AI In-Person Facilitator

What

AI agent that provides real-time group facilitation for in-person deliberative processes, managing participation equity, topic focus, conflict resolution, and shared understanding across multiple simultaneous breakout groups.

Why

Remove facilitator bottleneck ↓, enable quality conversation at scale ↑, reduce facilitation costs ↓, unlock mass participation ↑ in high-quality deliberative processes.

Problem Definition

High-quality facilitated conversation is bottlenecked by human facilitator availability, limiting deliberative processes to small groups or accepting lower-quality unfacilitated discussions. Without scalable facilitation, processes must restrict participation to affordable facilitator ratios or accept dominated conversations, topic drift, and breakdown in deliberation quality.

Definition of Success

Maintains >90% participant satisfaction and achieves target outcome of deliberations >90% of the time. AI provides the same service quality as trained human facilitators across all relevant languages at <$100/agent cost. For 100-person Citizens’ Assembly with 10 breakout groups: reduces facilitation costs from $100k (10 facilitators × 5 days × $2k/day) to $1k (10 AI agents × $100).

Requirements

• Facilitation skills: understand and follow facilitation guide; apply best practice for good facilitation (active listening, feeding back, group management); and ask well framed questions.
• Real-time conversation analysis: track 8-12 distinct voices (>95% accuracy for ~3-hour sessions); measure speaking time, interruption patterns, engagement levels; recognize emotional states; track argument structure; detect energy levels; adapt communication style for different participant backgrounds; and analyze themes and detect topic drifts.
• Intervention and communication: interrupt conversations appropriately; adjust timing or focus based on group dynamics; multi-modal interaction (voice and text); and ability to escalate to human.
• In-room connection: connect with microphones in a room; and connect with screens/speakers for instructions.
• Local system: local memory; autosave; and preferably run on self-sustained battery.
• Challenges to address: conversational timing; identifying voices against ambient noise; targeted response depending on participant language; recognising conflict and dealing with it appropriately; and keeping participants on track without leading.

Existing Limitations

Most current deliberative processes rely entirely on expensive human facilitators or accept lower-quality self-facilitated discussions. Early attempts at facilitation e.g. Stanford Online Deliberation platform, provide basic facilitation via turn-taking control but only prompts engagement and conversation flow, lacking conflict resolution, cultural sensitivity, or complex group dynamics management.

Milestones

1. Text-only CustomGPT managing pre-set questions with basic turn-taking for 6-person groups
2. Full facilitation guide following with dynamic questioning and quiet participant engagement
3. Voice + text system handling complex/sensitive topics with cultural adaptation across multiple simultaneous groups

Starting Points

1. Test Stanford system to identify strengths/weaknesses.
2. Experiment with CustomGPT limitations for group interactions.