人类对人工智能信任的接受度及脑认知机制研究:实证研究与神经科学实验的元分析

A study on the acceptance of human trust in artificial intelligence and brain cognitive mechanism:A meta-analysis of empirical studies and neuroscience experiments

人工智能技术广泛渗透到工作与生活的各方面,在带来巨大变革的同时,其自身固有的不确定性、脆弱性和恐怖谷效应,使得用户对其采纳抱有怀疑与犹豫的态度。人类对人工智能的信任问题由此产生。基于此,本文将信任纳入传统的技术接受模型之中,探讨在人工智能的使用环境下,信任与感知易用性、感知有用性和系统采纳意愿之间的关系。同时,将信任分为认知信任和情绪信任两个维度,研究了这两个维度的前因:透明度和拟人化。通过实证研究文献的元分析,验证了包含信任元素的人工智能接受模型的有效性。进一步,通过脑成像元分析方法,研究了信任概念在人类大脑中的发生位置和运行机理。研究结果表明:(1)信任对于技术接受模型中的感知有用性、感知易用性和采纳意愿均有显著的正向影响,证明信任是驱动模型中上述结构的重要因素。(2)信任区分为认知信任和情绪信任,透明度和拟人化分别是认知信任和情绪信任的重要前因。(3)信任的心理理念主要发生在大脑的壳核和颞上回,壳核的激活表征对奖励的预期,颞上回的激活表征同理心的激发,这些证据揭示了人机协作过程中“信任”的神经生理机制。本研究以信任为核心,拓展了人工智能使用环境下的技术接受模型,同时揭示了人类对人工智能信任的脑认知机制。这些理论成果为采纳人工智能技术的企业正确实施该项技术,提升用户的可信度提供了管理启示。

Artificial intelligence(AI)technologies are penetrating a wide range of work and life.While AI has brought great changes,its uncertainty,vulnerability and uncanny valley effect have made users skeptical and hesitant to adopt AI technologies.This leads to the issue of human trust in AI technologies.Current literature on the trust in AI tends to focus on AI technologies with very few investigating the impact of trust-related factors on the acceptance of AI technologies from a management perspective.Although AI has been successfully applied in autonomous driving,robotics,healthcare,and recommendation systems,a review of current literature shows that it is important to research into the factors affecting human trust in AI from the perspective of acceptance of AI technologies.Additionally,neuroscience experiment methods,which have been applied to management research,have been used to study the brain cognitive mechanism of users′trust in the management system.Against the backdrop of the above context,this study used two meta-analysis methods to analyze empirical studies and neuroscience experiments on human trust in AI.A theoretical framework for the acceptance model of AI trust was proposed,and the cognitive neural mechanism of trust was revealed.This study incorporated trust into the Technology Acceptance Model(TAM)to explore the relationship between AI trust and Perceived Ease of Use(PEU),Perceived Usefulness(PU),and Intention to Use(IU).The study divided trust into two dimensions:cognitive trust and emotional trust.It investigated the antecedents of these two dimensions:transparency and anthropomorphism.The validity of TAM on AI that included trust was verified through a meta-analysis of the empirical research literature.Further,this study investigated the activation regions and operating mechanism of trust in the brain by using the meta-analysis method of brain imaging.The meta-analysis of the empirical research results showed that:1)Through the heterogeneity tests and publication bias tests of meta-analysis method,the effectiveness of TAM including trust was verified by calculating the effect values among five groups of variables:transparency-trust,anthropomorphism-trust,trust-IU,trust-PU,and PEU-trust.2)PEU affected human trust in AI;trust strongly affected PU;the more human had trust in AI,the more they were willing to adopt AI systems.3)Anthropomorphism was an important factor influencing human emotional trust in AI,and transparency was the main factor influencing human cognitive trust in AI.The meta-analysis of neuroscience experiments results showed that:1)Two brain regions,the Putamen and the Superior Temporal Gyrus,were activated by trust.2)The function of the Putamen represented people′s expectation of reward.The activation of the Putamen confirmed that human trust in AI included both cognitive trust and emotional trust.The process of building trust was resulting from the combined influence of cognitive trust and emotional trust.3)The activation of the superior temporal gyrus confirmed that the trust between human and AI systems included empathy,and that humans would treat AI as their peers in the process of human-AI collaboration.In summary,this study used two meta-analysis methods to extend the application of TAM in the context of AI.It revealed the brain cognitive mechanism of human trust in AI.The theoretical contributions of this study are as follows:1)The study constructed TAM centered on trust,and clarified the impact of PEU-trust,trust-PU,and trust-IU in TAM.The study suggests that trust can be used to drive the elements in TAM when applied the model to AI systems.Increased human trust in AI could increase PU and IU,which could lead to the use of AI systems.2)The study considered trust from two distinct dimensions:cognitive trust and emotional trust.It conducted the meta-analysis of transparency and anthropomorphism as the antecedents of cognitive trust and emotional trust.In human-AI interactions,transparency could help calibrate trust by conveying information about AI′s uncertainty and vulnerability.Anthropomorphism,in its various applications,especially in robotics,could improve AI′s social ability and eliminate the emotional barrier between human and AI,is a key factor to enable trust between human and AI.3)This study provided the first meta-analysis of brain imaging data based on trust issues in human-computer interaction in the field of management.The study identified the location of“trust”activation in the brain,and revealed the brain cognitive mechanism of trust.The study has laid foundation for future research into economic prediction using human and AI interaction in the field of management.These theoretical contributions offer management insights for enterprises to appropriately adopt AI technologies and improve users′trust in AI technologies.