Overseeing Agents Without Constant Oversight: Challenges and Opportunities

TL;DR

This paper investigates how to improve human oversight of agentic AI systems by designing better action traces and interfaces, revealing challenges and potential solutions for effective verification.

Contribution

It introduces a novel interface design that reduces error detection time and explores the complexities of human verification of AI agents.

Findings

Current trace practices are cumbersome and limit efficacy.

The proposed design reduces error-finding time.

Participants' confidence increased without improving accuracy.

Abstract

To enable human oversight, agentic AI systems often provide a trace of reasoning and action steps. Designing traces to have an informative, but not overwhelming, level of detail remains a critical challenge. In three user studies on a Computer User Agent, we investigate the utility of basic action traces for verification, explore three alternatives via design probes, and test a novel interface's impact on error finding in question-answering tasks. As expected, we find that current practices are cumbersome, limiting their efficacy. Conversely, our proposed design reduced the time participants spent finding errors. However, although participants reported higher levels of confidence in their decisions, their final accuracy was not meaningfully improved. To this end, our study surfaces challenges for human verification of agentic systems, including managing built-in assumptions, users' subjective and changing correctness criteria, and the shortcomings, yet importance, of communicating the agent's process.