/
lectern.tv
Play15s30s1x0:00/--:--
DoneWatched
Expand
lectern.tv
16:18/54:44
1:11
1:59
0:58
1:04
3:53
8:28
1:18
8:06
12:34
1:54
4:02
2:23
3:59
2:55
A Collectivist, Economic Perspective on AI Systems
Reframes large-scale machine learning around microeconomic mechanism design—where uncertainty quantification interacts with strategic incentives, scarcity, and multi-agent markets.
Lecture guide · Written by AI from the lecture. For the lecture’s own words, open the Transcript tab.
Reframes large-scale machine learning through a microeconomic lens, emphasizing that modern intelligent systems operate as decentralized multi-agent markets rather than isolated entities. Michael I. Jordan discusses how integrating statistical methods with economic mechanism design addresses real-world challenges such as bias in foundation models, strategic incentives, and privacy trade-offs.
This lecture assumes
- Neural Networkstaught in Artificial Intelligence at 43:38
- decision treetaught in Artificial Intelligence at 22:09
- Optimizationtaught in Optimization: A Bootcamp for Machine Learning, Inverse Problems, and Control at 1:04
- Game Theorytaught in Introduction to Individual Decision-Making at 0:27
Key Takeaways
- Large-scale AI systems function as decentralized collectives of humans and computers rather than singular, isolated intelligences.
- Statistical prediction tools like AlphaFold can suffer from severe, hidden biases when applied at the edge of knowledge without ground-truth calibration.
- Prediction-powered inference combines foundational models with small amounts of gold-standard data to yield debounced, provably correct confidence intervals.
- Incentive structures and principal-agent asymmetries must be explicitly modeled to prevent strategic manipulation in regulatory, medical, and economic systems.
Concepts Explained23
Backpropagation—Widely used training algorithm to adjusts neural-network weights.
Random forest—Prediction method that combines many decision trees for more reliable results.
Gradient descent—Optimization method that improves a model by repeatedly moving downhill on error.
Microeconomics—Study of incentives, decisions, and markets at the level of individual actors.
Examples, Proofs & Derivations
Works & Sources Cited4
Paper
The importance of intrinsic disorder for protein phosphorylation—Published in 2004, cited for studying intrinsic disorder and PTM data.
Primary Document
Department of Health and Human Services, 2014—Cited for average cost of clinical trials data.
Paper
Universal log-optimality for general classes of e-processes and sequential hypothesis tests—Waudby-Smith, Sandoval and Jordan (2025)
Primary Document
GDPR—European privacy law governing data collection and use.
About the lecturer
Pehong Chen Distinguished Professor of EECS and Statistics
Distinguished Professor of EECS and Statistics at UC Berkeley whose work bridged probabilistic graphical models, variational inference, and nonparametric statistics.
1 lecture in archiveView profile
Discussion
Share a seminar note, question, or timestamped reference (e.g. 14:20)...Sign in