A Quantum Model of Trust Calibration in Human-AI Interactions.

Area of Science:

• Human-Computer Interaction
• Cognitive Science
• Artificial Intelligence

Background:

• Understanding human trust in artificial intelligence (AI) is crucial as AI systems become more integrated into daily life.
• Human reliability judgments of AI can fluctuate, necessitating models that capture this dynamic.
• Previous models, like Markov random walks, may not fully account for individual variability in trust.

Purpose of the Study:

• To compare the predictive accuracy of quantum versus Markov random walk models for human reliability judgments of AI.
• To identify neural correlates associated with changes in human trust and reliability assessments of AI systems.
• To explore the potential for real-time adaptation of AI models based on human trust indicators.

Main Methods:

• A mixed-methods experiment was conducted to explore reliability calibration in human-AI interactions.
• Behavioral data on human reliability ratings were collected and used to assess model performance.
• Electroencephalography (EEG) was employed to capture neural activity related to trust perturbations.

Main Results:

• The quantum model demonstrated superior predictive performance for evolving human reliability ratings compared to the Markov model.
• The quantum model's effectiveness may stem from its ability to represent within-subject variability in trust judgments.
• A distinct event-related potential (ERP) response in EEG data was identified, correlating with perturbed expectations of AI reliability.

Conclusions:

• Quantum models offer a more promising framework for predicting human reliability judgments in human-AI interactions.
• The identified EEG-based measure provides a potential neural correlate for trust and can be explored for real-time applications.
• This research opens avenues for developing adaptive AI systems that respond to human trust dynamics.