Lois Wong

I’ve had a longstanding obsession with silence—specifically, the idea of silence as a signifier and absence as a parsable signal. In Prince and Smolensky’s Optimality Theory, it is the non-words that reveal the underlying structure of a language. Debussy defines music as the silence between the notes; Cage shows how silence is compositional. Woolf cites unwritten books and Du Bois posits the “eloquent omissions and silences” in sorrow songs as articulations of pain and tragedy.

My work is one instantiation of this obsession. I'm interested in how archival silence and unrepresentative training data surface as algorithmic bias and model error, as well as building black-box evaluation methods to identify knowledge gaps and targeted remediation to close them. LMs learn from corpora that disproportionately preserve the profiles and perspectives of certain groups, e.g. those historically afforded an education and platform, while underrepresenting others due to factors like data quality, language, and uneven digitization. Those absences are not only historical; they are ongoing. People are still missing and being silenced, and developing methods to detect and surface that is ultimately where I want to take my work.

I have a BA in Linguistics from UC Berkeley and an MSE in Computer Science from Johns Hopkins University, where my research focused on information retrieval and NLP. Working across big tech, startups, nonprofits, and academia have given me a strong sense of how novel technologies are developed, adopted, and governed across different contexts and priorities. A significant part of my work involves building the education and infrastructure for broader participation in AI development and adoption as a means of surfacing gaps and reducing harm. Refer to my LinkedIn profile to learn more about my work as the first AI Librarian at the University of Chicago.