The Modern Trolley Problem - Responsible AI Principles
summary
Demonstrates training a model with Custom Vision (started with 15 images, ~1 hour to train ~100) and deploying it to an Azure Percept device to detect cones and LEGO figures; the robot uses bounding-box object detection (not semantic segmentation) so it sometimes mislabels a toy lion as a Lego man. Covers Responsible AI specifics: fairness checks, GDPR-style restrictions on facial recognition, telemetry for accountability, and recommendations for larger, diverse training sets and production-grade updates and security.
AI-written summary · about
chapters (7)
- 0:00Start
- 3:04What is the Trolley Problem?
- 12:04What is the Modern Trolley Problem?
- 29:54Responsible Al principles
- 33:18The 3 algorithms
- 36:55Machine Learning Algorithms
- 39:24Model Training Demo
related talks
01h 00mThe Modern Trolley Problem - Responsible AI Principles
01h 02mThe Modern Trolley Problem - Responsible AI Principals
01h 04mThe Modern Trolley Problem - Responsible AI Principles
01h 03mThe Ethical Dilemma of AI: Save the World or Forsake It?
01h 04mBuilding and Training your own Custom Image Recognition AI
43mIntersection of Awesome: Building AI enabled IoT solutions powered by the cloud
01h 00mLearn Live: Get started with AI on Azure
01h 02mAI on the Edge