Stanford CS547 HCI Seminar | Spring 2026 | Promoting Agency in Human-AI Interaction
Stanford Online · 44:20 · 2 days ago
AI systems designed to function as "advisors" rather than mere "assistants" increase human agency and long-term engagement. By prioritizing non-prescriptive communication—where the system facilitates user reflection rather than simply issuing commands—and utilizing probabilistic models to manage uncertainty regarding user goals, AI can provide more effective and personalized support.
- Advisor vs. assistant — Assistants automate objective tasks like coding or formatting, while advisors help users clarify their own intentions to support behavioral change .
- Motivational interviewing — This counseling style emphasizes avoiding unsolicited advice, focusing instead on extracting a user's personal reasons and motivations for change .
- Qualitative context — Incorporating subjective user values alongside quantitative data, like step counts, leads to more effective coaching and stronger user rapport .
- Prompt chaining — Structuring AI instructions into distinct phases forces the model to ask clarifying questions rather than immediately offering solutions, maintaining a facilitative role .
- Evaluation results — While standard AI models often gave unsolicited advice, the developed architecture maintained non-prescriptive communication strategies over 93% of the time .
- Bayesian planning — Treating a conversation as an uncertain state allows AI to view unknown user needs as a statistical variable, enabling the model to balance the trade-off between asking questions and offering advice .
- User engagement — Participants using the autonomy-promoting interface spent over five times as much time in the application compared to control groups using rigid, template-based systems .
How does modeling user state as a statistical variable improve AI responses during multi-turn conversations? What methods are used to determine when an AI should prioritize asking a question over providing a recommendation?