Motivating Transparent Communications about Bias in Healthcare Technology Development.

Abstract

As healthcare artificial intelligence (AI) systems advance, their capacity for bias (e.g., as a function of patient protected characteristics) increases as well and these limitations are often left undisclosed by developers. Here, the question arises - do supportive motivational messaging designed to increase buy-in inspire healthcare AI developers to transparently communicate about bias in their technology? Computer science students (Study 1: N=271; Study 2: N=209) were randomly assigned to receive a brief communication framed in either an autonomy-supportive (choice promoting) or controlling (judging and pressuring) way, emphasizing either personal benefits (gaining profit) of transparency or legal implications of non-transparency. Results showed that while communication type was not associated with behavioral intention to engage in an educational course on transparent communication about bias, both internal (self-directed) and external motivations were associated with greater intention to take a course to build transparency-congruent technology skills, as well as greater ethical voice - intention to speak up in the service of positive transparency-consistent cultural change, and lower antagonism - i.e., a lower critical perspective regarding the need for transparency. Findings suggest that universities and workplaces should provide students and developers with a broadly supportive motivational climate, rather than providing a singular brief training.