This conference talk from SREcon25 Americas explores the challenges and potential solutions when integrating AI into operational systems. Join Hazel Weakly from Nivenly Foundation and Sandeep Kanabar from Gen as they examine what happens when AI—often described as "garbage in, garbage out"—is incorporated into carefully tuned systems. The speakers share their personal experiences with biased and broken systems, discussing known AI drawbacks including data poisoning, bias, inaccessibility, and de-skilling. Rather than accepting a future dominated by inscrutable black box AI systems, discover practical strategies for addressing current issues and preventing future problems. Learn how to transform AI into a transparent, reliable tool that enables innovation rather than creating chaos in your systems.
Overview
Syllabus
SREcon25 Americas - Systems Thinking with Poisoned Systems
Taught by
USENIX
Related Courses
-
Ethics in AI Design
-
Safe Evaluation and Rollout of AI Models
-
Understanding AI Ethics and Governance - Building Responsible and Trustworthy Systems
-
MLOps Under Attack: Threat Modeling Modern AI Systems
-
Bringing Light to Risks Lurking in the Black Boxes of AI Models
-
Bias Identification and Mitigation in AI Systems
