Design knowledge for virtual learning companions from a value-centered perspective

Students who work alongside their studies do not lack information. They lack time, and on bad days they lack the will. Existing pedagogical agents attack these problems separately and functionally: a scheduling bot here, a goal-setting feature there, a chatbot tutor that answers questions. A literature review of 252 recent publications found only five that mention a learning companion at all, and none that offered design knowledge for one.

The rise of ChatGPT sharpened the gap. A system that hands over finished answers can undermine exactly the engagement that learning depends on. The paper’s premise is that companionship, a bonding relationship in which learner and agent work toward a shared goal, is a prerequisite for supporting motivation and time management, not a nice-to-have on top.

The unconventional move is to treat the companion not as a product with features but as a service in which value is co-created. Following the value-in-interaction model, that value decomposes across three layers, and the paper’s nine principles are organised by which layer they primarily serve. They are not separable: if the relationship is missing, the service is not used; if the service is useless, no relationship saves it.

Behind the principles stand several theories: people treat computers as social actors, so social cues work; conversations need common ground; flow theory and self-determination theory explain what keeps learners engaged; and Maslow’s higher-order needs frame learning as growth rather than knowledge transfer. The three relationship principles map directly onto the earlier design theory for virtual companionship, transferred to the learning context.

Alex greets learners by first name, has a youthful comic-style avatar accompanied by a dog that learns tricks when the student reaches a goal, uses emojis and memes, adapts to the learner’s learning style, keeps a course schedule with reminders to drink water and take a walk, explains the Pomodoro technique, runs a class-wide scoring model, and lets students message peers and teachers from the chat. Twenty-two design features were implemented in total.

Both groups appreciated the human and friendly interaction and welcomed the organiser functions. Then the two groups diverged. Students at the technical university found Alex too informal, the emojis childish, and liked the competitive game elements. Students in continuing education found Alex too formal, wanted more mutual praise and reaction to their mood, and found the game story distracting. Same prototype, opposite feedback.

The paper reads this as evidence for its central connecting principle. Companionship is perceived very differently by different people, so adaptation, both by the system and by the user, is what bridges the social relationship layer and the functional service layer. There is no one-size-fits-all learning companion.

The contribution is deliberately layered: nine principles for orientation, 28 meta-requirements for justification, 33 design features for implementation, and three instantiations that show the leeway in between. The comic avatar and the dog, for instance, are one way to be human-like without being human, suited to students in their twenties and probably wrong for mature professionals.

The paper is also candid about controversy. A focus group questioned whether human-likeness distracts from learning. Workshop participants asked whether an organiser that suggests appointments is a coach rather than a peer. Experts doubted that emotion detection is reliable enough not to damage the interaction. The answer to all three is the same: keep the human-like elements judicious, let the companion take different roles as human companions do, and let learners switch features off.

The paper was recognised twice by the community: with the Paul Gray Award for the most thought-provoking paper in Communications of the AIS in 2024, and with the AIS Senior Scholars’ Best Information Systems Publications Award in 2025. The workshop paper it grew from had already won the best paper award at the Pre-ICIS Workshop on Education and Research in 2022.