社会与工程心理学研究室知识库

人类正在进入智能化时代，智能系统在生产与生活中发挥着越来越重要的作用。在这一时代背景下，人类会与智能系统产生越来越多的交互。人机信任是人与智能系统交互的一个核心问题，关系到人与机之间是否能够真正有效融合共生的关键。有关人机信任的研究在近十年来得到了研究者的重视。

现有研究的一个主要思路是探讨人际信任中存在的现象是否能够迁移到人机信任之中。既有研究发现，在人际信任中起到重要作用的内群体偏好效应，也可以促进人机信任的发展。但现有研究主要是基于最简群体范式的假想群体，尚未有考察真实群体所带来的影响。在实践中，考察真实群体对人机信任产生的影响更有现实意义，但与假想群体不同的是，真实群体会有其历史、文化、胜任力和刻板印象的差异，当智能系统被贴上群体标签时(如中国的机器人)，这些差异会影响用户对系统的总体态度和行为。此外，用户群体认同等级的变化(如爱国者VS卖国贼)以及认同的不同凸显(如一个群体被外群体威肋、时)可能会影响他们对内群体的信任。那么，智能系统具备与被试相同或不同的群体身份(内群体一外群体)是否会影响用户对智能系统的信任?个体的群体认同感和群体身份的凸显是否会影响群体身份对信任的作用?针对这些问题本文进行了两个研究。

研究一探索了在真实群体中，拥有不同群体身份、不同可靠性水平的智能系统和不同群体认同个体对用户信任的影响，实验采用了2(群体身份)X2(群体认同)X3(可靠性)混合实验设计，并测量了被试的主观信任和行为信任。

结果发现，可靠性会影响被试对智能系统的信任，被试对高可靠性的系统信任高于低可靠性的系统;群体身份影响被试对智能系统的信任，被试对内群体身份的系统信任高于对外群体身份的系统;群体认同影响被试对智能系统的信任，在外群体胜任力略强的情况下，低认同的被试对内群体系统的信任低于外群体系统的信任，高群体认同的被试对内外群体系统的信任没有差异。

研究二探索了认同凸显对群体身份影响信任的作用。其中，研究二a探索了群体认同凸显是否会影响群体身份对信任的效应。实验采用了2(群体身份)X2(认同凸显)X2(可靠性)混合实验设计。结果发现，凸显的主效应显著，认同凸显组的被试对内群体的信任低于对照组。为了考察研究二a中的认同凸显为什么使被试的信任降低了，研究二b探究了不同凸显方式的作用，实验采用了2 (群体身份)X3(认同凸显)X2(可靠性)混合实验设计。结果发现，凸显的主效应显著，官方评价组的信任高于其他组;群体身份和群体认同有交互作用，高认同组的被试对内群体的信任高于对外群体的信任。

总的来说，本研究探索了真实群体中群体身份的作用，并首次探索了群体认同的程度和认同凸显对人机信任的影响。研究发现在真实群体中，智能系统的群体身份影响用户对其的信任，群体认同能调节群体身份的影响，这说明人际信任中群体认同的作用可以迁移到人机信任中，可以为人机信任的应用提供数据支持。但是群体认同凸显不能调节群体身份的影响，出现这一结果的原因还需要进一步探索。

In the era of artificial intelligence, intelligent systems play an increasingly important role in everyday life and interact with human beings at an unprecedented frequency. Human trust in such systems become a core research topic since it plays an important role in human-computer interaction. In the past decade, researchers have realized such importance and paid a lot of attention in this field.

Most studies adopted an approach to investigate whether the mechanisms found in the research of interpersonal trust can be transferred to the human-computer trust. Existing research has found that the effect of in-group favoritism, which plays an essential role in interpersonal trust, can also promote the development of human-computer trust. However, current studies only examined such effects in hypothetical groups using the Minimal Group Paradigm. While it is more practical to examine such impacts in real groups, unlike hypothetical groups, real groups are more complex. On the one hand, real groups have a different history, culture, capabilities, and stereotypes, which may influence users' general attitudes and behaviors toward an intelligent system when such a system is given a group tag (e.g., Chinese robots). On the other hand, the users may vary in their levels of group identity (e.g. patriots vs. traitors) and their identity may at different salience (e.g., when the group is threatened by an out-group), which both may influence their trust toward their in-group. Whether users may have different levels of trust toward an intelligent system which has different group membership (in-group vs. out-group) as users? o users' group identity and group membership salience also affect trust? To investigate these two questions, we conducted two studies.

Study 1 explored the impact of intelligent systems with different group memberships and different levels of reliability and individuals with different levels of group identity on user trust in real groups. This study used a 2 (group membership) x 2 (group identity) x 3 (reliability) mixed experimental design, in which group membership was a between一subject variable, group identity, and reliability were within-subject variables, and subjective trust and behavioral trust of the participants were measured. Results were found that reliability affected the trust of the participants in an intelligent system. The participants' trust of the system with high reliability was higher than the system with low reliability. The group membership affected the trust of the participants in the intelligent system. Participants' trust toward the in-group system was higher than the out-group. Then, group identity affected the trust of the participants in the intelligent system. In the situation that the out-group competence was slightly high, the low identity participants had less trust in the in-group system than the out-group system. There was no difference in the trust of the in-group and out-group systems among the participants with high group identity.

Study 2 explored the role of identity salience in moderating the influence of group membership on trust. Among them, Study 2a explored whether the salience of group identity affects the effect of group membership on trust. The study used a mixed experiment design of 2 (group membership) x 2 (identity salience) x 2 (reliability). The results showed that the main effect of salience was significant, and the trust on in-group of identity salience group was lower than that of the control group. In order to examine why the salience in Study 1 reduced the trust of the participants, Study 2b explored the role of different salience methods. The Study used a mixed experiment design of 2 (group membership) x 3 (identity salience) x 2 (reliability). The results showed that the main effect of manipulation was significant; the trust of the official evaluation group was higher than that of other groups. There was an interaction between group membership and group identity, and participants in the high identity group had higher trust in the in-group than the out-group.

In general, this study explored the role of group identity in real groups, and for the first time explores the extent of group identity and the impact of identity on human-machine trust. The study found that in real groups, the group membership of intelligent systems affects users' trust toward them, and group identity moderated the effect of group membership, which showed that the role of group membership in interpersonal trust could be transferred to human-computer trust, which can provide data support for the application of human-machine trust. However, group identity salience cannot moderate the effect of group membership, and the reason for this needed to be further explored.