Despicable machines: how computers can be assholes
summary
A deep learning model claims to infer criminality from still face images, reportedly reaching 90% accuracy using AlexNet; the pipeline trains on convicted-person photo sets and feeds into predictive policing and automated decisions. The system is opaque, produces spurious correlations, and amplifies human bias through a feedback loop, a phenomenon the speaker calls mass-washing. Empirical surprise and concrete metrics drive the risk: high numeric accuracy does not prove causal validity and the classifier demonstrably misattributes protected traits.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:27Human perception of machines
- 3:39Machines as decision makers
- 6:12Machine learning risks
- 7:38Biased ML model example
- 9:24Why models learn wrong patterns
- 12:36Data washing dangers
- 14:45Real-world implications
- 18:20Feedback loops in predictions
- 20:08Societal impact of ML
- 23:26Ethical hiring concerns
- 24:30Solving biased models
- 26:49Our responsibility
- 29:00Question and answer session
related talks
54mWe are the guardians of our future
31mDoes This Algorithm Have a Soul?
01h 01mWe are the Guardians of our Future
32mThe Miseducation of This Machine
47mEthical Machine Learning
34mNeural networks: insane in the membrane
44mComputers are Stupid: Protecting "AI" from Itself
45mThe Myth of Neutrality: How AI is widening social divides