Do LLMs Share Human-Like Biases? Causal Reasoning Under Prior Knowledge, Irrelevant Context, and Varying Compute Budgets

TL;DR

This paper evaluates whether large language models (LLMs) perform causal reasoning similarly to humans, finding they often rely on rule-like shortcuts and are less susceptible to certain human biases, with robustness improved by chain-of-thought prompting.

Contribution

It provides a comprehensive benchmark of 20+ LLMs on causal judgment tasks, revealing their reasoning strategies and robustness under various conditions, and compares them to human judgments.

Findings

Most LLMs exhibit rule-like causal reasoning rather than human-like biases.

Chain-of-thought prompting improves LLM robustness to irrelevant context.

A small interpretable model effectively summarizes LLMs' causal judgments.

Abstract

Large language models (LLMs) are increasingly used in domains where causal reasoning matters, yet it remains unclear whether their judgments reflect normative causal computation, human-like shortcuts, or brittle pattern matching. We benchmark 20+ LLMs against a matched human baseline on 11 causal judgment tasks formalized by a collider structure ($C_1 \rightarrow E \leftarrow C_2$). We find that a small interpretable model compresses LLMs' causal judgments well and that most LLMs exhibit more rule-like reasoning strategies than humans who seem to account for unmentioned latent factors in their probability judgments. Furthermore, most LLMs do not mirror the characteristic human collider biases of weak explaining away and Markov violations. We probe LLMs' causal judgment robustness under (i) semantic abstraction and (ii) prompt overloading (injecting irrelevant text), and find that chain-of-thought (CoT) increases robustness for many LLMs. Together, this divergence suggests LLMs can complement humans when known biases are undesirable, but their rule-like reasoning may break down when uncertainty is intrinsic - highlighting the need to characterize LLM reasoning strategies for safe, effective deployment.