Algorithmic UDAP
Talkat Bridging Prediction and Intervention Problems in Social Systems (Simons Institute, 2026) · 2026-01-12..01-18·
summary
Compares disparate impact and UDAAP by formalizing both in a simulated lending setting using Adult Income, German Credit and HMDA data, training many classifiers and retaining models in the top 5% accuracy and within 10% of the best; flags a highest‑accuracy baseline model and searches for less discriminatory alternative models that meet business constraints. Shows disparate impact requires alternatives with similar performance while UDAAP’s substantial injury and scaling trade‑off let regulators weigh harms vs benefits; cites recent state settlements (Earnest 2025, Pieces 2024) and a MelApp false‑advertising precedent.
AI-written summary · about
chapters (11)
- 0:00Algorithmic discrimination law intro
- 2:39Disparate impact limits
- 5:42Disparate impact vs UDAP
- 8:00Less discriminatory alternatives
- 11:19UDAP consumer protection
- 14:51Discrimination doctrines comparison
- 17:28Model fairness trade-offs
- 21:28UDAP beyond allocation
- 24:55State UDAP laws
- 27:57Future of algorithmic harm
- 28:22Q&A
related talks
35mBuilding Inclusive Machine Learning Models
46mA look at the methods to detect and try to remove bias in ML models
28mMeasuring Model Fairness
27mThe ethical algorithm
30mBuilding fairer models for finance
38mPrivacy: the Last Stand for Fair Algorithms
35mOpen Source Fairness
51mDigital Discrimination: Cognitive Bias in Machine Learning