Course Home Page for Ethical Algorithm Design (CIS 4230/5230) University of Pennsylvania, Spring 2025 (original) (raw)
Ethical Algorithm Design CIS 4230/5230
Spring 2025
Tuesdays and Thursdays 10:15 11:45AM ET
Annenberg 110
Instructor:
Prof. Michael Kearns
mkearns@cis.upenn.edu
Teaching Assistants:
Elliu Huang
elliuh@seas.upenn.edu
Alexandra Oh
alexoh@seas.upenn.edu
Simon Roling (head TA)
rolings@seas.upenn.edu
Arnab Sircar
asircar@seas.upenn.edu
Here is a list ofoffice hoursfor all course personnel. You may also request office hours by appointment.
Course Description
This course is about the social and human problems that can arise from algorithms, AI and machine learning, and how we might design these technologies to be "better behaved" in the first place. It is first and foremost a science or engineering course, since we will be developing algorithm design principles. You can get a rough sense of course themes and topics by visiting the websites for the pilot versions of this course offered in2021, 2020and2019.The first formal offering of the course was inSpring 2022,and you can also visit the sites forSpring 2023andSpring 2024.
Here are the lecture videosfrom the last pilot version in 2021. Please note that they will not correspond exactly to this year's lectures (and we will cover material not in the videos at all), and should not be viewed as a substitute for mandatory lecture attendance, but rather as a study aid.
Prerequisites: Familiarity with some machine learning, basic statistics and probability theory will be helpful. While this is not a theory class, you need to be comfortable with mathematical notation and formalism. There will be some simple coding and data analysis assignments, so some basic programming ability is needed.
Course content will include readings from the scientifc literature, the mainstream media and other articles and books.
Grades will be based on a mixture of quizzes, coding assignments, written homeworks, and a written midterm and final.
CIS 423/523 fulfills theSEAS Engineering Ethics Requirementfor these programs: ASCS, BE, CMPE, CSCI, DMD and NETS (but you should confirm with your academic adivsor to be certain).
| Lecture Dates | Topic | Lecture Notes | Readings, Assignments, and Announcements |
|---|---|---|---|
| Thu Jan 16 | Course Introduction and Overview | Lecture Notes | While they look ahead to material later in the semester, the following two (required) general-audience articles on the science of Responsible AI are a good preview of the spirit of the class, please read them in the first week of class or so: Responsible AI in the generative era,M. Kearns, Amazon Science blog, May 2023. Responsible AI in the wild: lessons learned at AWS,M. Kearns and A. Roth, Amazon Science blog, November 2023.A general-audience introduction to some of the themes of the course is given in the (recommended but not required) bookThe Ethical Algorithm: The Science of Socially Aware Algorithm Design,by M. Kearns and A. Roth.Also recommended but not required:Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy,by C. O'Neil. |
| Tue Jan 21Thu Jan 33 | Foundations of Machine Learning | Lecture Notes | . |
| Tue Jan 28Thu Jan 30Tue Feb 4 | Bias in Machine Learning: COMPAS and ProPublica | Lecture Notes | The following readings are required: ProPublica article on COMPAS ProPublica analysis Practitioner's Guide to COMPAS Core(no need to read, but we'll peruse a bit together in lecture) COMPAS Risk Assesment Survey(just skim) Northpointe response to ProPublica ProPublica github repository, including dataset(we'll look at the dataset a bit in lecture)(technical, just skim) |
| Tue Feb 4Thu Feb 6Tue Feb 11Thu Feb 13Tue Feb 18Thu Feb 20Tue Feb 25Thu Feb 27Tue Mar 4 | Science of Fair ML: Models and Algorithms | Lecture Notes Slides from a related talk | Readings: Inherent Trade-Offs in the Fair Determination of Risk Scores.J. Kleinberg, S. Mullainathan, M. Raghavan. (First 8 pages required) The Frontiers of Fairness in Machine Learning.Alexandra Chouldechova, Aaron Roth. (Required)Please play around with the followingGoogle demo site on fairness and ML.(Required) Equality of Opportunity in Supervised Learning.M. Hardt, E. Price, N. Srebro. (Intro required, rest optional; this is the post-processing/bolt-on method) A Reductions Approach to Fair Classification.A. Agarwal, A. Beygelzimer, M. Dudik, J. Langford, H. Wallach. (Intro required, rest optional; this is the in-processing/constrained optimization/game theory method) An Empirical Study of Rich Subgroup Fairness for Machine Learning.MK, S. Neel, A. Roth, S. Wu. (Intro required, rest optional; this the rich subgroup/preventing fairness gerrymandering method) An Algorithmic Framework for Bias Bounties.I. Globus-Harris, MK, A. Roth. (Read at least the Intro and Sections 5 and 6) Diversified Ensembling: An Experiment in Crowdsourced Machine Learning.I. Globus-Harris, D. Harrison, MK, P. Perona, A. Roth. (Skim, will discuss in class) |
| Thu Mar 6 | MIDTERM EXAM | 2023 Midterm | The midterm will be closed notes and closed book. |
| Tue Mar 18Thu Mar 20Tue Mar 25Thu Mar 27Thu Apr 3Tue Apr 8Thu Apr 10 | Security, Privacy and Related Topics | Lecture Notes | Readings: Broken Promises of Privacy: Responding to the Surprising Failure of Anonymization.P. Ohm. (not assigned, but for your perusal) Differentially Private Query Release Through Adaptive Projection.S. Aydore, W. Brown, M. Kearns, K. Kenthapadi, L. Melis, A. Roth, A. Siva. (not assigned, but will be discussed in lecture) Confidence-ranked reconstruction of census microdata from published statistics.T. Dick, C. Dwork, MK, T. Liu, A. Roth, G. Vetri, S. Wu. (not assigned, but will be discussed in lecture) Differential Privacy Overview.Apple. Skim for discussion in lecture. How One of Apple's Key Privacy Safeguards Falls Short.Wired magazine. Skim for discussion in lecture. Privacy Loss in Apple's Implementation of Differential Privacy on MacOS 10.12J. Tang, A. Korolova, X. Bai, X. Wang, X. Wang. Skim for discussion in lecture. Implementing Differential Privacy: Seven Lessons From the 2020 United States Census.M. Hawes. Skim for discussion in lecture. See how your community is moving around differently due to COVID-19.Google Covid Mobility Reports. Skim for discussion in lecture. |
| Tue April 15Thu April 17Tue April 22Thu April 24 | Ethical Algorithm Design in the Generative Era | Lecture Notes | Readings: Hallucination, Monofacts and Miscalibration: An Empirical Investigation.MK, M. Miao. (skim for discussion in lecture) AI model disgorgment: Methods and choices.A. Achille, MK, C. Klingenberg, S. Soatto. (required) |
| Tue April 29 | Review session for final exam | . | . |
| Fri May 9 noon - 2PM | FINAL EXAM | Location:CHEM 102 | The final exam will be cumulative, with perhaps a slight emphasis on later material. It will be closed book, closed notes, no devices of any kind. To help you study, here are the final exams fromSpring 2023andSpring 2024.Please remember that course material changes a bit from year to year, so some questions on these exams may not be relevant for your exam. |