Dan Liebling

I create AI systems for scientific research. Science AI should center the scientist, not just be yet another training or evaluation dataset. Currently I lead and engineer projects directly facing researchers in specialized domains. The goal is to show rigorous scientific impact rather than create a so-called "AI Scientist." Learn more about our team's Science AI research.

Previously, I lead a team that worked on on conversational speech recognition and machine translation at Google AI. Machine translation is generally seen as high-quality, yet our users continually report frustration in real-world settings. We developed models to improve Google Translate based on real human conversations, not just single utterances. Read more about our work in our CHI 2020 paper and other works.

You can view my academic contributions on Google Scholar. Follow me on BlueSky or Mastodon!

Previous work at Microsoft Research

I improved search (Bing) and other information work experiences (Office) by modeling user behavior and context. 

• Personal microtasking and AI-assited writing (with Jaime Teevan, Shamsi Iqbal, Kristina Toutanova, et al)

• Contextual search (with Susan Dumais, Paul N. Bennett, Ryen W. White, et al.)

• Eye tracking as input (with Shane Williams, Susan Dumais, et al.)

Civic Tech

For a while I was involved in the "civic tech" community. It tends to have a great mix of civically-minded people and tech people (which sometimes overlap). 

• Modeled pedestrian-car incidents

• #bikeraxx project to sense bike rack use on buses won honorable mention at Hack the Commute

• Court Whisperer project to enable mobile legal form filling won the Social Justice Hackathon

• Nimby Ninja was a low-barrier way for YIMBYs to refute NIMBY public comments.

Academic history

I studied Computation and Neural Systems at Caltech for my BS, and have a MS in Computer Science and Engineering from the University of Washington. In 2023, I was an adjunct lecturer in the Lingustics department there.