You can measure how different changes to an AI system affect participants' trust on a quantitative scale. But you'll miss a lot!
Our Ethnography of NASA engineers building and using an AI tool to build mission-critical space software finds that Trust has as much to do with a tool's location within an social and organizational context, as it has to do with aspects of the tool itself. This study also questions boundaries between AI "builders" and "users" as well as "prototype" and "final" systems, as we watched the tool be built, used, and then built and used some more, often by the same people. We propose that as AI tools proliferate into knowledge work teams, AI Trust is best conceived as an aspect of the the operator-human collaboration ("collaborative trust") as they work together and with a wider team to produce creative output.
In this presentation, I will present findings from our CHI 2021 paper, briefly contrasted with recent work on an Open Source Deepfake tool building community, in hopes that this will provoke an engaging discussion of Trust (and other oft-vague descriptors) in AI.
Link to NASA CHI 2021 paper, Link to Deepfake FAccT 2022 paper. Please email dwidder@cmu.edu if there is any way I can remove accessibility barriers to your participation in this talk and discussion.
Note that this is a Wednesday seminar
David Gray Widder (he/him) is a Doctoral Student researching AI Ethics in the School of Computer Science at Carnegie Mellon University, and previously at Intel Labs, Microsoft Research, and NASA’s Jet Propulsion Laboratory. He was born in Tillamook, Oregon and raised in Berlin and Singapore. He maintains a conceptual-realist artistic practice, advocates against police terror and pervasive surveillance, and enjoys distance running.