Trust, Automation Bias and Aversion: Algorithmic Decision-Making in the Context of Credit Scoring

Abstract

Algorithmic decision-making (ADM) systems increasingly take on crucial roles in our technology-driven society, making decisions, for instance, concerning employment, education, finances, and public services. This paper aims to identify peoples’ attitudes towards ADM systems and ensuing behaviours when dealing with ADM systems as identified in the literature and in relation to credit scoring. After briefly discussing main characteristics and types of ADM systems, we first consider trust, automation bias, automation complacency and algorithmic aversion as attitudes towards ADM systems. These factors result in various behaviours by users, operators, and managers. Second, we consider how these factors could affect attitudes towards and use of ADM systems within the con-text of credit scoring. Third, we describe some possible strategies to reduce aversion, bias, and com-placency, and consider several ways in which trust could be increased in the context of credit scor-ing. Importantly, although many advantages in applying ADM systems to complex choice problems can be identified, using ADM systems should be approached with care – e.g., the models ADM sys-tems are based on are sometimes flawed, the data they gather to support or make decisions are easily biased, and the motives for their use unreflected upon or unethical.