ExploreGen: Large Language Models for Envisioning the Uses and Risks of AI Technologies

LLMs can help AI teams identify overlooked risks: ExploreGen generated realistic use cases for facial recognition technology that AI practitioners hadn't previously considered, demonstrating that AI developers often miss potential applications and associated risks during early design stages.

Automated risk classification shows promise: The framework classified use cases according to EU AI Act regulations with 94.5% accuracy according to user ratings, suggesting LLMs can help standardize and systematize the currently difficult task of assessing regulatory compliance.

Brainstorming support improves responsible AI practices: The tool enabled "reflexivity, brainstorming, and deliberation" among practitioners, addressing a documented struggle that appears in model and data cards—suggesting LLMs can make responsible AI design processes more systematic and thorough.

Identified uses had significant potential impact: Practitioners rated the AI-generated use cases as having "high adoption potential and transformational impact" despite being unfamiliar with many of them, indicating the framework surfaces genuinely important scenarios that teams should consider during development.

Practical value across different roles: Both AI developers and compliance experts found ExploreGen helpful, suggesting the tool addresses a genuine need across multiple functions involved in responsible AI governance.