Controlling AI Agent Participation in Group Conversations: A Human-Centered Approach

TL;DR

This paper investigates how AI agents should participate in group conversations by studying a Slack bot named Koala in group ideation tasks. Through two mixed-method studies, Koala I reveals that AI participation is valuable but requires user controlled mechanisms to manage proactivity, tone, and placement. Koala II implements enhanced proactive behaviors and introduces a user facing control panel, demonstrating improved pacing, topic adherence, and user satisfaction, while still requiring further refinements. The authors derive a taxonomy for when, what, where, how, and who can control AI behaviors in groups, offering a human centered design framework to guide the development of mixed-initiative, proactive conversational agents in collaborative settings.

Abstract

Conversational AI agents are commonly applied within single-user, turn-taking scenarios. The interaction mechanics of these scenarios are trivial: when the user enters a message, the AI agent produces a response. However, the interaction dynamics are more complex within group settings. How should an agent behave in these settings? We report on two experiments aimed at uncovering users' experiences of an AI agent's participation within a group, in the context of group ideation (brainstorming). In the first study, participants benefited from and preferred having the AI agent in the group, but participants disliked when the agent seemed to dominate the conversation and they desired various controls over its interactive behaviors. In the second study, we created functional controls over the agent's behavior, operable by group members, to validate their utility and probe for additional requirements. Integrating our findings across both studies, we developed a taxonomy of controls for when, what, and where a conversational AI agent in a group should respond, who can control its behavior, and how those controls are specified and implemented. Our taxonomy is intended to aid AI creators to think through important considerations in the design of mixed-initiative conversational agents.

Figures (7)

• Figure 1: Koala as an AI participant in Slack. These screenshots (with human participant names redacted) show examples of Koala participating in a Slack channel. (A) The reactive variant of Koala replies to a user's question addressed to "Koala" (A.1) or to "@Koala" (A.2). (B) The proactive variant of Koala generates a proactive reply (B.1) to the conversation and a reactive reply (B.2) in response to a direct request.
• Figure 2: Koala operational logic. When users post a message in a Slack (A), the post triggers an event that is handled by the Koala backend (B) where control logic determines whether Koala should either immediately pass on replying or hand off for further evaluation in the autonomy control logic (C) where the LLM (D) generates a response that is further evaluated for potential posting in the channel.
• Figure 3: Study 1 overview. During each session, a group of participants sequentially completed three rounds of brainstorming and post-brainstorm surveys. The order of conditions (No AI, Reactive AI, Proactive AI) was kept the same for each group to assess the impact of increasing levels of AI autonomy on participants' experience. Brainstorming topics were assigned in a counterbalanced fashion to negate order effects.
• Figure 4: (a) Impact of Koala on the production of ideas and (b) participants' preferences for the different Koala variants.
• Figure 5: Control settings. Koala II's control settings provides users with the ability to control aspects of its interactive behaviors. These settings include (A) switching between proactive and reactive variants; (B) adjusting the threshold at which the proactive variant deems a generated response valuable to send to the group; (C) selecting whether Koala's responses should appear in the conversation or in a thread; and (D) managing how long messages are displayed.