关键词: 面子, 反馈效价, 反馈采纳, 人工智能

Abstract: Negative feedback from managers is often resisted by employees compared to positive feedback. In work situations, the negative feedback given by leaders will upset employees, making them focus on their gaining or losing face rather than whether the feedback itself is reliable. But negative feedback can help employees correct their mistakes and adjust their work status in time, which has an irreplaceable and important role. Therefore, it is important to explore how to provide feedback to employees in an appropriate way. Previous research on facilitating feedback adoption has primarily explored (1) how to accurately and rationally deliver the content of feedback and (2) how to do so by adjusting the order and timing of multiple feedback pushes. However, according to the Face theory in China, as the lowly person in the power relationship, employees instinctively feel nervous and uneasy when they receive negative feedback, and may even see the evaluation as a denial of their abilities. The impact of the gap caused by this hierarchical relationship cannot be resolved by handling the feedback itself. However, few studies have explored how to balance negative feedback from the perspective of feedback sources. We think the AI systems in work situations may help, as studies have found that employees feel less face loss from negative advice given by AI and rarely resist it. Based on the traditional face theory in China, this study establishes a mechanism for the impact of leadership feedback on employee feedback adoption.
Study 1 used a situational experiment to simulate a work scenario in which employees receive performance feedback to initially explore the effects of positive and negative feedback from AI systems and human leaders on employee feedback adoption. Study 1 adopts a 2 × 2 between-groups design, with the two variables manipulated being the valence of feedback (positive feedback/negative feedback) and the subject of the feedback (AI system/ human leader), comprising four experimental scenarios. Subjects would randomly fill in one of the four contexts, and the number of pushers was same in all contexts. Two hundred and twenty questionnaires were randomly distributed with the help of Credamo, a widely-used online questionnaire platform in China, and the subjects were required to be working employees. A total of 208 valid questionnaires were returned. The ANOVA results showed a significant interaction effect between feedback valence and feedback source. Relative to AI systems, positive feedback from human leaders yielded a higher willingness to adopt. Negative feedback from AI systems can obtain a higher willingness to adopt relative to human leaders.
Study 2 used a situational experiment to simulate a workplace scenario in which employees receive performance feedback to explore in depth the impact of positive and negative feedback from AI systems and human leaders on employee feedback adoption through the mediating mechanism of face. Study 2 targeted 300 questionnaires to current employees through the Seeing Numbers platform, with no crossover between the sample and Study 1. At the time of distribution, the subjects were randomly divided into four groups based on the valence of feedback (positive feedback/negative feedback) and the source of feedback (AI system/ human leader), and 75 questionnaires were distributed to each group, and a total of 287 valid questionnaires were returned in the end. Finally, it was found that when a human leader provided positive feedback, the positive influence on feedback adoption through face gaining was stronger, and when he provided negative feedback, the negative influence on feedback adoption through face loss was stronger.
Study 3 was a two-period follow-up experiment with a group of delivery workers. It was designed to explore the mechanism in real work scenarios. A total of 307 valid questionnaires were collected. The results of data analysis showed that the hypotheses were supported, which indicated the high reliability of our study.
The findings of this study inspire us to promote cooperation between humans and AI to fully utilize each other's strengths rather than focusing on achieving complete automation (i.e. replacing humans with AI). This study reveals the mechanism of how AI systems and real human leaders work together to improve feedback adoption, which enriches the research in the field of human-computer cooperation and serves as a guide for the proper use of AI systems.

Key words: face, feedback valence, feedback adoption, artificial intelligence