Ravi, Stevens, Hurt, Hanks, Lin & Anderson (2026). 📄 Full text (arXiv)
Ravi et al. investigate how the voice accent of a generative-ai conversational peer agent shapes learners' perceptions, trust, and interactional dynamics in k-12 group learning. While prior work examined agent accent effects in one-to-one settings, little is known about how these effects manifest in multi-party group contexts — a critical gap as AI peer agents enter collaborative classrooms.
The between-subjects mixed-methods study involved 33 teachers interacting with a GenAI voice agent in three accent conditions — British, Indian, and African American. Key findings:
1. The British-accented agent was largely treated as a tool and engaged with in detached, utility-based ways — less anthropomorphized, more like an external resource. 2. Indian- and African American-accented agents were more readily anthropomorphized and integrated as peers — building stronger trust, engagement, and reliance over time. 3. These role expectations influenced collaboration dynamics: turn-taking, questioning patterns, and perceived social presence all shifted based on accent condition.
The findings advance understanding of how GenAI's sociolinguistic design features shape group dynamics in CSCL (computer-supported collaborative learning), with implications for designing culturally inclusive AI partners. The work connects to human-ai-collaboration debates about teacher and AI roles in collaborative classrooms, and raises important questions about how accent may inadvertently reinforce or disrupt power dynamics in educational AI.
Related Pages
- human-ai-collaboration — Human-AI collaboration frameworks
- k-12 — K-12 AI in education
- equity — Equity and inclusion in AI education
- intelligent-tutoring — Intelligent tutoring systems
- teacher-role — Teacher roles in AI-mediated classrooms
- ai-literacy — AI literacy and understanding of AI capabilities