Martin Wattenberg

Also known as MARTIN P. WATTENBERG, Martin M. Wattenberg

Martin Wattenberg is a researcher and artist whose visualization and interpretability work helps people see how machine learning systems operate. After earning a mathematics doctorate at Berkeley in 1996, he made interactive journalism at SmartMoney.com, founded IBM Research's Visual Communication Lab in 2004, and joined Google in 2010, where he and Fernanda Viégas announced the People + AI Research initiative (PAIR) in 2017. From 2021 to 2026 he was a tenured professor at Harvard, and in 2023 he coauthored two interpretability papers: one on a board-state representation inside a GPT model trained on Othello moves, and one on Inference-Time Intervention for truthful answers. In December 2025, he, Viégas and Jess Weaver published Inside the Black Box, which introduced a dashboard that shows the assumptions a chatbot forms about its user. He is at Google DeepMind, where he co-leads a research team with Viégas on human-centered AI.

Now at: Google DeepMind
Born: 1970
Badges: Founder
Bridge

Years link to their sources.

Timeline

1991–1996

Mathematics Ph.D. at Berkeley

He earns an A.B. from Brown University in 1991 and an M.S. from Stanford University in 1992, then a Ph.D. in mathematics at U.C. Berkeley under Charles C. Pugh in 1996.
1996–2002

The Map of the Market

He works for Dow Jones on the personal finance and investing site SmartMoney.com from 1996 through 2002. His Map of the Market, created in 1998, is the first web-based treemap.
2002–2007

Many Eyes at IBM Research

He takes a position at IBM's Thomas J. Watson Research Center in Cambridge, Massachusetts, in 2002 and founds IBM Research's Visual Communication Lab in 2004. With Fernanda Viégas, he helps found Many Eyes, an IBM web site created in 2007 that makes visualization technology accessible to the public.
2010–2017

Big Picture and PAIR at Google

He and Fernanda Viégas start the visualization studio Flowing Media, Inc. in April 2010 and join Google as co-leaders of its Big Picture data visualization group in Cambridge, Massachusetts. In July 2017, the two announce the People + AI Research initiative (PAIR), which studies how people interact with AI systems and releases the open-source tools Facets Overview and Facets Dive.
2021–2026

Professor at Harvard

He joins Harvard in fall 2021 as a Gordon McKay Professor of Computer Science in the John A. Paulson School of Engineering and Applied Science. A tenured professor there until 2026, he co-leads the Insight + Interaction Lab with Fernanda Viégas.
2023

Othello-GPT and Inference-Time Intervention

With Fernanda Viégas, Hanspeter Pfister and Kenneth Li, he coauthors two language-model interpretability papers. Emergent World Representations, an ICLR 2023 oral, trains a GPT variant to predict legal Othello moves and finds an emergent nonlinear internal representation of the board state. Inference-Time Intervention, a NeurIPS 2023 spotlight, raises the truthfulness of the instruction-finetuned Alpaca model from 32.5% to 65.1%.
December 2025

An interpretability dashboard for chatbots

Inside the Black Box, by him, Fernanda Viégas and Jess Weaver, appears at the Berkman Klein Center in December 2025. It introduces an interpretability dashboard, developed by researchers at the Berkman Klein Center and Harvard's Insight and Interaction Lab, that reveals how chatbots infer characteristics such as age, gender, race and socioeconomic status.
As of October 2026

A research team at Google DeepMind

He is at Google DeepMind, working to help highly capable AI systems be useful and beneficial. With Fernanda Viégas, he co-leads a research team that takes a participatory, human-centered approach to AI.