Researchers develop mental model framework to predict and measure human trust in AI

Trust between humans and AI systems is crucial for safe and effective collaboration, yet measuring and predicting trust remains a complex challenge. Researchers have now developed a mental model-based framework that aims to capture the multidimensional nature of trust and enable AI systems to infer human trust levels.

How the framework works#

The proposed model formalizes intuitions about trust found in existing research literature, creating a structured approach to understanding trust perception. Rather than treating trust as a single metric, the framework recognizes that trust encompasses multiple dimensions—capturing how people judge factors related to an AI agent's performance, process, and purpose.

The researchers show how this framework can help define key concepts including:

• Trust evolution over time as humans gain experience with an AI system
• Human reliance decisions based on trust assessments
• How trust influences human decision-making
• What constitutes an appropriate level of trust for any given agent

Validating the approach#

The team tested their framework through human subject studies to verify two core hypotheses. First, they examined whether deliberately adjusting human beliefs about an agent—guided by predictions from their mental model—could produce measurable changes in trust. Second, they investigated whether modifications to specific belief dimensions would translate into corresponding shifts in how people perceive the agent's trustworthiness.

Why it matters#

This framework provides a foundation for building trust-aware decision-making systems that can adapt to human expectations. By enabling AI systems to model and predict trust dynamics, developers can design interactions that foster appropriate confidence levels and safer human-AI collaboration.