Trust toward humans and trust toward artificial intelligence are not associated: Initial insights from self-report and neurostructural brain imaging

Christian Montag, Benjamin Klugah‐Brown, Xinqi Zhou, Jennifer Wernicke, Congcong Liu, Juan Kou, Yuanshu Chen, Brian W. Haas, Benjamin BeckerPersonality 2023 · DOI

Abstract

The present study examines whether self-reported trust in humans and self-reported trust in products with built-in artificial intelligence (AI) are associated with one another and with brain structure. We sampled 90 healthy participants who provided self-reported trust in humans and AI and underwent brain structural magnetic resonance imaging assessment. We found that trust in humans, as measured by the trust facet of the personality inventory NEO-PI-R, and trust in AI products, as measured by items assessing attitudes toward AI and by a composite score based on items assessing trust toward products with in-built AI, were not significantly correlated. We also used a concomitant dimensional neuroimaging approach employing a data-driven source-based morphometry (SBM) analysis of gray-matter-density to investigate neurostructural associations with each trust domain. We found that trust in humans was negatively (and significantly) correlated with an SBM component encompassing striato-thalamic and prefrontal regions. We did not observe significant brain structural association with trust in AI. The present findings provide evidence that trust in humans and trust in AI seem to be dissociable constructs. While the personal disposition to trust in humans might be "hardwired" to the brain's neurostructural architecture, a corresponding significant link for the disposition to trust AI was not observed.

Introduction

As AI moves from science fiction to daily utility—powering everything from Siri to autonomous vehicles—understanding how humans decide to trust these systems is critical. Historically, trust has been studied as an interpersonal phenomenon with deep evolutionary roots, often linked to specific biological pathways and brain regions. This research seeks to answer a fundamental question: Is trust a "general" trait that carries over from humans to machines, or are they separate psychological processes? If they are separate, it suggests that the biological hardwiring that helps us navigate social cooperation does not necessarily dictate how we interact with technology.

Methods

The study employed a dual-layered approach, combining psychometric self-reports with structural neuroimaging in a sample of 90 healthy male participants (mean age ~22) recruited in China.

Psychometrics

Participants completed the NEO-PI-R to measure the "trust" facet of personality and two AI-specific scales: the ATAI (Attitude Toward Artificial Intelligence) and the TDP-AI (Trusting Diverse Products where AI is in-built), which covers products like Alexa, Siri, and human-like androids.

Neuroimaging

Participants underwent 3T MRI scans. The researchers utilized Source-Based Morphometry (SBM), a data-driven multivariate approach. Unlike standard Voxel-Based Morphometry, SBM uses Independent Component Analysis (ICA) to identify clusters of brain regions where gray matter density (GMD) covaries across individuals, providing a sensitive measure for subtle structural patterns related to personality.

Results

Behavioral Dissociation

The study found virtually zero correlation between how much a person trusts other humans and how much they trust AI. As shown in the correlation matrix below, the correlation between human trust (NEO-PI-R) and AI acceptance (ATAI) was non-significant (r = 0.041).

Correlation matrix showing the lack of association between trust in humans and trust in AI.

Neural Signatures

Neuroimaging results further supported this dissociation. SBM identified a specific "Component 1" encompassing the thalamus, striatum (putamen), and right middle frontal cortex. Higher levels of interpersonal trust were significantly associated with lower gray matter density in this component (r = -0.24, p = 0.023). Crucially, no significant associations were found between any brain structural components and trust in AI.

Brain regions associated with interpersonal trust (A) and the negative correlation between human trust and the identified SBM component (B).

Discussion

The findings suggest that interpersonal trust and trust in AI are dissociable constructs. The authors argue that interpersonal trust is an evolutionary survival mechanism designed for social exchange, while AI is a very recent phenomenon that the human brain has not yet developed specialized structural architecture for.

The fact that human trust correlated with brain structure while AI trust did not suggests that interpersonal trust may be more "hardwired" into our neurobiology. Trust in AI might be more "plastic"—driven by cognitive appraisal, media exposure, and direct experience rather than innate disposition. Interestingly, the data did not strongly support the "uncanny valley" theory; trust levels remained relatively stable across different types of AI products, though tools like Siri were trusted slightly more than human-like androids.

Limitations and Future Work

The study was conducted exclusively on male participants in China, a culture with high AI acceptance. Future research should include female participants and diverse cultural samples. Additionally, while this study focused on brain structure, future work could investigate functional connectivity to see if trust in AI is reflected in how brain regions communicate in real-time rather than their physical density.

Summary updated

Abstract

The present study examines whether self-reported trust in humans and self-reported trust in [(different) products with built-in] artificial intelligence (AI) are associated with one another and with brain structure. We sampled 90 healthy participants who provided self-reported trust in humans and AI and underwent brain structural magnetic resonance imaging assessment. We found that trust in humans, as measured by the trust facet of the personality inventory NEO-PI-R, and trust in AI products, as measured by items assessing attitudes toward AI and by a composite score based on items assessing trust toward products with in-built AI, were not significantly correlated. We also used a concomitant dimensional neuroimaging approach employing a data-driven s<i>ource-based morphometry</i> (<i>SBM</i>) analysis of gray-matter-density to investigate neurostructural associations with each trust domain. We found that trust in humans was negatively (and significantly) correlated with an SBM component encompassing striato-thalamic and prefrontal regions. We did not observe significant brain structural association with trust in AI. The present findings provide evidence that trust in humans and trust in AI seem to be dissociable constructs. While the personal disposition to trust in humans might be "hardwired" to the brain's neurostructural architecture (at least from an individual differences perspective), a corresponding significant link for the disposition to trust AI was not observed. These findings represent an initial step toward elucidating how different forms of trust might be processed on the behavioral and brain level.
BibTeX
@inproceedings{montag2023trust,
  author = {Christian Montag and Benjamin Klugah‐Brown and Xinqi Zhou and Jennifer Wernicke and Congcong Liu and Juan Kou and Yuanshu Chen and Brian W. Haas and Benjamin Becker},
  title = {{Trust toward humans and trust toward artificial intelligence are not associated: Initial insights from self-report and neurostructural brain imaging}},
  booktitle = {Personality},
  year = {2023},
  doi = {10.1017/pen.2022.5},
}
Figure