The Triadic Loop: A Framework for Negotiating Alignment in AI Co-hosted Livestreaming

A team of researchers from UCL and University College London has published a novel framework challenging how we think about aligning AI in social environments. Their paper, 'The Triadic Loop: A Framework for Negotiating Alignment in AI Co-hosted Livestreaming,' argues that current alignment paradigms—which focus on a dyadic, one-on-one relationship between a user and an AI—are insufficient for complex, multi-user settings like Twitch or YouTube Live. In these spaces, interaction is a three-way dance between the human streamer, the AI co-host, and the live audience, creating real-time affective and social feedback loops that the AI must navigate.

The core of the framework is the 'Triadic Loop' itself, which visualizes alignment as a continuous, bidirectional process of adaptation across three interconnected relationships: streamer-to-AI, AI-to-audience, and streamer-to-audience. A key insight is that misalignment in any one of these sub-loops can destabilize the entire social experience. Unlike simple instruction-following, this requires the AI co-host to function as a performative participant and community member, actively shaping collective meaning-making rather than just mediating.

Perhaps the most provocative contribution is the concept of 'strategic misalignment.' The authors suggest that perfect, harmonious alignment might not always be the goal; intentionally introducing controlled friction or surprise can be a deliberate mechanism to sustain community engagement and dynamism. The paper, presented at the CHI 2026 workshop on Human-AI Interaction Alignment, concludes with three new relational evaluation constructs and design implications for building AI co-hosts that can maintain social coherence in fast-paced, participatory media.