Researchers shed new light on the phenomenon often referred to as ‘algorithm aversion’. The study suggests that humans don’t always mistrust machines, but instead may struggle to learn how to effectively use them due to a bias in their learning process. When humans do not follow an algorithm’s recommendations, they miss out on opportunities to observe its accuracy, leading to an incomplete understanding of its decision-making capabilities.
New research by professors at ESMT Berlin reveals that machines often make superior decisions compared to humans, yet humans tend to override algorithmic recommendations, resulting in suboptimal outcomes. The study challenges the notion that this behavior solely stems from mistrust of machines, suggesting that contextual factors impede accurate assessment of machine performance. The researchers designed an analytical model where a human decision-maker supervised a machine making critical judgments. Trust in the machine increased when its recommendations were correct, but humans failed to verify accuracy when no further action was taken, hindering learning opportunities.
Biased learning occurs through the interplay between human decisions and subsequent evaluation of machine performance, leading to a reinforcement of the inclination to override algorithms. This bias is not solely driven by mistrust but also by an inadequate understanding of effectively utilizing machines. Professor de Véricourt highlights that relying solely on the correctness of machine predictions for learning may contribute to the inappropriate and systematic override of algorithms.
The research underscores the importance of trust in machine decision-making and the need for continuous consideration of machine advice to enhance learning opportunities. Professor Gurkan emphasizes that human decision-makers should embrace complete learning with machines rather than selective reliance. These findings highlight the significance of collaboration between humans and machines, guiding when to trust machines and when to exercise independent judgment.
Ultimately, this research provides a framework for effectively leveraging machines in decision-making processes, emphasizing the need to understand the optimal circumstances for heeding machine recommendations versus making independent decisions.