Reconciling Professional Vision and AI Recommendations

Abstract:  In this paper, based on an ethnographic study of doctors performing colonoscopies, we show how professionals bolster confidence in their decisions- especially when these decisions conflict with AI recommendations- by employing not only their professional vision but also semiotic work. We connect the uncertainty introduced by AI to the differences between human and AI reasoning and demonstrate how workers address these challenges through semiotic work enabling them to confidently validate or reject AI outcomes. Our findings reveal that the difficulty in developing a confident diagnosis while interacting with AI stems from a thinking that effective collaboration with AI requires understanding AI reasoning.  Contrarily,  we  propose  that  justifying  professional  vision  with  semiotic  work  allows  to confidently validate or reject AI outcome. We show how doctors can transform the uncertainty induced by AI  into  a  confident  decision  by  transforming  their  intuitive  professional  vision  into  an  explicit reasoning process which use semiotic work, the interpretation of additional signs that go beyond the object of scrutiny.

Bio:  During her PhD, she conducted ethnographic fieldwork with shopfloor workers producing molds for the automotive industry. From this fieldwork, she co-published a paper with Ruthanne Housing, highlighting how workers who lack official training or peripheral  learning opportunities can achieve computational literacy through what they term "Vicarious Coding." Based on the same fieldwork, she also  published  a  paper  in  collaboration  with  Jean  Clarke,  which  explores  how  workers  in  low-skill environments can fin meaning in their work by engaging in anomalous craft. Currently, she is involved in  a  hospital  project  examining  how  AI  infleunces  doctors'  work  practices.  Over  the  course  of  18 months,  she  has  observed  doctors  in  operating  rooms  and  has  been  analyzing  how  AI's recommendations integrate with and enhance professional vision.