Oregon State University Develops Talk Show Robot, Which Has Performed Dozens of Shows Across the United States and Received Positive Reviews

Scholars at Oregon State University have been studying how to make robots master humor and learn to perform better in talk shows. To this end, the team took the talk show robot Jon to perform dozens of shows and summed up some tips for success.

In our traditional impression, most robots have a cold metallic texture. Even voice assistants like Siri, although they have no physical image, their cold voices convey a robotic temperament.

There is a saying,“When human-machine interaction becomes as natural as human communication, the era of true intelligence will arrive.”

Scientists and engineers have been working hard in the field of human-computer interaction, and hope that robots and AI can acquire a sense of humor so that they can get along better with humans. After all, a sense of humor is one of the most valuable qualities of humans.

For example, Siri and Microsoft Xiaobing are both working hard to learn how to tell good jokes.

Let robots tell jokes: a good way to improve human-computer interaction

It is not easy to give robots or voice assistant AI a sense of humor, and it is even considered the ultimate challenge in the field.

(We have discussed the issue of AI having a sense of humor before, see"Let artificial intelligence learn humor, so that human-computer dialogue is not awkward". )

In May 2019, a robot named "Titan" appeared on the talk show "Swordsman". Claiming to be a comedian, it made jokes frequently and made the audience laugh.

However, the overly realistic performance of "Titan" in action and language interaction aroused doubts from the audience. It was later confirmed that it was indeed just a robot shell with a real person inside to control its actions and performances.

Since the development of social robots, their interactivity and understanding of human emotions have always been very challenging.A robot like "Titan" may have some elements of meeting the needs of the program, but it is still difficult to create a robot that can be controlled freely and is good at picking up jokes in reality.

A research team at Oregon State University has been working on human-computer interaction and robot sense of humor since 2011. In order to solve this problem and make social robots more attractive to people and more human-like with their sense of humor, Naomi Fitter, assistant professor of robotics, plans to start by building a talk show robot.

The team first selected the NAO robot developed by SoftBank Robotics as a carrier and performer to run the talk show program they developed.

So far, Naomi Fitter has toured the West Coast with her talk show robot Jon for 32 shows.Discovered how robots can gradually understand and learn human sense of humor.

32 performances, continuously improving performance skills

We know that talk shows generally consist of "opening remarks - jokes", and in addition to the jokes,The speed at which jokes are told, as well as the pause time between the opening remarks and the punchline, are also crucial to a talk show performance.

Since the current research direction is to make the humorous interaction between robots and humans more natural rather than automatically generating jokes, the researchers followed the general routine of talk shows and pre-wrote performance materials suitable for Jon.

The material is all from the robot's perspective, and themes include interpersonal relationships, daily life, politics, dreams and frustrations.

For example: Recently when my nephew asked me "where do robots come from", I was always embarrassed. I could only tell him that robots come from their mother - the delivery truck.

When performing, Jon uses the radio to judge the audience's reaction to decide whether the next joke needs a pause or needs to be emphasized.

The researchers wrote about 8 minutes of material, including 26 jokes and 22 emotional tags, and then used the NAO robot as a performer to run its comedy performance program.

Jon the Robot performed a total of 32 shows at two open mic venues, one in the Los Angeles area and one in Oregon.

  First study: Pausing skills improve performance 

The first study was conducted in the Los Angeles area and included 22 performances with an audience of 10-20 adults who had never seen a robot talk show before.

The robots performed in two modes.

The first mode is bad-timing mode: That is, after telling each joke, wait for a fixed 5 seconds before starting the next one.In this mode, the robot completed 12 jokes;

The team used two methods based on volume and sound count to determine whether the audience was still laughing and clapping.

Then the robot was given a baseline waiting time of 170ms. If the audience remained silent during this period, it would start. If there was no response to silence, it would wait until the 5-second timeout to start the next joke (experts suggest that 5 seconds is usually enough.)

The second mode is the appropriate timing mode:The robot decides to pause or continue speaking based on the audience's reaction. If the audience is laughing, it pauses; if the audience becomes quiet again, it continues.In this mode, Jon completed 10 jokes.

The study shows that if a robot comedian has good timing in his performance, that is, giving the audience appropriate reaction time, etc., it will be much more interesting than a robot that has no timing and performs on its own.

Because, when a joke is successful, the audience will often laugh or applaud, but because the robot will start the next joke after a fixed delay, the audience will miss part of the next joke, resulting in a failure.

  Study 2: Observing the Expression of Others While Speaking 

The second study was based on 10 performances in Oregon.The team took into account the different responses of the audience to the performance, which they called "adaptive performance."

The research team added jokes with tags to the joke material. The tags are divided intoIntense, Lukewarm and SilentThree types.

Each time the robot told a joke, it chose which "label" of joke it wanted to tell based on the level of laughter it received from the audience. In the second study, all performances had appropriate timing.

However, the results of this study showed thatWhether or not there is a label has little impact on the success rate of the performance.

  Conclusion: Timely release of the punchline will increase the probability of success 

Based on the above research, Fitter concluded: "In bad-timing mode, the robot always waited a full 5 seconds after each joke, regardless of the audience's reaction.

In appropriate-timing mode, the robot used a timing strategy to pause when laughter started and continue after it died down, just like a good human comedian.

So in general,When jokes are delivered at the right time, audiences respond more.Jon is already quite familiar with this technique.

Research significance: Improving robots' understanding of human humor

Fitter said the 32 performances provided the research team with enough data to identify significant differences between the different modes of robot talk shows.This data could help "autonomous social robots improve their sense of humor."

At the same time, this research also helps provide some answers to key questions about comedic social interactions. It will help researchers in the fields of robotics and AI,Understanding how human groups respond to entertaining social robots in real-world settings will also help comedians evaluate the success of a joke or routine.

For example, before the next performance, Li Dan and Chi Zi can let the robot rehearse for them first, and if the effect is not good, they can make improvements in time. So, in the future, talk show hosts may not need to perform on stage, but just focus on writing jokes behind the scenes?

Just like Jon's joke: If you like me, book me! Not only can I tell jokes, I can also replace your job ?

-- over--