Citation

Dark Matters: Bias in Speech AI

Author:
Diedrick, Johann; Lynn Stoever, Jennifer; Túbọ̀sún, Kọ́lá; Sprang, James Allister
Year:
2021

For this roundtable at Pioneer Works in June 2021, Jennifer Lynn Stoever, Kọ́lá Túbọ̀sún, and James Allister Sprang discussed Black speech, racial bias, creative possibilities, and speculative technological futures. The panelists explored the questions: What’s the history of the relationship between racial bias and Black speech? How are these systems already in our daily lives? How are these systems related to other voice technologies we might already be familiar with? How do these systems function?

The Bias in Speech AI Roundtable was part of the Dark Matters series. Dark Matters is an interactive web experience that spotlights the absence of Black speech in datasets that train voice assistants like Siri, Google Home, and Alexa. The project also reveals the racial bias and code switching that results. Through a three-dimensional visualization of major speech datasets, viewers come into contact with vacuums of space representing these data voids. Intertwined are narratives attesting to the resilient and resistant qualities of Black speech, suggesting how we might create more equitable futures.