Minimizing Risk Through Minimizing Model-Data Interaction - A Protocol for Relying on Proxy Tasks When Designing Child Sexual Abuse Imagery Detection Models

Learn about a risk-minimization protocol for developing child sexual abuse imagery detection models through this 11-minute conference talk from the Association for Computing Machinery. Explore how researchers from Brazilian and UK universities propose using proxy tasks to reduce direct interaction between machine learning models and harmful content during the development process. Discover the methodology for maintaining detection effectiveness while minimizing exposure to illegal material, examine evaluation practices for responsible system development, and understand the ethical considerations involved in creating AI systems for child protection. Gain insights into balancing technical requirements with safety protocols when working with sensitive and harmful digital content in machine learning applications.