Transparency/clarity key to AI trust and acceptance- study proves
Shin 20 (Donghee,College of Communication and Media Sciences, Zayed University, “User Perceptions of Algorithmic Decisions in the Personalized AI System:Perceptual Evaluation of Fairness, Accountability, Transparency, and Explainability”) 14 Dec 2020
https://www-tandfonline-com.proxy.lib.umich.edu/doi/full/10.1080/08838151.2020.1843357 acd
The existence of trust is key to promoting technology acceptance (Shin, ) as it facilitates openness and transparency in adoption process. Findings in the work of Lee () imply that trust plays a mediating role in algorithm acceptance. In the news recommendation context, trust is considered a mediator of relationships between behavioral intentions and individual characteristics, and algorithm technology (Shin & Park, ). It can be reasonably considered trust can play a mediating role in the context of AI. For estimating indirect effects for FATE and trust, a non-parametric bootstrapping approach was employed. Bootsrapping is one of the crucial parts in structural equation modelling when it comes to test mediation effects. Mediation analyses assessed the indirect effect of explainability on the association of FAT and trust. Bootstrapping techniques can be used when examining mediation to gain confidence limits for specific indirect effects. Variance accounted for (VAF) is to evaluate the indirect effect and the value of greater than 80% is full mediation, while greater than 20% but less than 80% is partial mediation (per Hair et al., 2013). The 95% confidence interval for the indirect effect via trust was obtained using bootstrapped resampling. Mediation is confirmed if such a confidence interval does not contain zero. The standardized indirect effect (Table 2) shows that exogenous latent constructs have partial mediation effects toward emotion through trust. The partial mediations observed in the test suggest that there are other potential mediators for trust or emotion.
The findings suggest that users trust AI recommendations as much as they understand FATE. The findings of the study provide evidence of the FATE model of human-AI interaction. The model illustrates that interacting with algorithms involves a series of interrelated cognitive processes wherein features of algorithms are used to formulate a heuristic of user motivation and to trigger action in AI services. The findings of this study offer interesting insights on the relationships between heuristics, quality, and trust in algorithm. First, using the HSM as a primary framework, this study examined how algorithmic features influenced users’ trust and emotion through two different routes of cognitive processing. Users’ heuristic process of FATE influences user trust and increased trust influence systematic processing of performance expectancy, which is positively associated with emotion and satisfaction. Just as transparency, accountability, fairness, and explainability are considered critical values in social systems, so too are they in algorithm-based services. Not only do qualities of FATE play a significant role in establishing trust, but they also play an anchoring role in developing user evaluation of performance as well as perceived usefulness and convenience of using AI. The model shows that users rely on FATE as heuristic tools to assess trust in algorithm. The model shows a dual process: Heuristic process is less resource demanding and less analytical as users normally do not have expertise to evaluate specialized algorithmic features, whereas systematic process is more effortful and more deliberate (whether personalized results are accurate, precise, and private) based on the established trust. Algorithm users develop their own processes of algorithmic trust based on cognitive processes related to FATE. User reactions to perceived usefulness and convenience are not automatic; rather, they are dependent upon or at least closely related to how users recognize, understand and process the information regarding FATE. Such a relationship can be described as heuristic insofar as users rely on FATE to determine their feelings on usefulness and convenience around algorithm services. In other words, users figure out usefulness and convenience in the context of algorithm according to the FATE of content. This finding is in line with the arguments of previous studies, which have shown the contextual nature of such variables (Shin & Park, ). Second, the findings suggest the significant role of trust in human-AI-interaction. With the pervasive role of algorithms in our lives, a question is how people trust an algorithm’s decision? How trust is formed and evolves in the course of adoption may provide important clues in designing and developing AI services. More and more people are aware that algorithms are not neutral and concern is rising that AI may have human prejudices. People would like to understand how algorithms work, how the processes work, and to what extent the results are fair. The model in this study shows a clue on how trust is gained and sustained and with what factors. Trust plays a liaison role by linking uncertain issues to measurable values and establishing confirmation for usage and adoption of personalized algorithms. It's worth noting that the model without trust mediation shows similar results but is less and powerful than the initial model with trust. Without this trust factor, overall paths remain significant but less robust with lower C.R. values. This signifies the role of trust as intermediate states that foster the development of engagement. Although previous research consistently has shown the role of trust in AI and algorithm (e.g., Alexander et al., ), this study empirically proves the role of trust in AI, its antecedents, intermediating role, and heuristic-systematic process. In algorithms, users get a sense of trust when they are assured with the level of FATE. When users trust algorithm systems, they tend to believe that system services are useful and “convenient” just like overriding effects (Shin & Park, ). The mediating role of trust between satisfaction and FATE supports the liaison role of trust in algorithmic processes: linking heuristic and system evaluation (Ferrario et al., ). Trust significantly mediates the effects of FATE on users’ satisfaction. Satisfaction promotes trust, and in turn, influences user perception of FATE. Higher satisfaction implies greater trust and suggests that users are more likely to continue to use an algorithm. Affording more user trust and assured emotion may warrant users that their individual data will be used by legitimate and transparent processes, thereby generating positive trust toward the AIs and the providers, ultimately leading to heightened levels of satisfaction. Previous research findings have confirmed the mediating role of trust in diverse contexts (Zhang et al., ). Based on the mediating role, it can be inferred that trust between users and algorithms is the underlying key factor in acceptance and experience of AI.