Consistent guidance, happier learners: A new SCIoI study on robotic tutors

By: Maria Ott

Paper DOI →

The first encounter with Pepper, the robot designed to teach among other tasks, brought with it the promise of reshaping education. Rather than being seen as just another machine, Pepper was crafted to be an engaging, interactive educator, supporting teachers in the classroom by personalizing interaction and adapting to the learner’s pace. It promised to possess the ability to bridge technology and human interaction, leading to a more dynamic, inclusive, and responsive learning environment. Yet, a new study titled “How adaptive social robots influence cognitive, emotional, and self-regulated learning”, published in scientific reports, and led by Helene Ackermann, Anna L. Lange, Verena V. Hafner, and Rebecca Lazarides from Science of Intelligence, reveals that when it comes to learning with and from an artificial agent or robot, predictability may matter more than personal adaptation. In this experiment, 120 participants tackled a vocabulary learning task with Pepper as their tutor, and the findings were both surprising and intriguing.

The study compared several teaching approaches to investigate how adaptive teaching impacts learning. In the non-adaptive conditions, Pepper either provided basic correctness feedback — simply indicating whether an answer was right or wrong — or consistently delivered hints every time a mistake was made. In contrast, in the adaptive conditions Pepper individually adjusted the number of hints based on the learner’s performance and self-reported enjoyment. In the more advanced personalized adaptive condition, the robot not only adjusted the frequency of hints, but also tailored the feedback to address the learner’s specific errors.

Contrary to what one might expect, the adaptive approach did not enhance performance or the learning outcome. In fact, when Pepper’s behavior fluctuated based on real-time assessments, participants reported enjoying the interaction less than in the condition where they always received a hint. It seems that the unpredictability of the adaptive feedback disrupted the flow of learning, making the experience less enjoyable compared to the consistent, enhanced guidance. In other words, when students know what to expect from a robotic interaction partner—even if it means consistently receiving hints, regardless of whether they were needed or not—they seem to feel more comfortable.

There was, however, an intriguing twist in the findings. When Pepper offered personalized adaptive feedback—directly addressing individual mistakes—learners were more likely to engage in self-regulated learning. They took extra time to reflect on their errors and adjust their strategies, ultimately boosting vocabulary test scores, suggesting a potential for fostering long-term learning habits through deeper self-reflection.

These results challenge the assumption that adaptive technology automatically leads to better educational outcomes. Instead, the study highlights the delicate balance robots must strike between being responsive and maintaining a consistent, predictable learning environment. For educators and developers, this is an important lesson. As schools increasingly turn to add technology to meet diverse learning needs, the human desire for reliability remains a critical factor in how students experience and benefit from these tools—especially social robots.

The work from the SCIoI invites us to rethink the design of adaptive systems. It appears that the key to effective robot tutors may not lie solely in their ability to adjust on the fly, but in finding the right mix between personalization and consistency. For now, it seems that a steady, predictable approach might better support student engagement and, ultimately, long-term learning success.

In a time where technology is set to play an ever-larger role in education, these findings remind us that sometimes the simplest approach—one that respects the need for routine and certainty—can make all the difference.

Research

An overview of our scientific work

See our Research Projects