When Should Managers Trust Generative AI? Calibrating Human Reliance from Algorithm Aversion to Over-Reliance

Abstract

Generative artificial intelligence (GenAI) is increasingly embedded in managerial decision-making, where it can summarize information, generate alternatives, evaluate scenarios, and provide recommendations. Yet the managerial value of GenAI depends not only on the quality of its outputs but also on how humans respond to them. Prior research presents an apparent tension. Studies of algorithm aversion show that decision-makers may reject useful algorithmic advice after observing algorithmic error, whereas research on algorithm appreciation shows that people may sometimes rely more heavily on algorithmic than human advice. More recent work raises the opposite concern: managers may over-rely on GenAI because fluent, confident, or authoritative outputs can appear credible even when they are incomplete, biased, or incorrect. This article integrates research on algorithm aversion, algorithm appreciation, human-AI collaboration, managerial expertise, information processing, and GenAI-enabled decision support to develop a framework of calibrated human reliance. The central argument is that effective human-GenAI decision-making requires neither greater trust nor greater skepticism, but alignment between the degree of human reliance and the reliability and decision context of AI advice. The framework distinguishes four reliance states: under-reliance, appropriate rejection, appropriate reliance, and over-reliance. It further proposes that reliance calibration is influenced by four contextual dimensions: managerial expertise, output verifiability, uncertainty, and decision consequences. The article contributes to the emerging literature by distinguishing trust as an attitude from reliance as decision behavior and by shifting managerial attention from whether GenAI should be trusted to when and how much managers should rely on its outputs. Keywords: generative artificial intelligence; trust in AI; reliance; algorithm aversion; algorithm appreciation; over-reliance; human-AI collaboration; managerial decision-making; calibration