A collection of AI Ethics education material developed specifically for scientists (typically taught to PhD students in scientific disciplines, but appropriate for general scientific audiences). This material has been used to provide AI Ethics instruction in a variety of contexts including the SLAC Summer Institute, The Argonne Training Program on Extreme-Scale Computing, The Large Synoptic Survey Telescope Corporation Data Science Fellowship Program and the SMART-HEP Network. The repository is divided into three sections.
This is 6-hour mini-course on AI Ethics and Responsible Data Science and includes 3 lectures and 2 guided, hands-on coding examples.
- Lecture 1: Overview
- a taxonomy of AI Ethics considerations (data collection and storage, task design and learning incentives, model bias and fairness, model robustness, system deployment and outcomes, and downstrem/diffuse impacs) including real world examples of each consideration.
- Why scientists should be engaging with AI Ethics
- Scientific frameworks for evaluating ML/AI models
- Discussion activities to practice evaluating real systems
- Hands-on 1: Exploring sources of unfairness and mistakes in ML models: a case study on COMPAS
- Lecture 2: Explainable AI
- Local explanation methods
- Global explanation methods
- Real examples applying examplainability methods to physics models
- Limitations of AI explainability/interpretability
- Transparency, documentation, and model auditing
- Hands-on 2: Implementing explainability methods SHAP and LIME
- Lecture 3: Other Topics in AI Ethics
- Model monitoring
- Quantitative fairness
- Privacy
- Regulation
- Participatory design
A condensed version of the course appropriate for an hour-long seminar or lecture.
- A Taxonomy of AI Ethics
- Why scientists should be engaging with AI Ethics
- Scientific frameworks for evaluating ML/AI models
Link to recording | Link to slides
The initial material for a semester long course on AI Ethics currently being taught to Data Science Masters students. Links to the currently available material are provided below and will be updated as the course progresses.
- Course Syllabus
- Lecture 1: Introduction to AI Ethics and Responsible Data Science
- Lecture 2: Algorithmic Bias
- Lecture 3: Building Trustworthy Models
- Lecture 4: Explainable/Interpretable AI
- Lecture 5: Generative AI and LLMs
- Lecture 6: Governance and Regulation of AI
- Lecture 7: Current Applications of AI
- Lecture 8: Real and Imagined Futures
- Lecture 9: The Role of Technology in Shaping Society
Please feel free to use these materials for your own learning or teaching (with appropriate credit attribution!). I am also very happy to consider guest lectures or similar endeavors, so please don't hesitate to reach out. If you have any suggestions please feel free to open a pull request or contact me at st3565 at columbia.edu