying the actual cases of shared laughter, which isn't easy because as you know, most laughter is actually not shared at all," said Inoue. "We had to carefully categorize exactly which laughs we could use for our analysis and not just assume that any laugh can be responded to." [...] The team said laughter could help create robots with their own distinct character. "We think that they can show this through their conversational behaviours, such as laughing, eye gaze, gestures and speaking style," said Inoue, although he added that it could take more than 20 years before it would be possible to have a "casual chat with a robot like we would with a friend." "One of the things I'd keep in mind is that a robot or algorithm will never be able to understand you," points out Prof Sandra Wachter of the Oxford Internet Institute at the University of Oxford. "It doesn't know you, it doesn't understand you and doesn't understand the meaning of laughter."

"They're not sentient, but they might get very good at making you believe they understand what's going on."
This discussion has been archived. No new comments can be posted.

Scientists Try To Teach Robot To Laugh At the Right Time More

Scientists Try To Teach Robot To Laugh At the Right Time

Comments Filter:
  • Isn't this the episode with Joe Piscopo?
  • "Identify all the people who are laughing at the right time"

  • Now there's a useful invention.

  • why this project is bogus (Score:5, Interesting)

    by Walt Dismal ( 534799 ) on Friday September 16, 2022 @09:29PM (#62888667)
    I'm a researcher in this field. Let me tell you why this project is off the mark and bogus. Let's say I tell a joke about dead babies and unloading them with a pitchfork, or a 'joke' about people dying on 9/11. At the end, this statistics-trained AI will laugh. It does not understand humor, it's just a stimulus-response robot. It has NO cognitive understanding of meaning. Such understanding of humor is quite difficult to achieve in a general AI system. Humor perception requires among other things the ability to understand situations within a cultural context. Such context is one of the things determining whether something is funny. For example, if I gave you a joke containing the string 'DEADBEEF' in the right context, like a bovine programmer debugging code with that string, some of you might see humor in it. I hate it when AI researchers make claims that cheapen the field and raise false expectations. Even Google has made deceptive claims when it announced it had an AI that does humor. Turned out to be a pun-generator that simply generated sentences with words with dual meanings in a context. That alone is not enough to do real humor.

    • Re: (Score:2)

      by swell ( 195815 )

      It's even more complicated than that. Even if the robot hears a joke and responds, how does it know which response is appropriate? There are many ways to laugh, as TFS tells us. Even humans, especially geeks, have difficulty knowing the correct laugh for a particular joke in a particular setting.

      And now, even lol has lost its meaning, almost never representing real laughter.

      • And now, even lol has lost its meaning, almost never representing real laughter.

        It sure lost something in the last decade or so, lol.

    • I find your point interesting, and it makes me wonder if that is just a labeling problem. Imagine you have a 9/11 joke labeled as "bad" in the USA but "good" in a terrorist camp in Afghanistan. With such a label, couldn't the AI, with an additional input of audience or speaker know which reaction to make. Now I see the problem with this, it isn't generalizable, it requires heavy labelling, in the same way Google has an if statement for maps involving Taiwan, based upon whose asking. That sort of labelin

  • Now we can get fake laugh tracks.

  • The machines will have the last laugh.
  • You lay on the ground, chest crushed by a giant metal object shot out of a makeshift air canon. You were there to check on alert about unusual AI behavior. Into your view swims a pair of bipedal robots. One begins laughing "Hahaha, good one. Right in the center of mass." the world fades to black.

  • It might get better at laughing than people, and when that happens, we might not need human laughers any more.

  • When you hear a Republican talk about protecting children, with their party full of child molesters...that's when you have to laugh. It might not be a happy laugh, but nevertheless, you have to admire their audacity.

    If you have access to party affiliation, this algorithm will not be difficult to write.

  • They just need to fork Mark Zuckerberg.

  • the article was amazing and I like it.

  • "They're not sentient, but they might get very good at making you believe they understand what's going on."

    One might say the same thing about babies.

  • Have the robot do what everybody else does - wait for the laugh track to tell you what's funny.

  • First, they need to teach Kamala to laugh at the right time. No, on second thought, that's a much harder problem. Inadequate hardware.

  • What's funny is supposed to be the thing that is the hardest to translate between one culture and another. But, if the AI learns which collections of words to laugh at in a given situation, it should also be able to say something funny in that situation.

Slashdot Top Deals

After all is said and done, a hell of a lot more is said than done.

Close