BoardroomAI: Human-Steerable Multi-Agent Deliberation via Evolving Decision Graphs
There's this new preprint on arXiv, identified as 2608.13046, that introduces BoardroomAI, a framework for enhanced discussions among multiple agents with human guidance. Unlike traditional systems where humans just set the stage for agents to work solo, BoardroomAI keeps humans engaged in the process. They can question assumptions, tweak constraints, shift priorities, introduce new information, or guide the decision-making. The framework includes four main features: a decision graph that lays out all relevant elements; an intervention compiler for translating human actions into graph changes; a method for tracking which parts of the graph are affected; and a protocol to keep discussions flowing smoothly. It's tailored for decision-making in organizations and adjusts based on changing information and priorities.
Key facts
- BoardroomAI is a framework for human-steerable multi-agent deliberation.
- It uses a typed decision graph to represent evidence, assumptions, constraints, claims, objections, alternatives, risks, decisions, semantic dependencies, and specialist responsibility.
- An intervention compiler converts confirmed human actions into explicit graph updates.
- Dependency-aware propagation identifies affected subgraphs and preserves unaffected artifacts.
- The system allows humans to challenge assumptions, modify constraints, change priorities, introduce evidence, or redirect the decision process.
- The paper is available on arXiv under identifier 2608.13046.
- The framework is designed for organizational decisions where evidence and priorities evolve.
- The approach contrasts with conventional transcript-based multi-agent systems.
Entities
Institutions
- arXiv