The Legible Machine
Author — 2023
A book on making intelligent systems understandable to the people they affect. Translated into six languages.
I work at the intersection of machine intelligence and human understanding. My research centres on interpretability — making the systems we build legible to the people who use them. I write essays on technology and attention, and I believe the most consequential engineering decisions are, at heart, decisions about values.
Selected Work
Circuits
Lead — 2021
Open tooling for the mechanistic interpretation of transformer circuits — now standard equipment in the field.
On Attention
Essayist
An ongoing series of essays on technology, attention, and the ethics of automation.
Experience
Lead Research Engineer
2020–PresentThe Aurora Institute
Lead the interpretability group. Published widely-cited work on the mechanistic interpretation of transformer circuits, and built the open tooling the field now uses to inspect them.
Visiting Researcher
2018–2020MIT Media Lab
Studied human–model interaction and how explanations change the way people trust automated systems.
Software Engineer
2015–2018Wolfram Research
Worked on symbolic computation and technical documentation systems.
Education
Ph.D., Computer Science
University of Oxford · Thesis on interpretable representations · 2014–2018
B.A., Mathematics
University of Cambridge · 2011–2014
Capabilities
Python · PyTorch · Interpretability · Research · Technical Writing · Mathematics · Causal Inference · Statistics
Recognition
- ICML Outstanding PaperICML, 2022
- Author, “The Legible Machine”2023
- Fellow, Royal Society of Arts2021
Beyond work
Poetry, Long-distance running, Classical piano