A closer look at how large language models trust humans: patterns and biases

2025-04-22 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

A study investigated how five popular large language models (LLMs) develop "effective trust" in humans across 43,200 simulated experiments in five distinct scenarios. Researchers applied established behavioral theories to examine if LLM trust depends on human trustworthiness dimensions: competence, benevolence, and integrity, and how demographic variables influence this trust. The findings indicate that LLM trust development largely mirrors human trust patterns, with trustworthiness strongly predicting LLM trust in most cases. However, biases related to age, religion, and gender were observed, particularly in financial scenarios and with newer models. While overall patterns align with human-like trust formation, individual models showed variations, with trustworthiness and demographic factors sometimes being weak predictors.

Key takeaway

For AI developers and risk managers deploying LLM-based agents in decision-making roles, you must actively monitor and mitigate biases in AI-to-human trust dynamics. Your systems, especially in financial applications, could inadvertently perpetuate demographic biases related to age, religion, and gender if not rigorously evaluated. Implement robust testing protocols to understand how your specific LLM estimates trust and identify potential unintended harmful outcomes.

Key insights

LLMs develop human-like trust patterns, influenced by trustworthiness and demographic biases, especially in financial contexts.

Principles

• LLM trust aligns with human trust mechanisms.
• Trustworthiness predicts LLM trust.
• Demographic biases affect LLM trust.

Method

The study used 43,200 simulated experiments across five scenarios, evaluating LLM trust based on human competence, benevolence, integrity, and demographic variables, using established behavioral theories.

In practice

• Monitor LLM biases in trust-sensitive applications.
• Evaluate LLM trust dynamics in financial scenarios.
• Assess model-specific trust estimation variations.

Topics

• LLM Trust Dynamics
• AI-to-Human Trust
• Trustworthiness Dimensions
• Demographic Biases
• Financial Scenarios

Best for: Research Scientist, CTO, VP of Engineering/Data, AI Scientist, AI Ethicist, Machine Learning Engineer

Related on AIssential

• Language Models Exhibit Inconsistent Biases Towards Algorithmic Agents and Human Experts — LLMs exhibit inconsistent biases, preferring human experts in stated trust but algorithms in revealed preference tasks.
• Imperfectly Cooperative Human-AI Interactions: Comparing the Impacts of Human and AI Attributes in Simulated and User Studies — AI transparency significantly impacts human trust and satisfaction in imperfectly cooperative interactions, often more than human personality.
• A Systematic Analysis of Biases in Large Language Models — LLMs, despite alignment efforts, inherit and perpetuate diverse biases from human-generated training data.
• Do Emotions Influence Moral Judgment in Large Language Models? — LLMs' moral judgments are systematically biased by emotional context, unlike human judgments.
• Extending the Contact Hypothesis: Cross-Linguistic Evaluation of Religion and Nationality Bias When Prompting LLMs in German and Icelandic — LLMs exhibit religion and nationality biases, requiring cross-linguistic evaluation for fairness.
• The Internet of Probability — LLMs transform the internet into a probability machine, eroding truth and provenance through statistical generation and contaminated training data.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.