$ au^2$-Bench: Evaluating Conversational Agents in a Dual-Control Environment

2025-06-09

AI Summary

Existing dialogue AI evaluation benchmarks assume unidirectional control—where only the AI invokes tools—overlooking real-world collaborative scenarios requiring active user participation and joint environment modification (e.g., telecom network operations). Method: We propose $τ^2$-bench, the first benchmark supporting bidirectional control. It formalizes dual-agent dynamic collaboration via Decentralized Partially Observable Markov Decision Processes (Dec-POMDP), integrates a programmable task generator, and introduces a tool-constrained, high-fidelity user simulator. It further enables fine-grained root-cause analysis of reasoning errors and collaborative communication failures. Contribution/Results: Experiments reveal significant performance degradation of state-of-the-art dialogue agents under bidirectional control, empirically validating collaboration bottlenecks. $τ^2$-bench is reproducible, extensible, and diagnosable—establishing a foundational infrastructure for next-generation dialogue agent evaluation.

Technology Category

• Multiagent Systems: Adversarial Agents
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

Existing benchmarks for conversational AI agents simulate single-control environments, where only the AI agent can use tools to interact with the world, while the user remains a passive information provider. This differs from real-world scenarios like technical support, where users need to actively participate in modifying the state of the (shared) world. In order to address this gap, we introduce $ au^2$-bench, with four key contributions: 1) A novel Telecom dual-control domain modeled as a Dec-POMDP, where both agent and user make use of tools to act in a shared, dynamic environment that tests both agent coordination and communication, 2) A compositional task generator that programmatically creates diverse, verifiable tasks from atomic components, ensuring domain coverage and controlled complexity, 3) A reliable user simulator tightly coupled with the environment, whose behavior is constrained by tools and observable states, improving simulation fidelity, 4) Fine-grained analysis of agent performance through multiple ablations including separating errors arising from reasoning vs communication/coordination. In particular, our experiments show significant performance drops when agents shift from no-user to dual-control, highlighting the challenges of guiding users. Overall, $ au^2$-bench provides a controlled testbed for agents that must both reason effectively and guide user actions.

Problem

• Evaluating AI agents in dual-control conversational environments
• Addressing gaps in single-control AI benchmarks
• Testing agent coordination and user guidance capabilities

Innovation

• Dual-control environment for agent-user interaction
• Compositional task generator for diverse tasks
• Reliable user simulator with constrained behavior