Learning by Teaching: Why Children Learn More When the Robot Is the Pupil

Some of the most imaginative educational robots are deliberately bad at things, because a child who has to teach a struggling robot often ends up learning more than one who is taught by a clever one.

Flipping the classroom

The obvious role for a robot in a school is the tutor: patient, tireless and endlessly willing to repeat the same explanation. But anyone who has ever had to explain something to someone else knows that teaching is a remarkably effective way to learn. You have to organise what you know, spot the gaps in your own understanding and keep track of whether your pupil is following. Educational researchers have long studied this learning by teaching effect, and social robots offer an unusually vivid way to harness it.

The protégé effect

A helpful starting point comes from work on teachable software agents. Chase, Chin, Oppezzo and Schwartz studied students working with teachable virtual agents and described what they called the protégé effect: learners put more effort into learning when they were doing it on behalf of an agent they were responsible for. Feeling responsible for someone else's progress, even a digital someone, seems to be motivating in a way that personal study often is not.

Robots add a physical body to this idea, along with gaze, gesture and the sense of a real presence in the room, which may make the feeling of responsibility stronger still.

The care-receiving robot

In a 2012 paper in the Journal of Human-Robot Interaction, Tanaka and Matsuzoe proposed turning the usual arrangement on its head. Rather than building a robot that teaches or cares for children, they introduced a care-receiving robot into a classroom at an English language school for Japanese children aged between three and six. The children's job was to teach it English verbs.

Their results suggested the idea was workable, and that the robot could help children learn new verbs efficiently. Just as interesting were the observations of how children went about teaching, gathered through classroom observation and interviews with parents. Young children, it turned out, readily adopted the role of teacher and developed their own ways of instructing the robot, insights the authors used to suggest how such robots could be designed better.

Teaching a robot to write

The CoWriter project, by Hood, Lemaignan and Dillenbourg at EPFL, applied the same principle to handwriting. A NAO humanoid robot, whose fingers are far too clumsy to hold a pen well, was paired with a tablet that displayed simulated handwriting. Using statistical models learnt from adult writing, the robot could produce deliberately deformed letters, and children then demonstrated better versions. With each demonstration the robot adjusted its own letter shapes.

In preliminary classroom studies reported in 2015, children aged six to eight engaged readily and improved the robot's writing until they were satisfied with it. The design goal was not simply practice. By casting the child as the expert, the system aimed to stimulate reflection on what makes a letter well formed, along with empathy and self-esteem, in children who might otherwise feel discouraged.

A 2016 article in IEEE Robotics and Automation Magazine by Lemaignan and colleagues set out the wider approach, opening with the stories of children who struggled with handwriting and whose difficulties had begun to affect their confidence at school. The emotional dimension is part of the point: a child who has been labelled as needing special help gets to be the one who helps.

Where it fits in the bigger picture

A 2018 review in Science Robotics by Belpaeme, Kennedy, Ramachandran, Scassellati and Tanaka surveyed social robots in education more broadly. The authors described robots acting as tutors and as peer learners, and concluded that they can improve both cognitive and affective outcomes, sometimes matching human tutoring on restricted tasks. They attributed much of this to physical presence, something conventional learning software lacks, while also setting out the substantial technical challenges that remain.

Learning by teaching neatly sidesteps some of those challenges. A robot that is supposed to be a novice is allowed to make mistakes, misunderstand and be slow. Its limitations become part of the pedagogy rather than a bug.

Cautions

The evidence base remains small. Many studies involve modest samples, short deployments and narrow tasks such as vocabulary or letter formation, and novelty may account for some of the enthusiasm. There are also design questions about how badly the robot should perform: too competent and the child has nothing to teach, too incompetent and the child becomes frustrated or learns the wrong lesson.

Still, the idea has a rare elegance. Instead of trying to build a machine smart enough to teach a child, these researchers built one just foolish enough to be taught, and in doing so put the child back at the centre of their own learning.

References

1. Tanaka, F. and Matsuzoe, S. (2012). Children Teach a Care-Receiving Robot to Promote Their Learning: Field Experiments in a Classroom for Vocabulary Learning. Journal of Human-Robot Interaction, 1(1). Link
2. Hood, D., Lemaignan, S. and Dillenbourg, P. (2015). When Children Teach a Robot to Write. Proceedings of the Tenth Annual ACM/IEEE International Conference on Human-Robot Interaction. Link
3. Lemaignan, S., Jacq, A., Hood, D., Garcia, F., Paiva, A. and Dillenbourg, P. (2016). Learning by Teaching a Robot: The Case of Handwriting. IEEE Robotics and Automation Magazine. Link
4. Belpaeme, T., Kennedy, J., Ramachandran, A., Scassellati, B. and Tanaka, F. (2018). Social robots for education: A review. Science Robotics. Link
5. Chase, C. C., Chin, D. B., Oppezzo, M. A. and Schwartz, D. L. (2009). Teachable Agents and the Protégé Effect: Increasing the Effort Towards Learning. Journal of Science Education and Technology. Link

Feature drafted with AI from the published research listed above, for social-robotics.com.