AI Agent Accents Alter K-12 Group Dynamics: Study Finds British Treated as Tool
British-accented AI seen as utility; Indian and African American voices anthropomorphized as peers.
A new study from researchers (Prerna Ravi et al.) explores how voice accent of a GenAI conversational peer agent affects human-AI collaboration in K-12 group learning. In a between-subjects mixed-methods experiment, 33 teachers interacted with an AI agent that spoke with a British, Indian, or African American accent. The study, published on arXiv, analyzed surveys, group interactions, and artifacts to understand how sociolinguistic design features shape group dynamics.
The results showed stark differences: the British-accented agent was perceived as a utility tool, engaged in a detached, task-oriented manner. In contrast, the Indian- and African American-accented agents were anthropomorphized and treated as peer collaborators, fostering higher trust and emotional engagement. These findings highlight that voice accent can inadvertently bias role assignment in AI-mediated group work, urging designers to consider cultural inclusivity when building educational AI partners.
- 33 teachers in a between-subjects study tested British, Indian, and African American voice accents on a GenAI peer agent.
- British-accented agent was treated as a detached tool; Indian and African American accents led to anthropomorphization and peer integration.
- Accent influenced trust, engagement, and reliance over time, affecting group collaboration dynamics.
Why It Matters
Voice design choices in AI agents can unintentionally bias group roles, demanding culturally aware persona development for equitable learning.