fast.ai co-founder, AI ethics and education
Rachel Thomas
Profile
Rachel Thomas co-founded fast.ai with Jeremy Howard in 2016, and that bet — that deep learning could be taught top-down to working developers instead of bottom-up to PhDs — shaped how a generation of engineers actually learned to build with neural networks. The Practical Deep Learning for Coders course has run for nearly a decade and has been taken by hundreds of thousands of students, many of whom went on to become professional ML engineers without ever sitting through a traditional graduate program. If you know anyone who “learned AI by doing fast.ai,” Thomas is half the reason that path exists.
She has a math PhD from Duke, was an early data scientist at Uber, and in 2019 founded the Center for Applied Data Ethics at the University of San Francisco. Her signature move there was building a free Applied Data Ethics course aimed at working engineers, not philosophers — concrete material on disinformation, bias, surveillance, and what she calls the “tyranny of metrics.” Her essay “The Problem with Metrics is a Fundamental Problem for AI” is probably the single piece of hers most worth reading: a sharp argument that optimizing any proxy metric eventually corrupts the thing the metric was meant to measure, and that this is not an edge case but the default behavior of systems like YouTube’s recommender or essay-grading AI.
Thomas is now Professor of Practice at Queensland University of Technology’s Centre for Data Science and has moved back into research at Answer.AI, the lab Howard launched in 2023. Her recent focus has shifted toward AI in medicine — the places where the hype meets actual patient harm, and where careless benchmarks lead to real bodies. She has also written extensively on the retention crisis for women in tech and the lack of diversity in AI research, pushing these as systems problems rather than HR problems.
For developers learning AI, Thomas matters because she made the field less gatekept and because her ethics work is pragmatic rather than abstract. She is not telling you to stop building; she is telling you that if you are going to build, you should understand what metrics you are optimizing and who pays when the system is wrong.
Books
Deep Learning for Coders with Fastai and PyTorch The book version of the fast.ai course, co-authored with Jeremy Howard — a top-down, code-first introduction to deep learning.Key Articles & Papers
The Problem with Metrics is a Fundamental Problem for AI Applied Data Ethics — a new free course, essential for all working in tech USF Launches New Center for Applied Data Ethics AI & Medicine: Promise & Peril Medicine's Machine Learning Problem What HBR Gets Wrong About Algorithms and BiasControversies
In 2020, Thomas publicly resigned from an advisory board at the Stanford Institute for Human-Centered AI (HAI) over concerns about the institute’s lack of diversity and its close ties to big tech funders. She argued that ethics-washing — using prominent ethics researchers as cover for business-as-usual — was an active harm, not just a missed opportunity. The critique was pointed but in character: she has been consistent that institutional structures and incentives matter more than individual good intentions.
Spotify Podcasts
