Averse to Algorithms or Averse to Uncommon Decision Procedures?

Abstract: Organizations are increasingly adopting artificial intelligence to handle contract negotiation, adjudicate disputes, and to even make large-scale strategic decisions. Nevertheless, a large and growing body of literature claims that people are averse to adopting the recommendations of algorithms, arguing that they punish the mistakes of algorithms more harshly than equivalent mistakes committed by human decision-makers. Across four studies, we show that much of what appears as Algorithm Aversion can instead be explained by an aversion to counter-normative decision procedures. Largely, algorithms are excessively penalized only to the extent they are unconventional for a particular domain. Changing the norm can, in fact, reverse the effects of algorithm aversion. This implies that as norms shift and adoption of algorithmic decision aids (e.g., LLMs) in strategic contexts grows, people may be increasingly comfortable outsourcing control over important resolution processes and decision-making tasks to artificial intelligence.

Keywords: Algorithm aversion, artificial intelligence