AI Algorithm Recommendation And Consumer Response In The Intelligent Er

Artificial Intelligence（AI）,as a strategic technology leading the new round of technological revolution and industrial change,has a strong spillover effect,and it is expected to contribute about$13 trillion to the global economy and increase the average annual growth rate of global GDP by 1.2%by 2030.At the same time,AI is driving industries to maintain a high market growth rate by accelerating the digitization of industries,for example,algorithm-based AI recommendations have made great strides in various fields such as e-commerce（e.g.,JD.com and Amazon）,content distribution（e.g.,Tik Tok and Netflix）,and health treatment（e.g.,Quark and Watson Health）.However,the development of AI recommendations is also facing several negative concerns like 1)the"black box"concerns,reflecting algorithms’inability to explain their decision processes and logic to users,and 2)the"information cocoon"concerns,i.e.,consumers are sometimes forced to be bound by the algorithm in a convergent and biased preference network,both of which have greatly damaged consumer trust and experience of human-computer interaction.Therefore,it is crucial to deal with such concerns to further deploy AI recommendations in the future.Existing studies on AI recommendations involves multidisciplinary fields such as computing science,artificial intelligence,sociology,and management.Specifically,existing research on AI recommendations in the field of artificial intelligence or computing is mainly oriented to a purely technical approach.Although AI algorithms have made great progress and development in the field of computing and artificial intelligence,the technical field usually views AI recommendation as a neutral tool based on efficiency and accuracy evaluation.However,the goal of creating superior accuracy in the technology field does not necessarily achieve the goal of creating a valuable consumer experience in the marketing field,as the former is mainly focused purely on technical aspects from a broad perspective,ignoring the heterogeneity of application scenarios and consumer psychology.Therefore,it is necessary to integrate consumer insights and specific application scenarios into AI technology,and then explore how and why consumers respond to the algorithm-based AI recommendations from the perspective of human-computer interaction.However,the existing research in the field,particularly based on the perspective of interpretability and granularity of AI recommendations,is mainly at the theoretical level of opinion-based discussion,and there is few empirical studies.Prior empirical studies in the marketing field are mainly focused on exploring consumer responses to algorithm-based（vs.human-based）recommendations in different contexts,and their mechanisms are mainly based on competency perceptions.However,the existing theoretical mechanism of competence is still insufficient to explain the phenomenons of the"black box"and"information cocoon"of algorithm-based AI recommendation mentioned above,indicating that the existing research on consumer response to AI recommendations and its mechanism as well as boundary conditions is still insufficient.That is,some key theoretical mechanisms have not received sufficient attention from marketing scholars,for instance,the effect of the interpretability and autonomy perceptions of AI recommendations on consumer responses is increasingly crucial but yet understudied.To fill in this gap,this current research attempts to focus on the"black box"concerns from the interpretability perspective and focus on the"information cocoon"concern from the granularity perspective of AI recommendations.Specifically,we investigate the variances in consumer need for interpretability of AI recommendations in different decision-making domains and how to improve the interpretability of AI recommendations to promote consumer responses.Also,we investigate when and why consumers prefer coarse-grained（vs.fine-grained）AI recommendations in varying decision-making domains and among different populations.Building on the heuristic-systematic model and self-determination theory,this research proposes six hypotheses and empirically tests them through six experimental studies across different experimental scenarios（e.g.,healthcare,music,sneakers,and facial products）,different operations of consumer responses（e.g.,trust,purchase intention,click-through rate,satisfaction）,different manipulation of the independent variables,and different sample pools.Specifically,study 1 and study 2 are conducted to test H₁,i.e.,to explore the variances in consumer need for interpretability of AI recommendations within different decision-making domains,and study 3 is conducted to test H₂and H₃,i.e.,to examine the effect of post-hoc explanations in improving consumer responses and the mediating role of interpretability perceptions of AI recommendations,study 4 is conducted to test H₄,i.e.,to examine the varying effect of different types of post-hoc explanations in improving consumer trust in AI within different domains,study 5 is conducted to test H₅,i.e.,to explore consumer responses on AI recommendations with varying granularity in different decision-making domains,and study 6 is conducted to test H₆and H₇,i.e.,to examine the moderating role of consumer beliefs in free will and the mediating role of autonomy underlying granularity of AI recommendations and consumer responses.The results demonstrate that 1)consumer need for interpretability（NFI）for AI recommendations is higher in the utilitarian（vs.hedonic）decision-making domain,2)post-hoc explanations improve consumer responses（i.e.,trust and purchase intention）to AI recommendations and this facilitating effect is stronger in the utilitarian（vs.hedonic）domain,3)interpretability perceptions mediate the relationship between post-hoc explanations and consumer response and its mediating effect is stronger in the utilitarian decision-making domain,4)attribution-based post-hoc explanations are more likely to promote consumer trust in the utilitarian decision-making domain,whereas user-based explanations are more likely to promote consumer trust in the hedonic domain,5)consumers are more satisfactory with and exhibit higher click-through rate toward coarse-grained（vs.fine-grained）AI recommendations in the hedonic decision-making domain,while consumers respond more positively to fine-grained ones in the utilitarian domain,6)consumers with higher belief in free will respond more positively to coarse-grained AI recommendations（i.e.,satisfaction and click-through rate）relative to fine-grained AI recommendations,while consumers with lower belief in free will exhibit no significant difference in their responses,7)autonomy perceptions mediate the relationship between granularity and consumer response,further,this mediating effect only holds in the hedonic decision-making domain and among populations who hold high belief in free will.Our work has several important theoretical and practical contributions.First,this research,to the best of our knowledge,is the first to propose and define consumer need for interpretability（NFI）of AI recommendations,and empirically explore the differences in consumer NFI in different domains;Second,this research clarifies the varying role of post hoc explanations on consumer responses in different decision-making domains,and analyzes the matching effects of different post-hoc explanation types with decision-making domains on consumers trust and purchase intention;Then,this research explores the effect of recommendation granularity on consumer responses and analyzes the matching effect of granularity with decision-making domain and heterogeneous populations;Finally,this research explores the mediation mechanism of interpretability perceptions and autonomy perceptions on consumer response and its boundary conditions,thus deepening the literature on consumer response patterns to AI recommendations.As for managerial implications,first,this study presents consumers’varying need for interpretability of algorithm-based AI recommendations in different decision domains,which helps entities identify the significance of the"black box"concern in different application scenarios.Second,the facilitative effect of post-hoc explanations on consumer trust and response proposed in this study provides entities with an effective tactic to overcome or mitigate the"black box"concern.Then,the interactive effect of the different types of post-hoc explanations and decision domains proposed in this study helps AI algorithm engineers to develop differentiated XAI technologies for different application domains.In addition,the moderating effects of decision domains and consumer free will beliefs can help entities identify the significance of consumer autonomy in different decision-making domains and among different populations.Finally,our proposed framework can help make sense of and predict consumer responses to AI recommendations as well as its psychological mechanism,and also deepen the understanding of the new generation of AI.