Preliminary results on the introduction of humanlike avatars to online learning

Humanlike avatars are starting to act as tutors in online learning environments. We conducted a series of studies, in collaboration with IDC Herzliya, to analyze various aspects of humanlike tutors and the expansion of immersive learning. Our preliminary findings indicate that learners’ response to humanlike avatars varies significantly according to personality type. We show several ways by which humanlike avatars induce high emotional engagement and learner satisfaction.

In the past year we were involved in a truly exciting and pioneering research that analyzed the introduction of humanlike avatars to online learning. We have some preliminary results that we’re happy to share, while we prepare more detailed papers.

We believe interactions with humanlike avatars would soon become an important part of the learning experience in universities, schools and the workplace. Therefore, the following is possibly a first look on where online learning and training is headed to.

But let’s start from the beginning.

About a year ago we collaborated with IDC Herzliya’s (Israel) Data-Tech track, headed by Dr. Gail Gilboa Freedman, and aided by Mr. Or Peretz. 25 students ran 5 sets of A-B tests on 575 online learners, using Trenario’s platform, under the supervision of Moti Shatner, co-founder of Trenario.

The study spanned across several aspects of the experience of learning with humanlike avatars.

So what did we find? Here are our preliminary findings:

Finding 1: The Uncanny-Valley Effect still exists, but sometimes it is reversed.

Humanlike avatars that imperfectly resembles actual humans tend to provoke a feeling of eeriness, even revulsion, in their (human) observers. That’s according to the Uncanny-Valley hypothesis that was first introduced by Prof. Masahiro Mori in 1970.

With the introduction of humanlike avatars to online classrooms, is this hypothesis still standing? And if it does, how can we decrease its effect?

In order to explore these questions we created two almost identical online courses — one with humanlike avatars, the other with more cartoon-style, less humanlike avatars.

We found that a feeling of eeriness still appears, but it changes according to the learner’s personality type. For example, introverts prefer a cartoonish avatar over a humanlike one. Humanlike avatars make them feel a little uncomfortable (a classic Uncanny-Valley effect). Nevertheless, other personality types responded differently. Specifically, learners who are emotionally unstable preferred humanlike avatars, and felt eerie with cartoon-like avatars.

Insight: in the future, online learning platform will offer the right “humanlike level” of avatars, according to the learner’s personality type.

Finding 2: If we allow the learners to choose their humanlike tutors, they become much more engaged and satisfied.

Today, learners cannot change and adjust their tutors according to their likings. But what if they could do that?

In this experiment, a humanlike tutor opened the course by asking the learner to choose between two avatars. The tutor gave some details about the avatars and showed their pictures. After the learners made their choice, the course continued with the selected avatar. Learners in a control group got a course in which they could not choose their avatars.

As expected, learners who were offered to choose their tutors showed much higher levels of satisfaction and engagement. Nevertheless, the ability to choose the avatar did not affect the learner’s feedback on the actual material being taught. That is, the overall learning experience was much more satisfying, yet it did not affect the learner’s sense of criticism when we asked for feedbacks on specific features of the course narrative.

Insight: In the future, when online learning platforms will allow learners to change and adjust avatars to their likings, engagement and satisfaction levels will be much higher.

Finding 3: When learners are passive during online sessions it reflects badly on their attitude towards their humanlike tutor.

It is not easy to conduct an online learning session in which all learners are active. This is even more challenging, as most of online learning is based on video clips (MOOCs) and Zoom (or like software) sessions. With humanlike avatars conducting the session, we can better control the activity level of the learners. How will it affect the learners?

To test that, we created two almost identical courses. In one of them, the learners had to constantly respond to the avatars. In the other, the learners were much more passive, i.e. were hardly asked to respond. For both courses, the underlying material was the same.

We found several differences in learner’s reactions to the two courses. Most interesting was the learners’ perception of their tutors. In more passive settings the tutor was perceived as more arrogant, even annoying. This phenomenon was even more observant among learners that were profiled as extroverts. For them being passive and not interacting with the tutor was highly frustrating.

Insight: A passive learner reflects badly on the tutor. In future blended learning environments, avatars could help tutors in creating highly interactive sessions, thereby improving the learners’ perception of the tutors.

Finding 4: When humanlike tutors talk about their virtual achievements, learners show higher emotional engagement.

It is already common knowledge that emotional engagement is key to the success of the learning session — it increases satisfaction levels and boosts long term impact.

But how can a tutor create emotional engagement? We ran two almost identical online courses in which a humanlike tutor talks to the learners and responds to them. The difference was in the way the tutor opened the session. In course A (the control group), she just said hello and went on to the learning part. In course B the tutor talked about herself in length, even bragged about her (fictional) academic degrees, personal achievements and a book that she wrote. Only then did she continue to the learning part.

Remember the learner knew that this is a virtual avatar. Still, this self-introduction made a significant effect. We found that an avatar that talks about herself, even if it’s clear that all is fictional, creates a significant positive emotional response by the learners. Moreover, even if the avatar brags about her achievements, this does not create alienation (on average), but rather increases the learners’ emotional engagement.

Insight: Online tutors should not be too focused on the material at hand. They can create significant emotional engagement by talking about themselves.

Finding 5: In an immersive learning environment, humanlike avatars that induce empathy highly increase overall learner satisfaction.

Immersive learning, in our case learning through interactions with avatars, often introduces the learners to a “story” or a “scene” in which avatars play various roles. How should these avatars “act” in order to increase user satisfaction?

We created two almost identical courses to better study this question. In the base course, a virtual manager was talking to the learners, asking for their advice on a management crisis that occurred in his company. In course B, the manager was interacting in the same manner, but some of the words and phrases were replaced by highly emotional words, i.e. we increased the emotional level of the story.

We found that learners showed a much higher level of satisfaction in course B, in response to the highly emotional avatar. Learners connected with the avatar, and felt they had to “help him”. As expected, this emotional connection also resulted in better knowledge retention.

Insight: Immersive learning sessions should be emotionally charged to increase user satisfaction and knowledge retention.

Summary

The introduction of avatars to assist in online learning creates a great opportunity for tutors to be able to design and conduct much more personalized and targeted sessions. This approach can also help in addressing the online learning challenges raised by the COVID-19 pandemic. We hope this research will help advance more studies on immersive, interactive online sessions with humanlike avatars.

About the research

For this study we created a “base” course (the “A” course), that trained the user on how to best cope with a management crisis. The course was interactive and immersive. A humanlike tutor talked to the learner, posed questions and responded to the learner’s reactions. Another humanlike avatar, a “manager” consulted with the learner.

We then created 5 other courses (the “B” courses) that were each slightly different from the A course, and tested various aspects of humanlike online tutoring. After they finished doing the course, all learners were asked to answer a detailed questionnaire that also included a standard personality test.