Human-Guided Data Exploration

Human-Guided Data Exploration

Collaborative AI Safety Through Interactive Data Augmentation

Amplio is an interactive tool that helps practitioners identify and address potential model weaknesses through human-guided data augmentation, particularly for discovering "unknown unknowns."

  • Combines human creativity with AI assistance to generate diverse, high-quality test cases
  • Enables systematic exploration of potential edge cases and failure modes
  • Demonstrated effectiveness with professional red teamers for enhancing model safety
  • Provides an interactive workflow for efficiently navigating complex data spaces

This research advances security practices by creating a structured approach to identify potential harms before deployment, making AI systems more robust against unpredictable inputs and reducing safety risks.

Exploring Empty Spaces: Human-in-the-Loop Data Augmentation

21 | 104