The research paper on which this feature article is based will be presented on May 12, 2021 at CHI 2021, where it received a best paper honourable mention.
CHI is the premier international conference on human–computer interaction. Held from May 8 to 13, 2021, the 2021 ACM CHI Virtual Conference on Human Factors in Computing Systems was originally slated to be held in Yokohama, Japan.
We often hear that laughter is the best medicine, but it may also be a powerful teacher. Humour lowers stress and anxiety, enhances self-esteem, and boosts motivation, fostering an enriching emotional and social environment in which students are more engaged and eager to learn.
Teachers who use humour in the classroom often see improved retention and recall in their students. But what if a teachable virtual agent — an agent with artificial intelligence and the ability to converse naturally — used humour when interacting with students? Teachable agents are being used ever more commonly in computer-mediated education because learning-by-teaching is often a more enriching experience for students than is learning on their own.
“We wanted to explore the use of humour by virtual agents to see whether a humorous conversational agent enhances learning experiences and outcomes,” said Jessy Ceha, a PhD candidate at the Cheriton School of Computer Science, who led the study. “To our knowledge, this is the first systematic study that looks at the use of humour by a teachable virtual agent.”
Humour is a part of an individual’s personality and the role it plays in the way a person engages with others has generated considerable interest. One of the most prominent contributions in the field of personality research has identified four styles of humour — affiliative, the use of benign humour to enhance social relationships; self-defeating, the use of humour to enhance relationships at one’s own expense; self-enhancing, the use of humour to enhance oneself; and aggressive, the use of humour to enhance oneself at the expense of others.
“We chose to explore the effects of affiliative and self-defeating humour in a teachable agent,” Jessy said. “These two styles are known to be conducive to social well-being and in building interpersonal relationships. We also looked at a neutral style, one where the virtual agent expressed no verbal humour.”
The study explored the following research questions — does a virtual agent’s use of affiliative or self-defeating humour affect the perception of the agent by participants, their attitudes toward the task, and their ability to recall material, and does the humour style of participants interact with that of the virtual agent?
The experiment took place on the Curiosity Notebook, an online learning-by-teaching platform developed to allow students to instruct a virtual agent.
“Using the Curiosity Notebook, participants teach a virtual agent called Sigma, which was represented by a static avatar, how to classify sedimentary, igneous and metamorphic rocks,” said Ken Jen Lee, a PhD candidate at the Cheriton School of Computer Science, who collaborated on the study.
Curiosity Notebook’s interface has a panel on the left side with a number of articles and pictures about the topic to be taught. Sentences within the articles can be selected and used to teach the virtual agent.
“The platform supports learning-by-teaching, the idea that students themselves do not know the topic, but by teaching the virtual agent about it they learn more about the topic themselves,” explained Edith Law, a Professor at the Cheriton School of Computer Science who supervises Jessy and Ken Jen. “The Curiosity Notebook is quite flexible in that we can make manipulations in the agent’s behaviour to study how it changes the learning-by-teaching process.”
Curiosity
Notebook’s
teaching
interface.
Articles
with
text
and
photos
are
on
the
left.
A
chat
window
is
on
the
right
and
it
is
here
that
a
student
and
Sigma
interact.
The
interaction
in
this
example
shows
Sigma’s
use
of
affiliative
and
self-defeating humour
as
well
as
the
neutral
condition (view
a
larger
static
image).
During
the
experiment,
however,
Sigma
was
in
a
single
condition
(affiliative,
self-defeating
or
neutral)
for
the
duration
of
the
interaction
with
the
participant.
Participants converse with the agent through a chat window on the right side of the screen, Ken Jen said. “At first, the notebook is empty because Sigma doesn’t know anything about rocks, but as the participant teaches Sigma, facts begin to fill up the notebook. For example, if I type shale, the notebook begins to be filled with information about shale and now Sigma knows that shale is a kind of rock.”
Sigma’s Notebook showing a list of rocks taught so far (left) and a page of notes on shale with an explanation provided by a participant (right).
When a participant is ready to interact with Sigma, he or she selects one of seven buttons at the top right of the teaching interface. Within Teach, a participant can select text to describe or explain a rock’s feature or to compare features. Within Check, a participant can correct information or quiz Sigma about rocks. And within Entertain, the fun fact button allows the agent to ask the participant to provide an interesting fact about rocks and occasionally an explanation why they thought it was interesting. The tell joke button allows participants to tell Sigma a joke.
Although participants can choose the type of interaction to have with Sigma, the virtual agent drives the conversation by asking questions and making statements.
Most often, the jokes were riddles — questions that rely on a play on words for comedic effect. For example, in the affiliative humour condition Sigma might say, “What’s a rock’s worst enemy? Paper, haha!” In the self-defeating humour style, Sigma might say, “You know that feeling when you’re taught something and understand it right away?... Yeah, not me! Haha!”
Fifty-eight participants at the University of Waterloo volunteered to take part in the study and they were randomly assigned to one of the three conditions. Regardless of the condition, participants interacted with Sigma one-on-one — a single participant with the teachable virtual agent in one of the three conditions.
“We found that participants who interacted with the agent with an affiliative humour style showed an increase in motivation during the teaching task and an increase in the amount of effort they put into teaching the agent,” Jessy said. “The participants who interacted with the agent with a self-defeating style also showed an increase in effort, but they didn’t find it as enjoyable an experience. We think the agent’s self-defeating style of humour may have decreased their self-confidence in teaching it.”
“We also looked into how the participants’ own humour style interacted with the humour style of the agent,” Jessy continued. “When participants with an affiliative humour style interacted with the agent that also had an affiliative humour style they were more likely to put in more effort for that agent. With the self-defeating agent, participants who had more of a self-defeating humour style rated their own teaching ability higher with the agent, so we also saw these interactions between participant’s characteristics and the agent’s characteristics. This indicates that care must be taken in the design of agents because a one-size-fits-all approach to humour will not be the most successful.”
The research team also found that humour accentuates the human likeness of an agent by giving it a personality. This is especially important in settings where an agent interacts with users over long periods of time, where having more of a personality can be helpful to increase users’ commitment to a task.
“We learned a lot, but there’s so much more to know about humour in virtual agents,” Jessy said. “The timing of the humour, how frequently humour can be used, and whether styles of humour should change are all things we may explore next.”
To learn more about the research on which this feature article is based, please see Jessy Ceha, Ken Jen Lee, Elizabeth Nilsen, Joslin Goh, Edith Law. Can a Humorous Conversational Agent Enhance Learning Experience and Outcomes? CHI '21: Proceedings of the 2021 CHI Conference on Human Factors in Computing Systems, May 2021, Article: 685: 1–14. https://doi.org/10.1145/3411764.3445068
This multidisciplinary research was funded by the Natural Sciences and Engineering Research Council of Canada (Discovery Grant RGPIN-2015-0454) and the University of Waterloo Interdisciplinary Trailblazer Fund, held by principal investigators Edith Law from the Cheriton School of Computer Science and Elizabeth Nilsen from the Department of Psychology.