Your AI Is Still Biased - Even After You Checked

Explore the ongoing challenge of AI bias in this 47-minute conference talk that challenges the misconception that bias is a one-time problem solved during development. Learn why AI systems continue to develop bias after deployment through retraining, updates, and real-world interactions, requiring continuous monitoring rather than initial checks alone. Discover how bias emerges during testing phases, model retraining cycles, and regular system usage, even when original development followed best practices. Examine practical examples demonstrating how retraining and personalization features can inadvertently introduce new biases into previously clean models. Understand the critical need for cross-functional collaboration involving designers, product managers, developers, testers, legal teams, and users in ongoing bias detection and mitigation efforts. Master strategies for integrating "de-biasing" processes into everyday workflows throughout the entire AI lifecycle, moving beyond traditional checklist approaches to embrace bias management as a continuous organizational responsibility for building fairer and more trustworthy AI systems.