This post marks the first in a series that I hope to write this year on redesigning my quantitative methods course for a world where generative AI is ubiquitous, incredibly powerful, and potentially dangerous. While there is plenty of thoughtful prose that has been written about generative AI and learning, I find that there is not enough about the nitty-gritty choices, tradeoffs, and compromises that all teachers are making, or will have to make, in order to fulfill our mission. This may sound negative, but I’m actually quite excited - these are the kinds of decisions that are central to the act of teaching! My hope is that documenting this process will be useful to others and help generate bigger-picture insights that contribute to our shared work.

Who I am

For those who don’t know me, I’m a faculty member at the Harvard Kennedy School. I teach courses in statistics for graduate students and online mid-career students, and I teach courses on generative AI. I think a lot about technology and teaching and make lots of online materials to facilitate learning. Most recently I’ve spent time writing and talking about generative AI and learning, and creating tools to help incorporate generative AI into my and others’ teaching.

If you want to know more, check out my website. If you don’t have time for that, use this chatbot I built (with help from Claude Code in building an MCP server) that can search and synthesize content from my website.

My priors

Rather than wax philosophical about my stance toward AI, how I think it’s transforming the world, the ways in which I think it is likely to transform education, and the ways in which I think it should transform education, I’m going to lay out the set of beliefs that inform my thinking here:

1. Generative AI gives us the unprecedented ability to meet every student where they are and help them build skills that they value, something that I have strived to create throughout my teaching career.
2. Generative AI tools change the set and relative importance of skills we teach in our courses. Some skills become more important, others less important or irrelevant, and new skills emerge.
3. Commercial generative AI offerings are currently structured in a way that is at odds with how lasting learning happens. In short, they are trained and tuned to help you accomplish things (which relates to #2), not to help you learn how to accomplish them. And yes, I include each and every one of the various “study modes” in this bucket. That healthy friction that makes learning happen is too easy to turn off with these modes.
4. Generative AI is advancing at a fast enough clip that “updating a course for AI” will have to happen on an ongoing, adaptive basis. I can remember feeling pretty self-satisfied when I designed my first assignment asking students to “explain why this LLM’s answer is wrong”; this week Claude Code completed the entire multi-week final assignment of my course, including a statistical analysis, memo, technical appendix, and PowerPoint presentation, in 10 minutes.

My course

The course that I’m working on transforming is API-201: Quantitative Analysis and Empirical Methods, a required introductory statistics course for first-year Master’s in Public Policy students. Here’s a syllabus with some information redacted to give you a sense of its goals and purpose. Some notes about it that I think are salient with respect to this effort:

• It’s a required course. There are plenty of students who begin the course frustrated that they have to take it and wanting to spend only the required amount of time on it and nothing more. I believe many of them change their minds (or should I say…) by the end, but that takes time.
• It’s a well received course. As far as required courses go, students leave pretty happy! This means that (a) I want to be sure to preserve the core components that make it a success, and (b) I have some leeway with students to experiment with new approaches and still maintain goodwill.
• While it is an introductory statistics course, it differs quite a bit from what you might see in a statistics department. The core goals are to instill in students a love of data, provide them with the tools to scrutinize analyses effectively, and make them able to immediately use these concepts in the real world. That means, for example, that we spend less time on different statistical tests and more time on the implications and limitations of statistical significance.
• Students will have taken anywhere from zero to six statistics courses before – it’s a very heterogeneous group!
• The section I teach uses Excel for analysis; my colleague Jonathan Borck teaches another Excel section. There is a different pair of sections that uses R, which tends to spend more time on coding and less time on the conceptual parts of the course.
• Generative AI is already incorporated to a significant extent. We have course-specific tutor bots, and exercises that involve students using generative AI to reproduce analyses. More on that later.

What’s next

This was a lot of background, but hopefully provides a sufficient overview for future posts. Up next I’ll outline what aspects of generative AI I’ve incorporated already and how I’m thinking about the redesign. Hope you’ll join me!