Human-Agent Teaming: Trust Calibration in Semi-Autonomous Agentic AI

ABSTRACT

As Agentic AI systems assume more responsibility in high-stakes domains, calibrating human trust remains a critical challenge. Under-trust leads to inefficient micromanagement, while over-trust results in catastrophic failures when agents hallucinate. This paper introduces the 'Dynamic Trust Calibration' (DTC) framework, which models human-agent teaming as a cooperative game. By forcing agents to explicitly quantify their confidence levels and proactively request human-in-the-loop intervention during high-uncertainty tasks, the DTC framework improves joint task performance by 38% in simulated air-traffic control and medical diagnostics scenarios.