An Empirical Study Of Mitigating Algorithm Aversion In AI Prediction

As artificial intelligence,machine learning,and algorithmic prediction continue to advance,algorithms are increasingly outperforming humans in a wide range of predictive tasks.Despite their superior performance,people often prefer human recommendations over those of algorithms,a phenomenon known as algorithm aversion.This thesis examines the factors that influence algorithm aversion and proposes interventions to mitigate it.We conducted a literature review on algorithm aversion and algorithm appreciation and developed a theoretical model of the factors influencing algorithm aversion based on task characteristics,human characteristics,and algorithmic characteristics.We then conducted three experiments with a sample of 404 subjects to investigate the effects of task type,familiarity,expertise,and the learning ability of algorithms on algorithm aversion.Our data were analyzed using paired samples t-tests,ANOVA,and regression analysis.The findings suggest that(ⅰ)algorithm aversion is more likely to occur in individual task conditions;(ⅱ)algorithm aversion is higher for subjective task types;(ⅲ)users’ perceived subjectivity of tasks can be changed by highlighting the benefits of quantitative task analysis;task type,expertise,and familiarity have a significant effect on algorithm aversion,and learning ability of algorithms moderates the effect of task type and expertise on algorithm aversion.This study contributes to our understanding of when algorithm aversion arises and proposes two interventions to mitigate it:shaping users’perceptions of task objectivity and demonstrating algorithmic learning ability.The findings have theoretical and practical implications for the future design of artificial intelligence and algorithms.