6.883 Data-Driven Decision Making and Society — Spring 2021Instructors: Aleksander Mądry, Asuman Ozdaglar and Shibani Santurkar
TA: Kai Xiao
Teaching staff email: email@example.com
Time and place: MW 2:30-4 pm on Zoom (you will need an MIT Zoom account to access this meeting—email us if you don't have such account)
Units: 3-0-9 AAGS
The last decade brought us tremendous advances in the power and sophistication of the data-driven decision-making techniques that are at our disposal. Encouraged by this progress, we are witnessing a broad deployment of these techniques in the real world. They now touch on—and sometime even govern—just about every aspect of our lives.
However, as much as these techniques were deployed with the promise of bringing a decisively positive change, it has become abundantly clear that they often are a mixed blessing, at best. Indeed, it turns out that the interface of algorithmic decision-making and society is rife with subtle and non-obvious interactions, undesirable feedback loops, and unintended consequences.
How should we make sense of and navigate these issues?
The goal of this class is to survey some of the key challenges emerging in the context of societal impact of data-driven decision making as well as to create a forum where the students can discuss potential approaches to addressing these challenges.
This class will be a mix of background lectures (given by the instructors and guest lecturers) and student-led discussion sessions. These discussions will involve the leading student team selecting (in consultation with the instruction staff) 1-2 papers to constitute the core discussion material for that session. They will also be in charge of preparing a set of discussion-seeding questions, preparing a short (5-10 min) framing presentation to kick off the session as well as preparing a brief (1-2 pages) report summarizing the discussion afterwards.
All the students would be expected to read the papers to be discussed (or their specified parts) and submit a brief, paragraph-long answer to a (pre-defined) question ahead of time. Finally, each student will be asked to write a short (3-4 pages) essay on one of the discussed topics or a theme that cuts across a number of topics.
Grade will comprise leading the class discussion (as a part of the team) [50%], involvement in class discussion (as a participant) [20%], and final-project essay [30%].
- If you plan to attend this course (either for credit or as a listener), please fill out this form.
- (Wednesday, 2/17) Introduction and Course Overview; Machine learning (non-)robustness.
Related readings: Introduction to adversarial robustness, Survey of data poisoning methods.
- (Monday, 2/22) ML interpretability and human-ML (mis)alignment.
Related readings: Towards A Rigorous Science of Interpretable Machine Learning, The Mythos of Model Interpretability, Adversarial Examples Are Not Bugs, They Are Features, Robustness Beyond Security: Representation Learning.
- (Wednesday, 2/23) Datasets.
Related readings: Big Data's Disparate Impact, Bias in Computer Systems, The Hidden Biases in Big Data, "Raw Data" Is an Oxymoron.
- (Monday, 3/1) Causality.
Related readings: Chapter 4 of the "Fairness and machine learning" book, Tutorial on Causal Inference and Counterfactual Reasoning.
- (Wednesday, 3/3) ML fairness.
Related readings: "Fairness and machine learning" book, "The Trouble with Bias" talk, Big Data's Disparate Impact.
- (Tuesday, 3/9) Sequential decision making.
Related readings: Introduction to Multi-Armed Bandits, Introduction to Online Convex Optimization, Fairness Is Not Static: Deeper Understanding of Long Term Fairness via Simulation Studies.
- (Wednesday, 3/10) Game theory.
- (Monday, 3/15) Human-technology interactions.
- (Wednesday, 3/17) Discussion: Data Provenance.
Related readings: Large Datasets: A pyrrhic win for computer vision?, Designing Ground Truth and the Social Life of Labels.
- (Monday, 3/29) Discussion: Facial Recognition and Bias.
Related readings: Gender Shades, One Month, 500,000 Face Scans: How China Is Using A.I. to Profile a Minority, Do These A.I.-Created Fake People Look Real to You?.
- (Wednesday, 3/31) Discussion: Data-aggregation and Privacy.
Related readings: Aspects of Data Ethics in a Changing World: Where Are We Now?, It’s Not Privacy, and It’s Not Fair.
- (Monday, 4/5) Discussion: Bias in Natural Language Processing.
Related readings: Word embeddings quantify 100 years of gender and ethnic stereotypes, PowerTransformer: Unsupervised Controllable Revision for Biased Language Correction.
- (Wednesday, 4/7) Discussion: Unintended Consequences of Social Media.
Related readings: Social Media, News Consumption, and Polarization: Evidence from a Field Experiment (only read Introduction and Section 5, pages 1-4 and 22-26), The truth behind filter bubbles: bursting some myths, The dashboard SplitScreen.
- (Monday, 4/12) Discussion: ML and Criminal Justice.
Related readings: Criminal justice, artificial intelligence systems, and human rights, Machine Bias.
- (Wednesday, 4/14) Discussion: Long-term Impact of Algorithmic Fairness.
Related readings: Attacking discrimination with smarter machine learning, On the Long-term Impact of Algorithmic Decision Policies: Effort Unfairness and Feature Segregation through Social Learning.
- (Monday, 4/21) Discussion: Strategic Interaction Between Humans and Platforms Powered by ML.
Related readings: Peer-to-Peer Markets, Surge Pricing Solves the Wild Goose Chase (only read Abstract and Sections 1, 6, 8), Tax-Induced Inequalities in the Sharing Economy (only read Abstract and Sections 1, 6, 7)
- (Monday, 4/26) Discussion: Law and Big Data.
Related readings: Big Data’s Disparate Impact (only read Sections II.A and II.B), Watch this 40 minute talk by Prof. Sandra Wachter, (optional) Can A.I. Be Taught to Explain Itself?.
- (Monday, 4/28) Discussion: Evaluating AI Rewards Systems and AI Alignment.
Related readings: Concrete Problems in AI Safety, Commentary: Facebook's Algorithm vs. Democracy.
- (Monday, 5/3) Discussion: Machine Learning and Health Care.
Related readings: Ethical Machine Learning in Health Care, Using AI ethically to tackle COVID-19.
- (Wednesday, 5/5) Guest lecture: Daron Acemoglu.
Related readings: AI’s Future Doesn’t Have to Be Dystopian, Automation and New Tasks: How Technology Displaces and Reinstates Labor.
- (Monday, 5/10) Guest lecture: Adam Berinsky and David Rand.
Related readings: This is how you stop fake news, The Psychology of Fake News.
- (Monday, 5/17) Guest lecture: Karen Levy.
Related readings: The Dark Side of Numbers: The Role of Population Data Systems in Human Rights Abuses.
- (Wednesday, 5/19) Guest lecture: Abby Everett Jaques.