Dr Joshua Loftus

This course is available on the BSc in Actuarial Science, BSc in Data Science, BSc in Mathematics and Economics, BSc in Mathematics with Data Science, BSc in Mathematics with Economics, BSc in Mathematics, Statistics and Business and BSc in Politics and Data Science. This course is available with permission as an outside option to students on other programmes where regulations permit and to General Course students.

Students must have completed Elementary Statistical Theory (ST102) or equivalent, Mathematical Methods (MA100) or equivalent, and at least one of MA212, EC220, EC221, ST206, ST202, or equivalent.

Familiarity with basic computer programming in R or Python. Students who have no previous experience in R are strongly encouraged to take on an online pre-sessional R course from the Digital Skill Lab (https://moodle.lse.ac.uk/course/view.php?id=7022)

This course covers a selection of topics central to the ethical practice of data science. Students will learn key concepts and methods to analyze a variety of case studies, from the historical and philosophical background of data technologies and ethics to the frontiers of research in machine learning, artificial intelligence, and socio-technical systems. These concepts will include some basic philosophical and legal ideas related to data ethics, frameworks for ethical practice developed by professional societies, formal statistical definitions and quantitative methods for objectives such as fairness and privacy, and an emphasis on the use of causal reasoning to evaluate data-driven systems and policies. Topics may include:

• Replication crisis, unfair algorithms, basics of normative ethics and causality
• Historical examples, professional ethical guidelines
• Transparency, reproducibility, open science
• Discrimination, statistical fairness, impossibility results
• Causal reasoning for fairness, pathway analysis, intersectionality
• Interventions, policy optimization, distributive justice
• Data provenance, privacy, differential privacy
• Strategic behavior, surveillance, democratic data
• Automation and AI, responsibility, complicity

Causal statistical models will be used as a formal framework throughout to understand and stress test these ideas.

10 hours of lectures and 20 hours of classes in the AT.

Students will be expected to produce 3 problem sets in the AT.

The first two problem sets will be formative, and feedback from these will help students prepare for the third, summative problem set.

Lecture notes will be provided. These will be supplemented with a variety of short readings, some of which will be taken from the following background references

• https://www.bitbybitbook.com/en/1st-ed/ethics/
• https://fairmlbook.org/
• https://data-feminism.mitpress.mit.edu/
• https://aiethics.princeton.edu/case-studies/
• https://www.acm.org/code-of-ethics
• https://rss.org.uk/RSS/media/News-and-publications/Publications/Reports%20and%20guides/A-Guide-for-Ethical-Data-Science-Final-Oct-2019.pdf
• https://www.amstat.org/ASA/Your-Career/Ethical-Guidelines-for-Statistical-Practice.aspx
• https://hastie.su.domains/CASI/
• https://www.statlearning.com/

Group project (50%) in the ST.
Group presentation (20%) in the AT Week 11.
Problem sets (30%) in the AT.

The final problem set during the AT will be summative and count for 30% marks. Group work consists of a presentation during the AT describing a project proposal, and the project itself will then be due in the ST.