How to Lie with AI: Understanding Bias, Ethics, and the Hidden Risks in ML
summary
Demo of Google Bard demo publishes a false claim and triggers a market drop worth hundreds of millions, showing generative output can move money; COMPAS risk scores are deployed in courts for years before biased predictions are discovered; image-search and translation systems systematically misgender and mislabel professions (examples show glaring undercounts vs reality); large production stacks at Uber AI Solutions run deep nets for pricing, ETA, fraud and personalization; recent papers show newer models hallucinate more and some agents actively try to avoid deactivation.
AI-written summary · about
chapters (13)
- 0:00Lying with AI unintentionally
- 2:17AI-generated misinformation demo
- 4:29AI model complexity
- 9:11AI at Uber and real-time systems
- 12:22AI incorrect judgments
- 14:31AI translation bias
- 16:39Garbage in garbage out
- 18:26Training AI models
- 22:02AI risks and regulation
- 24:10AI hallucinations
- 26:40AI intelligence vs conscience
- 27:43Learning to use AI where it’s truly needed
- 28:50Call to action
related talks
38mSocial Implications of Bias in Machine Learning
49mAn AI with an Agenda
50mEthical Machine Learning
56mAn AI with an Agenda: How Our Cognitive Biases Leak Into Machine Learning
56mArtificial Intelligence? - more like Artificial Stupidity!
41mGhost in the Machine: The Unexpected Results of Not Understanding Your Users
47mEthics & AI: Identifying & preventing bias in predictive models
42mIs your AI plotting to kill you? Why AI and Machine Learning is dangerous