Algorithmic Fairness and Service Failure: Why Firms Should Want Algorithmic Accountability

Co-Authors: Kalinda Ukanwa (USC) & Bill Rand (NCSU)

Abstract:

The past years have witnessed growing consumer concern about the fairness implications of the widespread adoption of AI in customer relationship management (CRM). To protect consumers against bias from algorithmic service decisions, regulators have introduced legislation holding firms accountable for the fairness of their algorithmic decisions. However, regulators have yet to invest in the systematic monitoring of algorithmic fairness, with the laws typically tasking firms with the detection and elimination of algorithmic bias. The resulting lack of transparency reduces firms’ ability to manage consumer expectations, leaving it to consumers to assess the – perceived – fairness of CRM outcomes. We posit that, to this end, consumers gather information about firm actions from their immediate social network, and we build a mathematical model to characterize how beliefs of bias may propagate within the market – even if the firm is using a fair algorithm. We show that, paradoxically, the lack of algorithmic transparency may lead to a divergence between consumer perceptions and the judicial view regarding the fairness of firm actions – under certain conditions, a firm with a fair algorithm can be perceived by the population as less fair than a firm with a biased algorithm. Using agent-based modeling, we also demonstrate how a watchdog institution may help in correcting such misperceptions.