ETD

• AOT
• Cognitive Load
• Confirmation Bias
• Conversational-AI
• Human-Computer Interaction
• Opinionated AI system
• Selective Exposure
• Trust in LLM

Large Language Models (LLMs) are increasingly adopted as substitutes for traditional search engines, providing users with synthesized answers rather than ranked lists of sources. While this shift offers efficiency and accessibility, it raises concerns about the amplification of cognitive biases such as selective exposure and confirmation bias, which in turn may foster echo chamber effects.
This thesis investigates the relationship between user trust in LLMs, cognitive load, and confirmation bias in information-seeking tasks. Building on

We designed and conducted a controlled experiment with 51 participants, examining how belief-consonant responses from conversational systems influence users’ information querying, cognitive load, and trust in the system. We further explored the moderating role of individual cognitive traits, such as Actively Open-Minded Thinking (AOT), in mitigating these dynamics.
These results contribute to the interdisciplinary discourse at the intersection of computer science, psychology, and human-computer interaction, providing empirical evidence on how belief-aligned AI systems can shape information behavior.