Recommender System Objectives
Level:
Third-Year Machine Learning
Class Time:
One 2-hour class
Last Modified:
Wed 06 April 2022
This module gives machine learning students some technical and ethical concepts that are relevant to the design of recommender systems. Students first discuss the technical advantages and disadvantages of different options for what recommender systems may optimize, including clicks, screen time, and other measures of engagement. Next, they evaluate several hypothetical recommender systems in order to identify the ethical problems that may arise when a recommender system optimizes for these qualities. Through a series of polls, they then consider whether their evaluations of those recommender systems can be formulated in terms of the concepts of harm to others or manipulation. The module concludes with a reflection exercise that asks students to critically evaluate the impact of one of three options (controls, payments for user data, and "well-being" metrics) on the ethics of recommender systems.
This module was developed by Steven Coyne and Roger Grosse, with contributions from Emma McClure. Diane Horton, David Liu, and Sheila McIlraith provided feedback on this module.
Materials
In-class Material
- Lesson Plan
- Part 1: Computer Science Slides (pdf, tex source)
- Part 2: Ethics Slides
Homework