The Modern Trolley Problem - Responsible AI Principles
summary
Demo trains an object detector with only 15 training images to recognize Lego cones and people, then deploys the model to an Azure Percept deployment on a small robot car. Bounding boxes are loose—bounding boxes not semantic segmentation—so the system mislabels a dog as a Lego person but ultimately avoids collisions; works as a proof of concept but is far from production‑safe. The talk also cites Tesla emergency‑brake behavior and changing UK liability rules.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:10Classic trolley problem
- 9:09Modern warfare dilemmas
- 10:32Autonomous vehicles ethics
- 15:55Legal responsibility crashes
- 26:40Microsoft responsible AI
- 37:12Responsible AI principles
- 40:00Reliability and safety
- 42:14Computer vision tech
- 46:07Lego robot demo
- 54:20Privacy and security
- 57:10Inclusiveness for disabilities
- 58:35Transparency and trust
- 1:00:19Accountability and ethics
- 1:02:30Closing remarks and Q&A
related talks
58mThe Modern Trolley Problem - Responsible AI Principles
01h 02mThe Modern Trolley Problem - Responsible AI Principals
01h 00mThe Modern Trolley Problem - Responsible AI Principles
58mAn introduction to Machine Learning using LEGO
01h 03mThe Ethical Dilemma of AI: Save the World or Forsake It?
15mUtility vs Sanctity: Designing an ethically acceptable AI
03mMaking good robots
35mBreaking black-box AI