Abstract:

The rapid advancement of artificial intelligence (AI) has prompted a significant shift in the nature of human-AI interactions, transforming AI from a mere tool into a collaborative agent embedded in social contexts. As AI systems become more autonomous and interactive, the dynamics of trust in AI have evolved. Trust is no longer simply a one-way human trust toward machines, but also involves mutual trust between humans and AI, AI’s trust in humans, and even trust among AI agents themselves. Despite the growing importance of these complex relationships, there is a lack of comprehensive theoretical models that integrate human-human trust with human-AI trust, particularly in terms of how trust evolves, updates, and transfers across different agents and contexts.

The proposed research aims to address these gaps by developing a novel experience-based framework for trust formation and transfer that integrates insights from social psychology and engineering psychology. Specifically, this proposal focuses on the dynamic nature of trust in human-AI interactions, where trust is not a static belief but a process that evolves over time through experience and is transferable across different agents and contexts. The core innovation of this framework lies in its dual-subject approach, treating both humans and AI agents as trustors and trustees, and in its focus on trust experience transfer. This allows for a unified theoretical model that explains how trust is not only learned through interactions but also transferred from one interaction partner to another, facilitating trust-building across human and AI agents. The research will explore three key research questions: (1) How do different trust subjects-humans and AI agents-learn trust through experience? (2) To what extent can trust-related experiences be transferred across agents, and what factors influence this transfer? (3) How do individual characteristics of the trustor and interaction characteristics moderate the learning and transfer of trust? By addressing these questions, the research will contribute to a deeper understanding of the mechanisms underlying trust formation in human-AI collaboration.

The proposed model of trust experience transfer posits that trust is a dynamic, evolving process where individuals (or agents) update their trust beliefs based on feedback from previous interactions. This process is not confined to a single trust target, but rather extends across different agents, facilitating the transfer of trust from one relationship to another. For example, trust built through prior human-AI interactions may inform future human-human interactions or trust in other AI agents. Additionally, the model accounts for individual differences in trustors (e.g., age, anthropomorphism, AI characteristics, etc.) and contextual factors (e.g., interaction context and interaction relationship), which influence how trust is learned and transferred. To empirically test this framework, the research will employ an experimental paradigm involving both human participants and AI agents. This paradigm will manipulate the trust subject (human or AI agent) and the nature of their interaction, allowing for the examination of how trust develops and transfers between different agents. The experiments will use a combination of behavioral data and computational models to analyze trust-building processes in both human-human and human-AI interactions. The proposed methodology represents an innovation in trust research, as it explicitly incorporates AI as a trustor and allows for the examination of multi-agent trust dynamics in controlled settings.

The anticipated contributions of this research are threefold: First, it will develop a unified model of trust that integrates human-human and human-AI trust, providing a comprehensive framework for understanding trust dynamics in mixed-agent systems. Second, the research will introduce the concept of trust experience transfer, demonstrating how trust can generalize across different agents and contexts, offering new insights into the adaptability and fluidity of trust. Finally, the study will provide empirical evidence regarding the role of individual and contextual factors in trust development and transfer, offering practical implications for the design of trustworthy AI systems. This research has significant implications for the design and deployment of AI systems, particularly in contexts where AI must interact with multiple agents and adapt to dynamic trust environments. By understanding how trust is built and transferred across human and AI agents, the research will inform the design of AI systems that can better calibrate trust, improve collaboration, and ensure the ethical deployment of AI in socially sensitive contexts.

In conclusion, this proposal aims to advance the field of human-AI trust by providing a dynamic, experience-based framework for understanding how trust is learned and transferred across different agents. By integrating social psychology with engineering psychology, this research will offer both theoretical insights and practical guidance for the design of more reliable and trustworthy AI systems.

Key words: human-machine trust, experience transfer, human-AI mutual trust, interpersonal trust