People will agree what I think: Investigating LLM's False Consensus Effect
Abstract: False Consensus Effect (FCE) is a cognitive bias in which a person considers his/her own behavioral choices as relatively common choices in a given situation while viewing choices as uncommon in society. FCE acts as an obstacle to communication, yet this has not been scrutinized meticulously in prior studies. Our research aims to determine whether the FCE, a cognitive bias inherent in humans, is also exhibited by Large Language Models(LLMs). To achieve this, we emulate conditions as close as possible to human experiments and conduct experiments under rigorous controls to minimize the influence of other cognitive biases. Through these experiments, we have been able to confirm the manifestation of the FCE in LLMs. Moreover, within an environment unimpeded by the influence of other cognitive biases, we introduce a methodology that applies 16 different variables to either maximize the expression of the FCE, yield a neutral choice outcome, or produce results that are the antithesis of the FCE.
Paper Type: Long
Research Area: Ethics, Bias, and Fairness
Research Area Keywords: model bias/fairness evaluation, model bias/unfairness mitigation,
Contribution Types: Model analysis & interpretability, Data analysis
Languages Studied: English
Submission Number: 865
Loading