Modeling Distinct Human Interaction in Web Agents

TL;DR

This paper introduces a dataset and models to predict human intervention in web agents, improving collaboration and user satisfaction by understanding when users are likely to intervene during web tasks.

Contribution

It presents CowCorpus, a dataset of human-agent web interactions, and trains language models to predict user intervention, enhancing agent adaptability and collaboration.

Findings

Intervention prediction accuracy improved by 61.4-63.4%.

User-rated agent usefulness increased by 26.5%.

Identified four distinct user interaction patterns.

Abstract

Despite rapid progress in autonomous web agents, human involvement remains essential for shaping preferences and correcting agent behavior as tasks unfold. However, current agentic systems lack a principled understanding of when and why humans intervene, often proceeding autonomously past critical decision points or requesting unnecessary confirmation. In this work, we introduce the task of modeling human intervention to support collaborative web task execution. We collect CowCorpus, a dataset of 400 real-user web navigation trajectories containing over 4,200 interleaved human and agent actions. We identify four distinct patterns of user interaction with agents -- hands-off supervision, hands-on oversight, collaborative task-solving, and full user takeover. Leveraging these insights, we train language models (LMs) to anticipate when users are likely to intervene based on their interaction styles, yielding a 61.4-63.4% improvement in intervention prediction accuracy over base LMs. Finally, we deploy these intervention-aware models in live web navigation agents and evaluate them in a user study, finding a 26.5% increase in user-rated agent usefulness. Together, our results show structured modeling of human intervention leads to more adaptive, collaborative agents.