Modeling the quantum-like dynamics of human reliability ratings in Human-AI interactions by interaction dependent Hamiltonians

TL;DR

The paper addresses modeling fluctuating trust in human–AI interactions. It adopts a quantum random walk framework with interaction-dependent Hamiltonians to capture alignment/misalignment effects on reliability judgments. A 10-level cognitive state vector is initialized from a Gaussian around the initial rating, evolves under trial-time steps scaled by gamma, and collapses at block boundaries into a new initialization. Three Hamiltonians corresponding to match, mismatch, and no-response guide the diffusion, with parameters alpha, beta, sigma, gamma optimized by MAE across participants, yielding good fits (MAE ~ 0.3–0.6) and notable cross-participant variation, particularly in gamma/gradient terms while sigma remains relatively stable. The results demonstrate the potential of QRW dynamics to model trust fluctuations in H-AI interactions and point to EEG-guided real-time control as a promising future direction.

Abstract

As our information environments become ever more powered by artificial intelligence (AI), the phenomenon of trust in a human's interactions with this intelligence is becoming increasingly pertinent. For example, in the not too distant future, there will be teams of humans and intelligent robots involved in dealing with the repercussions of high-risk disaster situations such as hurricanes, earthquakes, or nuclear accidents. Even in such conditions of high uncertainty, humans and intelligent machines will need to engage in shared decision making, and trust is fundamental to the effectiveness of these interactions. A key challenge in modeling the dynamics of this trust is to provide a means to incorporate sensitivity to fluctuations in human trust judgments. In this article, we explore the ability of Quantum Random Walk models to model the dynamics of trust in human-AI interactions, and to integrate a sensitivity to fluctuations in participant trust judgments based on the nature of the interaction with the AI. We found that using empirical parameters to inform the use of different Hamiltonians can provide a promising means to model the evolution of trust in Human-AI interactions.

Figures (7)

• Figure 1: Markov and quantum models. The top depicts a Markov model. The black dots represent definite cognitive states associated with a reliability rating. The human agent currently inhabits the cognitive state associated with a reliability rating of 2. In a RW walk model, there can be a transition to adjacent states or back to the current state. Probabilities are associated with these transitions. The quantum model (bottom) illustrates that the human agent simultaneously inhabits all cognitive basis states which are like wave peaks. The concentric circles represent the amplitude of the wave peaks at each basis state, which represent probabilities. Wave energy transfers to neighbouring wave peaks.
• Figure 2: Variation of reliability ratings of AI for three participants
• Figure 3: Results of in silico experiments showing the dynamics of the cognitive state amplitudes under different Hamiltonians; Each Panel shows the initial state vector S0 (left vertical bar graph); the temporal evolution of the state propensities for each of the states (central 10 graphs); and finally the modeled reliability rating (bottom single graph). Panels A and B show the evolution using the positive Hamiltonian; Panels C and D with the central Hamiltonian; and Panels E and F with the negative Hamiltonian. The Hamiltonian used impacts which of the states are being explored.
• Figure 4: Reliability modeled with the Quantum Random Walk Model, adopted to our WoZ Trust experiment. Panel A: The cognitive state vector is centred around the initial reliability rating reported by the participant; the mix of Hamiltonians (positive, negative and neutral) affect the evolution of the state throughout the first block. After finishing 28 trials, the participant rates the AI reliability, the state collapses and is 'frozen' until the start of the second block. The difference between the modeled and human-provided reliability rating (blue line) is used to find optimal model parameters for this participant. Panel B: the full experiment has 20 blocks of 28 trials each; these 20 blocks are divided into three parts with short breaks.
• Figure 5: Goodness of fit for all participants, characterized by the mean absolute error. Each participant is marked. The MAE lies between 0.3 and 0.6 (corresponding to 3 and 6 percent on the reliability scape).