A Comprehensive Evaluation of Cognitive Biases in LLMs

2024-10-20

Proceedings of the 5th International Conference on Natural Language Processing for Digital Humanities

AI Summary

This study investigates whether large language models (LLMs) exhibit human-like cognitive biases—systematic deviations from rational judgment—across 30 canonical bias categories. Method: We propose the first general-purpose, automated cognitive bias evaluation framework, integrating prompt-driven controllable scenario generation, structured test-case design, multi-model batch response collection, and systematic bias annotation and analysis. We construct an open-source benchmark dataset comprising 30,000 annotated samples. Contribution/Results: Empirical evaluation of 20 state-of-the-art LLMs reveals that all models exhibit at least one bias category, and all 30 biases are statistically substantiated. Significant inter-model variation in bias prevalence and severity is observed. This work delivers the first large-scale empirical evidence base and open evaluation infrastructure for LLM alignment, bias mitigation, and trustworthy AI development.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Bias, Fairness & Privacy

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

We present a large-scale evaluation of 30 cognitive biases in 20 state-of-the-art large language models (LLMs) under various decision-making scenarios. Our contributions include a novel general-purpose test framework for reliable and large-scale generation of tests for LLMs, a benchmark dataset with 30,000 tests for detecting cognitive biases in LLMs, and a comprehensive assessment of the biases found in the 20 evaluated LLMs. Our work confirms and broadens previous findings suggesting the presence of cognitive biases in LLMs by reporting evidence of all 30 tested biases in at least some of the 20 LLMs. We publish our framework code to encourage future research on biases in LLMs: https://github.com/simonmalberg/cognitive-biases-in-llms

Problem

• Evaluates cognitive biases in large language models
• Develops test framework for detecting model biases
• Assesses 30 biases across 20 LLMs systematically

Innovation

• Novel test framework for large-scale LLM evaluation
• Benchmark dataset with 30,000 cognitive bias tests
• Comprehensive assessment of 30 biases across 20 LLMs