信任能降低公众对人工智能技术的风险感知吗?

Can trust reduce the public’s risk perception of artificial intelligence technology?

论文基于一项全国性抽样问卷调查,从主效应和交互效应两个层面,探究信任是否以及如何塑造公众对人工智能技术的风险感知。结果显示:对政府官员和科学家的信任,均显著降低了对人工智能技术的风险感知;信任政府官员的影响力,不仅高于信任科学家,也高于人工智能知识;对科学家能力信任的影响力高于对科学家的道德信任。除了主效应,信任与知识之间还存在相互强化的交互效应,即信任对技术风险感知的影响力随着人工智能知识水平的上升而增强,随着知识水平的下降而减弱。论文进一步讨论了上述研究结果对风险管理与风险传播的政策意义与未来研究方向。

Few scholars doubt the importance of trust in explaining variation in public perception of technological risk. Relatively little, however, is known about what "trust" actually consists of in any given circumstance. What do people mean when they say "I trust"? In other words, trust is hard to measure in any society. On the one hand, it has multiple targets, such as various institutions, social groups, and general people;On other hand, the substance of trust in a given target has multiple dimensions, such as competence, objectivity, and fairness. In addition, although the main effects of trust on risk perceptions has been debated at great length, it remains unclear whether the influence of trust will be moderated by other important factors, such as knowledge. This present study is motivated by a concern to address this gap described above in the literature of risk analysis. Using a nationally representative survey of Chinese adults(N=1557), we address two major issues:(1) distinguishing between specific types of trust and further comparing the main effects of them on risk perceptions of artificial intelligence;(2) investigation the interaction effects between trust and knowledge. Results show that both of trust in regulators and in scientists are significantly and negatively correlated with public perceived risk. More specifically, the level of trust in regulators is lower than in scientists, but the strength of impact of trust in regulators is higher than that in scientists, and even than knowledge of artificial intelligence, which is measured by perceived familiarity of artificial intelligence. With regard to the different dimensions of trust in scientists, the competence-based trust has much stronger influence than value-based one. In addition to the main effect, there is a complementary interaction effects between trust and knowledge on risk perception. That is, the negative effects of the three kinds of trust on risk perception of artificial intelligence increases with increased levels of knowledge of the technology, and vice versa. There are several of policy implications. First, we should continue to invest in science education and science popularization initiatives in order to mitigate public risk perception of artificial intelligence. Second, it is critical to pay much attention to the importance of trust as an element of "social capital" and take initiatives, such as social dialogue and public engagement, to build, maintain, and strength institutional trust, especially trust in government. Third, because of the complementary effect between knowledge and trust, it is not sufficient to invest in either knowledge communication or trust cultivation;instead, policy makers and opinion leaders should focus on knowledge and trust simultaneously. We recognize that our study has several limitations. Above all, without experimental or longitudinal data, it is difficult to test the causal relationship validly, because the correlations between variables may be caused by some unobserved factors, such as personality. For this reason, the reader should exercise caution in interpreting the results from our statistical models. The second is that theoretically we classify trust into four categories based on different targets and dimensions, but empirically we only measure three type of trust because of limit of data. Finally, we only discuss the interaction effect between trust and knowledge, but other factors may also moderate the impact of trust. Therefore, we encourage future research pay much attention to these problems and advance our understanding of public attitudes towards technological innovations.