Search Results (3)

Large Language Models (LLMs) are increasingly deployed in decision-support systems across high-stakes domains, yet their susceptibility to cognitive biases—systematic deviations from rational judgment well-documented in human psychology—remains poorly understood in quantitative terms. Existing studies typically examine a narrow set of biases, test a single model family, and rely on qualitative assessments of bias presence. In this work, we present a rigorous experimental framework, inspired by the methodology of experimental physics, for the systematic quantitative measurement of cognitive biases in LLMs. We introduce the Bias Strength Index (BSI), a normalized metric with associated confidence intervals that quantifies the magnitude of bias on a continuous scale, and we decompose the total uncertainty into statistical and systematic components—the latter arising from prompt reformulation. We evaluate a comprehensive taxonomy of eleven cognitive biases (including anchoring, framing effect, confirmation bias, availability heuristic, sunk cost fallacy, bandwagon effect, status quo bias, and others) across eight state-of-the-art LLMs from seven families: GPT-4.1 Mini, Claude 3.5 Sonnet, Gemini 2.5 Flash, Llama 3.3 70B, Llama 3.1 8B, Mistral Large (mistral-large-2411), DeepSeek V3, and MiniMax M2.5. Each bias is probed through multiple semantically equivalent prompt variants, with N = 100 independent trials per configuration, yielding a dataset of over 70,000 model responses. Our results reveal that all tested models exhibit non-zero bias effects for multiple bias categories, though with markedly different profiles. A trial-level Generalized Linear Mixed-Effects Model (GLMM) analysis finds statistically significant bias effects in 27 of 43 testable bias–model combinations (62.8%) after multiple-comparison correction, while a more conservative variant-level test—which requires effects to generalize across prompt formulations—yields only one significant result, highlighting the dominant role of prompt-induced systematic uncertainty. Framing and primacy/recency effects are near-universal, while susceptibility to other biases varies substantially across model families. We further evaluate three debiasing strategies—zero-shot chain-of-thought, adversarial counter-prompting, and role-based prompting—applied at inference time without modifying model weights. Our findings provide a quantitative foundation for auditing cognitive biases in LLMs and highlight the bias-dependent effectiveness of prompt-based debiasing techniques. Full article

Background/Objectives: Sunk cost bias—continuing failing endeavours due to prior investments—is among the most studied decision-making biases. Despite decades of vignette-based research, these measures lack systematic psychometric validation. We examined whether widely-used sunk cost scenarios reliably measure the same psychological construct. Methods: Across two experiments (N = 395), we tested established sunk cost vignettes, including classic scenarios from Arkes and Blumer (1985). English-speaking participants from Prolific Academic completed vignettes alongside cognitive reflection and social desirability measures. We assessed internal consistency and intercorrelations between scenarios. Results: Internal consistency was consistently poor (ω = 0.14–0.57) with weak intercorrelations between scenarios. Even highly similar vignettes correlated only moderately. External validity was problematic, showing inconsistent relationships with cognitive reflection and social desirability across vignettes. Conclusions: These measurement failures have critical implications for neuroimaging research, where unreliable behavioural measures may be mistaken for genuine neural differences. The field needs systematic categorization of scenarios to identify which vignettes engage specific psychological processes and neural circuits, enabling more targeted theoretical development. Full article

Due to the randomness of participants’ movement and the selfishness and dishonesty of individuals in crowdsensing, the quality of the sensing data collected by the server platform is uncertain. Therefore, it is necessary to design a reasonable incentive mechanism in crowdsensing to ensure the stability of the sensing data quality. Most of the existing incentive mechanisms for data quality in crowdsensing are based on traditional economics, which believe that the decision of participants to complete a task depends on whether the benefit of the task is greater than the cost of completing the task. However, behavioral economics shows that people will be affected by the cost of investment in the past, resulting in decision-making bias. Therefore, different from the existing incentive mechanism researches, this paper considers the impact of sunk cost on user decision-making. An incentive mechanism based on sunk cost called IMBSC is proposed to motivate participants to improve data quality. The IMBSC mechanism stimulates the sunk cost effect of participants by designing effort sensing reference factor and withhold factor to improve their own data quality. The effectiveness of the IMBSC mechanism is verified from three aspects of platform utility, participant utility and the number of tasks completed through simulation experiments. The simulation results show that compared with the system without IMBSC mechanism, the platform utility is increased by more than 100%, the average utility of participants is increased by about 6%, and the task completion is increased by more than 50%.  Full article