Algorithmic decision making for social good

Introduction

AI and algorithmic decision-making systems are increasingly built and deployed in practice, with many purporting to “do good”. In this course, we’ll investigate what “good” looks like, what the right problems are to work on, and discuss considerations for the effective design and deployment of algorithmic decision-making systems. We'll look at both avoiding unintended harms as well as moving towards desirable, socially beneficial outcomes. Along the way, we’ll study algorithms across machine learning, optimization, market design, and reinforcement learning applied to societal problems spanning sustainability, education, healthcare, and government operations.

Course details

Grading

• In-class paper presentations (30%)
• Project (45%)
• Participation (20%)
• Paper responses (5%)

There are no formal prerequisites, but this course expects mathematical maturity and ability to engage with state-of-the-art research from operations research, computer science, and other disciplines.

Schedule

Please note that this schedule is subject to change.

Resources

Relevant organizations

• EAAMO (Equity and Access in Algorithms, Mechanisms, and Optimization)

Additional readings

• Channing & Ghosh. preprint. AI for Scientific Discovery is a Social Problem
• Guha et al. NeurIPS 2023. LegalBench: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Models
• Kleinberg and Raghavan. PNAS 2021. Algorithmic monoculture and social welfare
• Lakkaraju, Kleinberg, Leskovec, Ludwig, Mullainathan. KDD 2017. The Selective Labels Problem: Evaluating Algorithmic Predictions in the Presence of Unobservables
• Perdermo, Zrnic, Mendler-Dünner, Hardt. ICML 2020. Performative Prediction
• Wang, Morgenstern, Dickerson. Nature Machine Intelligence 2025. Large language models that replace human participants can harmfully misportray and flatten identity groups