Como o processo de elicitação de preferências influencia a aceitação de recomendações de inteligência artificial

Abstract

The adoption or aversion to a recommendation from an algorithm is a topic that arouses interest in researchers from fields as diverse as artificial intelligence, information systems, marketing, or consumer behavior. However, few empirical studies on the behavior of individuals in relation to algorithms are available in the literature and little is known about the subjective reasons that lead consumers to adopt or reject a recommendation made by an artificial intelligence agent. Studies indicate that the accuracy of the algorithm could influence acceptance, different studies indicate that the type of task in which the recommendation fits would influence acceptance. However, these studies presented a certain ambiguity in the conclusions and other reasons could lead to the acceptance of the recommendation. Like others researches that point to the misunderstanding of the internal logic of the algorithm, it would provoke the rejection of the algorithm's recommendation. On the other hand, there is an initial interaction between the algorithm and the user called the preferences elicitation process (PEP), a stage by which the user informs his preferences and needs used by the algorithm in preparing the recommendation, which can influence the acceptance of the recommendation. Studies indicate that this interaction process influences satisfaction with the recommendation because it provides a feeling of transparency and relevance in the formulation of the recommendation, creates an expectation of quality in the recommendation and can help in understanding the internal logic of the algorithm. However, few experiments were found that analyzed the PEP with the satisfaction of the recommendation and lacks further empirical studies. Another topic found in the literature, but also little explored, refers to the influence of subjective knowledge on the domain of recommendation on user satisfaction. So, does PEP influence satisfaction with the recommendation? Does the user's subjective knowledge attenuate or enhance the acceptance of the recommendation? Thus, the main objectives of this work are to study whether the preferences elicitation process (PEP) influences satisfaction with the recommendation as well as whether the individual's knowledge of the recommendation domain influences consumer satisfaction. To achieve these objectives, an exploratory experiment was carried out that manipulated different PEPs, considering the level of subjective knowledge about the recommendation domain. The analysis of the PEP alone showed marginally significant effects, however when analyzed together with domain knowledge, the results indicate significant differences in satisfaction. Thus, it was possible to verify the influence of PEP along with subjective knowledge on user satisfaction. Therefore, this exploratory research has contributed to the literature not only with new learning about the preferences elicitation process per se, but also with new studies involving the PEP and subjective knowledge about the recommendation domain. At the same time, the research collaborates with the practice in the corporate world, to bring new ideas for companies to improve their PEPs in operation or develop new strategies that provide greater satisfaction to consumers