Toward Foraging for Understanding of StarCraft Agents: An Empirical Study

Users struggle to find relevant information about AI agents: The study found that even experienced participants faced "difficult foraging problems" when trying to understand how an AI agent worked, meaning they had trouble locating the specific information they needed to assess its behavior.

Important information gets overlooked: Participants "entirely missed events that were important to them," suggesting that current explanation systems may not surface critical details about agent decision-making, potentially leading to incomplete or incorrect assessments of AI system performance.

Users face high cognitive burden: Accessing information about AI agents required participants to bear "high cognitive, navigation, and information costs," indicating that explanation systems are often unnecessarily complex and difficult to navigate for non-technical users.

Explainable AI design needs user-centered research: The study demonstrates that developing better AI explanation systems requires understanding how actual users search for and process information, rather than simply providing more data—a finding that should inform how companies and researchers design transparency tools.

Current explanation approaches may force poor choices: Participants "reluctantly chose to ignore actions they did not want to ignore," suggesting that poorly designed explanation interfaces force users to make compromises that undermine their ability to properly evaluate AI systems.