The Dark Side of AI
summary
Argues that accuracy numbers hide real harm: a model at 95% accuracy still produces the 1% error that triggers actions and user rejection. Describes a released mixing console with AI that provokes social backlash (“waste of R&D”, job-loss fears) and extracts a design checklist: state purpose, show performance and confidence, allow easy override, and make automation levels adaptable. Applies the checklist to SAE levels 0–5 and to the Boeing 737 MAX MCAS failures to show missing feedback, training, and intermediate automation.
AI-written summary · about
chapters (14)
- 0:00Start
- 1:53ALGORITHMS
- 2:56MACHINE LEARNING PREDICTIONS
- 3:55DATA PREDICTIONS ACTIONS
- 5:181% ? ACTIONS ?
- 9:18FEAR OF AUTOMATION IN MUSIC
- 11:24ACTIVE USER PARADOX
- 13:15A METAPHOR FOR TRUSTABLE AUTOMATION
- 13:31AUTOMATION AS A COGNITIVE PROCESS
- 13:59HEARTBEAT
- 14:31BREATHING
- 15:363. Decision selection 4 Action implementation
- 21:05POSSIBLE APPROACHES
- 22:59AVIONICS Boeing 737 Max Case Study
related talks
49mAn AI with an Agenda
09mML and AI systems will make mistakes
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
31mAvoid mistakes building AI products
52mSome pitfalls in AI
47mEthics in AI Panel
01h 05mNavigating the Security and Privacy Landscape of Modern AI
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone