Title: Algorithmic Fairness

Time & Date: 3-5pm, April 5, 2022


Outline of Activities:

  • Pre-module survey: 1% marks for completing, due by 3pm ET on April 5
  • In-class module: not graded, to be conducted in-class 3-5pm on April 5
  • Post-module survey: 1% marks for completing, due by 11:59pm ET on April 15
  • Graded essay: 3% weight, due by 11:59pm ET on April 15
  • Note: The above module activities carry a total of 5% marks, which will be part of the 10% marks for course participation announced at the start of the course.

Readings: It would be helpful to read through (or at least glance through) the following papers before the in-class module.

  • A survey on fairness in machine learning and algorithmic decision-making (link)
  • A book chapter on participatory budgeting, which is the topic of one of the in-class module activities (link)

Module Overview:

In this module, we consider fairness in algorithmic decision-making. The module begins with an overview of various mathematical approaches to fairness used in computer science, many inspired from ideas on economic theory.

This is followed by a breakout activity, in which we consider various real-world contexts and try to reason about which factors should or shouldn't matter for fairness. For example, when fairly distributing COVID vaccines to various countries, should the size of their population matter? How about their age distributions? How about their GDP? After the breakout activity, we synthesize the responses and reflect on them.

Next, we focus on the real-world problem of participatory budgeting, in which public funds must be fairly allocated to public projects in a city based on the preferences of its residents. We have another breakout activity, in which we look at sample participatory budgeting elections, and reason about which outcomes would be fair. After the breakout activity, we again synthesize the responses and reflect on them.

Before concluding, we consider the hotly debated issue of whether fair decision-making is best left in the hands of algorithms or in the hands of humans. We evaluate arguments for both positions and think about what different stakeholders would want in a given situation.


Goals:

  1. Introduce students to the topic of algorithmic fairness.
  2. Explain different mathematical approaches to fairness.
  3. Equip students with tools to reason about fairness in algorithmic decision-making.
  4. Practice thinking and communicating about what fairness requires in a given real-world context.
  5. Practice thinking about a situation from the perspectives of different stakeholders.
Agenda:

  • 1:10-1:15: Introduction
  • 1:15-1:30: Overview of various approaches to algorithmic fairness (provided by the instructor)
  • 1:30-1:45: Breakout activity 1: Which factors should or shouldn't matter in fair algorithmic decision-making?
  • 1:45-1:55: Synthesis of the outcome of breakout activity 1
  • 1:55-2:05: Break
  • 2:05-2:20: Introduction to an application: participatory budgeting
  • 2:20-2:35: Breakout activity 2: Which outcomes do you find fair in example participatory budgeting scenarios?
  • 2:35-2:45: Synthesis of breakout activity 2
  • 2:45-2:55: Hot topic: Algorithms versus humans
  • 2:55-3:00: Conclusion