[Paper Review] From OECD to India: Exploring cross-cultural differences in perceived trust, responsibility and reliance of AI and human experts

Vishakha Agrawal, Serhiy Kandul+2 more|arXiv (Cornell University)|Jul 28, 2023

Ethics and Social Impacts of AI4 citations

TL;DR

This cross-cultural study compares trust, responsibility, and reliance on AI versus human experts between OECD and Indian participants using a task-based simulation. Results show OECD participants trust AI more but assign less responsibility to it; Indians trust humans more and assign equal, high responsibility to both AI and human experts, highlighting cultural divergence in moral accountability for AI systems.

ABSTRACT

AI is getting more involved in tasks formerly exclusively assigned to humans. Most of research on perceptions and social acceptability of AI in these areas is mainly restricted to the Western world. In this study, we compare trust, perceived responsibility, and reliance of AI and human experts across OECD and Indian sample. We find that OECD participants consider humans to be less capable but more morally trustworthy and more responsible than AI. In contrast, Indian participants trust humans more than AI but assign equal responsibility for both types of experts. We discuss implications of the observed differences for algorithmic ethics and human-computer interaction.

Motivation & Objective

• To investigate cross-cultural differences in trust, responsibility, and reliance on AI versus human experts between OECD and Indian populations.
• To address the lack of representation of Indian and Global South samples in AI perception research, which is predominantly based on Western data.
• To examine how cultural dimensions—particularly individualism vs. collectivism—shape perceptions of moral responsibility in human-AI collaboration.
• To test whether task-based interaction reveals different patterns than self-report scales in assessing trust and reliance on AI.
• To contribute to algorithmic ethics and human-computer interaction by identifying culturally contingent patterns in AI accountability.

Proposed method

• An interactive, scenario-based simulation game was designed where participants made decisions under risk and uncertainty, choosing between AI and human expert advice.
• Participants were assigned to either OECD or Indian samples and completed four missions involving 'save' or 'kill' decisions with expert recommendations.
• Mixed-effects linear regression models were used to analyze reliance, trust, and responsibility, with fixed effects for mission, scenario, and expert type, and random intercepts for individuals.
• The model included interaction terms for culture, expert type, and scenario to test for differential reliance and responsibility across groups.
• Standardized scales measured perceived trust, moral trust, and responsibility toward AI, human experts, and AI developers.
• The experimental design ensured that human participants retained final decision authority, simulating real-world human-AI collaboration with override capability.

Experimental results

Research questions

• RQ1 How do perceptions of trust in AI and human experts differ between OECD and Indian participants?
• RQ2 What are the cross-cultural differences in perceived responsibility for AI and human experts, including AI developers?
• RQ3 How does reliance on AI versus human advice vary across cultures and decision scenarios?
• RQ4 To what extent do cultural dimensions such as individualism-collectivism shape responsibility attribution in human-AI teams?
• RQ5 Does the presence of a human 'last word' in decision-making reduce responsibility gaps for AI in Western samples?

Key findings

• OECD participants trusted AI more than human experts in terms of capability, but perceived humans as more morally trustworthy and responsible.
• Indian participants trusted human experts more than AI, but assigned equal responsibility to both AI and human experts, with mean responsibility scores significantly above the midpoint of the scale.
• There was no significant difference in reliance on AI versus human advice in either culture, though OECD participants slightly reduced reliance in 'save' scenarios.
• Reliance on expert advice increased in later missions among OECD participants but remained stable among Indian participants.
• Indian participants attributed significantly higher responsibility to AI and its developers than OECD participants, who showed no significant responsibility attribution to AI or its producers.
• The interaction between culture and scenario type revealed that cultural differences in reliance patterns were minimal, but responsibility attribution diverged markedly, especially toward AI and its creators.