Ryan Calo, Assistant Professor, University of Washington at MLconf Seattle 2017
summary
Identifies three concrete policy problems: introducing due process for algorithmic decisions when courts and hospitals use opaque risk scores (ProPublica/COMPAS) that trade off transparency and accuracy; protecting privacy when ML derives intimate traits from public data and when data concentration gives big firms exclusive advantage; and the unequal distribution of harms and benefits where wealthy actors can demand explanations while vulnerable people cannot. Points to technical responses like federated learning and differential privacy and to forums such as FAT ML and We Robot.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 2:01What is new
- 2:25The 1980s
- 5:10Reintroducing societal expectations
- 7:03Judges in Israel
- 8:40The Black Box
- 11:05Privacy
- 15:10Risks and Benefits
- 17:54Artificial Intelligence
- 19:30Policy Challenges
- 21:05Questions
- 21:48Special challenge with AI
- 22:42Strict liability
- 23:36Negligence
- 24:24Legal Standard
- 26:20How to get involved
related talks
01h 20mUnlocking societal benefits of AI with privacy protective technologies.
54mEthical Machine Learning
05mInserting privacy controls and due process into semi-autonomous algorithms
01h 06mPanel on Interpretability in the Law
33mMachine Learning in the age of increasing Data Privacy Consciousness
11mAddressing Privacy, Bias, and Fairness in AI Systems
43mPanel: Privacy Design Patterns for AI Systems: Threats and Protections
05mPrivacy, identity, and autonomy in the age of big data and AI