← CHI 2026 map

CHI '26 · Best paper · full-paper review · confidence high

Novobo: Supporting Teachers' Peer Learning of Instructional Gestures by Teaching a Mentee AI-Agent Together

Jiaqi Jiang , Kexin Huang , Huan Zeng , Duo Gong , Roberto Martinez-Maldonado , Pengcheng An

DOI PDF Program page Page JSON

Novobo is a notable CHI contribution because it shifts teachable agents from student-facing learning support into teacher professional development and ties that move to a plausible social mechanism. The strongest contribution is not merely the LLM pipeline, but the interaction framing: by having teachers jointly teach an AI apprentice, the system appears to lower interpersonal pressure and make tacit gesture knowledge easier to discuss, demonstrate, and refine together.

Video Figure

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

Novobo stands out because it does something more interesting than adding AI to teacher training: it changes the social role AI plays. Instead of positioning the system as an evaluator that judges teachers’ gestures, it casts the system as an apprentice that teachers mentor together. That reframing matters because the paper’s central problem is not only technical generation of gestures, but also the social difficulty of peer learning around tacit, embodied teaching practices. The introduction makes clear that existing approaches can be expert-heavy, prescriptive, or socially awkward, especially when peers hesitate to criticize one another directly. Within that framing, Novobo’s contribution is a well-formed artifact and interaction design that uses verbal commentary, bodily demonstration, and a skeletal mirror to support collaborative reflection. The evidence base is reasonably strong for the paper’s intended scope. The user study covers 30 teachers in 10 collaborative sessions and is explicitly aimed at understanding how the teachable agent supported peer learning and how teachers experienced the design features. The abstract and introduction both support the claim that teachers externalized and shared tacit knowledge and experienced reduced pressure when teaching the AI mentee together. In addition, the technical evaluation is not trivial window dressing: the pipeline was compared against a baseline using the same base model, and it significantly outperformed that baseline on four quality metrics. That helps establish that the system’s generated content was at least credible enough to support the interaction design. Still, the paper should be read with clear boundaries. The limitations section is substantive and real: gesture output is text-based and therefore indirect; the skeletal mirror depends on camera position and may constrain movement capture; the participant pool is narrow in geography and AI familiarity; the current setup assumes synchronous collaboration; and the study does not measure individual teacher learning gains or student outcomes. So this is not evidence that Novobo broadly improves teaching effectiveness. Rather, it is strong evidence for a promising social-technical mechanism in a bounded context: an AI apprentice can serve as a focal object that helps teachers exchange, articulate, and co-construct embodied professional knowledge with less interpersonal friction. That is a meaningful and publishable CHI contribution.

What Changed

Canon before

Prior dominant assumptions include that training teachers' nonverbal instructional gestures is time-consuming, isolated, or prescriptive, e.g., relying on expert video review or binary AI feedback; peer learning for teachers' tacit gesture knowledge is obstructed by social hesitancy to criticize peers; and teachable agents have been almost exclusively designed for students rather than teacher professional development.

Departure from common sense

Contrary to assumptions that training instructional gestures requires expert-driven or prescriptive approaches and that peer feedback is hindered by social pressure, this work argues that a jointly taught AI mentee can create a lower-pressure setting for teachers to externalize tacit gesture knowledge and learn from one another.

Actual novelty

This paper presents Novobo, a teachable AI-powered apprentice agent for teachers’ collaborative learning of instructional gestures, combining verbal and bodily inputs, a multi-agent LLM pipeline, and a skeletal mirror interface, then evaluating how this design supports externalization, peer exchange, and co-construction of practical knowledge.

Evidence

Evidence comes from two complementary evaluations in the provided sections: a user study with 30 teachers in 10 collaborative sessions examining peer-learning support and experience of the design, and a technical evaluation showing the gesture-generation pipeline outperformed a single-LLM baseline on pedagogical meaningfulness, appropriateness, naturalness, and understandability. The qualitative claims are strongest around reduced peer pressure, externalization of tacit knowledge, and collaborative knowledge construction.

Limits

Method limits

The system currently conveys generated gestures through text, which requires teacher interpretation and may introduce variability. The skeletal mirror is camera-dependent and can fail to capture the full body or gesture when positioning is poor. The study emphasizes interaction processes rather than direct measurement of individual learning gains or student outcomes.

Deployment limits

Participants were exclusively teachers from high socio-economic status cities and were relatively familiar with AI in educational contexts, so transfer to other settings is uncertain. The current approach also requires synchronous communication, which may be difficult for teachers with busy schedules.

Boundary conditions

The paper is best supported for collaborative teacher study-group contexts focused on instructional gestures, especially synchronous sessions where teachers jointly mentor an AI apprentice. Claims should not be generalized to all teacher populations, lower-resource contexts, asynchronous settings, or to downstream improvements in teaching performance or student learning without further study.

Position in field

The work extends teachable-agent research beyond student learning into teacher professional development and contributes a concrete HCI artifact for embodied, tacit-skill peer learning. Its main field position is as a design pattern showing how reframing AI as a mentee rather than evaluator can reduce social friction in peer feedback.

Abstract

Instructional gestures are essential for teaching, enhancing communication and student comprehension. Current training methods for developing these skills can be time-consuming, isolating, or overly prescriptive, e.g., watching lengthy, one-size-fits-all videos. Conversely, research suggests that developing these tacit, experiential skills requires teachers’ peer learning, where they learn from each other and build shared knowledge. While much HCI exploration has applied learning-by-teaching to students’ peer learning, little has explored this approach for teacher professionalization. We present Novobo, an apprentice AI-agent stimulating teachers' peer learning of instructional gestures through verbal and bodily inputs. An evaluation with 30 teachers in 10 collaborative sessions showed Novobo prompted teachers to externalize and share tacit knowledge through dialogue and movement. Teaching an AI mentee together reduced their pressure, facilitating peer exchange and the co-construction of practical knowledge. This work contributes a novel design and empirical insights into how teachable AI-agents can facilitate peer learning in teacher professionalization.