Offered at the Dept. of Electrical and Computer Engineering, McGill University
Upcoming Offering: Winter 2023
Course outline: Course Outline
Instructor: AJung Moon
Credits: 3 credits
Activity schedule (per week, for 13 weeks): 2x1.5 hour lectures, 1 hour tutorial, 5 hours of self-study
Prerequisites: (ECSE 202 or COMP 250 (Introduction to Computer Science)) and (ECSE 205 (Probability and Statistics for Engineers) or MATH 323 (Probability)) or instructor approval
Proposed Co-requisite: COMP 451 (Fundamentals of Machine Learning) or COMP 551 (Applied Machine Learning) or ECSE 551 (Machine Learning for Engineers).
NOTE: Co-requisite means you need to have taken these courses already or will be taking the course(s) alongside ECSE557. If you are a non-McGill student looking to take the course through inter-university credit transfers etc., or are graduate students with similar course credits from other institutions, please send me an email about it.
Recommended but not required: COMP 424 (Artificial Intelligence), ECSE 526 (Artificial Intelligence)
Past offering: Winter 2022 (Course outline),
This course is intended for engineering undergraduate students who are interested in learning about the social and ethical implications of intelligent systems. It is also intended to provide a richer selection of courses for students pursuing a minor in AI or interested in pursuing a career in AI.
There are no assigned textbooks for the course. However, students wanting to have more reference material to follow alongside the lectures/tutorials, you'll find the following books helpful (all of these are open access or avail. as e-books via McGill library):
Chapters from Spiekermann, S. (2016), Ethical IT Innovation: A Value-Based System Design Approach
Chapters from Barocas S., Moritz H., Narayanan A. (2019) Fairness and Machine Learning
Chapters from Nielsen A. (2020), Practical Fairness: Achieving Fair and Secure Data Models
The following is a rough list of topics I attempt to cover. But every semester is a little different. It will likely change as we go, especially as students bring their own interests to the table. For now, the most accurate plans for the topics to be covered can be found on the course outline listed above.
Lecture 1: First class, A brief review of ethics
Intro to the course. How is the course different from professional ethics, engineering codes of conduct, research ethics, and other ethics discussions? Why is it important? A quick review on the three (western) schools of thought in ethics; technology ethics and where does moral psychology fit in?
Lecture 2: A brief introduction to AI
What are autonomous intelligent systems, and what’s the difference between AI, ML, robotics?
Tutorial 1: Worked examples on ethics theories
Lecture 1: Values I & Human Rights
What are values? What are normative values? What are principles/standards? The Montreal AI Declaration. What are the human rights and what’s in the UDHR?
Lecture 2: Values II
Conceptualizing key values: responsibility and accountability; fairness, equality, bias & diversity; truth, trustworthiness, & trust; transparency, explainability, interpretability; automation & autonomy; privacy and human rights; Identifying and prioritizing values.
Tutorial 2: Worked examples on value identification and prioritization
Lecture 1: Ethically sensitive data & informed consent
Personally identifiable data; what makes data sensitive? What data can/can’t you collect? What are protected data? Components of informed consent;
Components of data quality; secondary use of data; data mining; case studies
Lecture 2: Anonymity & privacy
Anonymity and pseudonymity; k-anonymity; l-diversity principle; privacy guarantees
Tutorial 3: Worked examples on k-anonymity, and l-diversity principles
Lecture 1: Algorithms for sharing and recommending
Knowledge & knowledge sharing (social media); truth and objectivity; recommender engines and filter bubbles; online advertisement; case study on Microsoft Tay
Lecture 2: Midterm 1 – Part I: Fundamentals
Midterm exam during class time
Tutorial 4: Midterm review
Lecture 1: Bias and Fairness I
Machine bias; fairness as a concept, conditions for fairness perception; fairness definitions and metrics; transparency and recourse; case study
Lecture 2: Bias and Fairness II
Case study
Tutorial 5: Fairness evaluation worked example
Lecture 1: Privacy as a concept
What is privacy? Who has privacy? What are privacy harms and what do we do with them? Privacy laws in Quebec and Canada, and GDPR; surveillance technologies; facial recognition; anonymity in public spaces; encryption and politics;
Lecture 2: Considering privacy in design
Using privacy assessment tools; privacy by design; concepts behind privacy-preserving techniques (differential privacy, federated learning)
Tutorial 6: Worked case studies on privacy
Lecture 1: Automation bias and shared control
Automation bias, human factors, shared control, accountability gap. Case studies
Lecture 2: User preference and manipulation
Types and components of dark patterns, positive vs. negative liberty; behaviourism in technology; nudging;
Tutorial 7: Dark patterns worked example
Lecture 1, 2: Ethics analysis processes and ethics in design practices
Stakeholder identification; datasheets for datasets; design methods; AI ethics principles; toolkits; ethically aligned design. Case study.
Tutorial 8: Ethics analysis worked example I
Lecture 1: Robots vs. algorithms
Introduction to service and consumer robotics; how the physical robot complicates ethics issues
Lecture 2: Human-robot interaction
Anthropomorphism; Uncanny Valley theory; racializing and gendering of robots; trust and manipulation in HRI; owners vs. clients; individual vs. public; case study
Tutorial 9: Ethics analysis worked example II
Lecture 1, 2: Issues in killer robots, autonomous cars, care robots
What do the real lethal autonomous weapons systems (LAWS) look like? International humanitarian law and the LAWS debate; rules of engagement, laws of war; Levels of automation; ethics dilemmas; design assumptions; Robots for eldercare and childcare;
Tutorial 10: Project support session
Lecture 1: Artificial moral systems, top-down AMA
What is an AMA? Types of moral systems; science fiction vs. reality; top-down, bottom-up, and hybrid systems; Design of a top-down AMA; why the Asimov’s Three Laws of Robotics don’t work
Lecture 2: Bottom-up & hybrid AMA
Utilitarian calculus and machine learning; learning from user preferences and behaviours; virtue ethics in AMA
Tutorial 11: Project support session
Lecture 1: Automation, economics, and automated inequality
Discussions on what economists say about the role automation plays in society; Do AIS lead to more inequality in our society? Case study-driven discussion
Lecture 2: Power and technology activism
Case study-driven discussion
Tutorial 12: Project support session
Lecture 1: Project presentations I
Lecture 2: Project presentations II & course debrief
Tutorial: none