Is your AI plotting to kill you? Why AI and Machine Learning is dangerous
summary
A survey of concrete failures and proofs of concept: the New York Times feeds a Bryant Park camera into AWS Rekognition to match people against public profiles for about $60 cloud demo; University of Antwerp and MIT/IBM publish printed adversarial patch t-shirt research that partly fools person detectors; a 100k faces dataset produces 100k generated faces indistinguishable from stock photos; GPT-2 is first withheld then released; Lyrebird voice clones from 40 sentences voice clone enable fraud; Amazon hiring AI and Microsoft Tay show clear bias and rapid degeneration.
AI-written summary · about
chapters (27)
- 0:00Intro
- 0:58What is AI
- 1:28History of AI
- 3:48Image Recognition
- 4:43Bad Things
- 5:04Facial Recognition
- 7:39The New York Times
- 9:24Public Cameras
- 10:22Adversarial Patches
- 11:50Printed TShirts
- 12:45Drawing
- 15:10AI Faces
- 16:59AI Text
- 18:08Beer Review
- 19:59Lyrebird
- 23:20AI can be dangerous
- 24:55Deepfakes
- 26:30AI
- 27:09People
- 27:27Problems
- 31:52People are awful
- 32:13Common mistakes
- 33:49Limitations of AI
- 37:28How can you protect yourself
- 38:00Building models
- 39:54Mix teams
- 41:44Conclusion
related talks
37mThe Shadows That Follow AI Generative Models
30mHow to Lie with AI: Understanding Bias, Ethics, and the Hidden Risks in ML
49mAn AI with an Agenda
31mAvoid mistakes building AI products
39mDefensible Application Security for the Artificial Intelligence Era
26m(Un)ethical Artificial Intelligence: How to Keep the AI Fair for Everyone
41mWhen AI Failures Reflect Our Own Social Biases
36mDisinform your Surroundings: AI and disinformation campaigns