Algorithm Aversion

Our latest research offers a detailed exploration of algorithm aversion, defining it as the human bias toward under-trusting or avoiding automated decision systems, even when these systems demonstrably outperform human capabilities. This disuse, which stands in opposition to automation over-reliance, is fuelled by psychological factors such as expecting machines to be flawless and exhibiting a general distrust of non-human agents. We illustrate the serious, real-world consequences of this bias across fields like medicine, law, and autonomous driving, where valuable AI tools are ignored after users witness a failure. To systematically analyse these behaviours, the essay employs our Cognitive Susceptibility Taxonomy (CST), which classifies reactive states like Trust Oscillation and A-Noosemic Withdrawal that follow a system error. Finally, we conclude by advocating for specific design interventions focused on trust calibration, including progressive disclosure and thoughtful transparency overlays, to ensure appropriate reliance and foster productive human-AI collaboration.