From Causal Plausibility to Causal Reliability: Evaluating LLMs as Calibrated Direct Causal-Edge Classifiers
From Causal Plausibility to Causal Reliability: Evaluating LLMs as Calibrated Direct Causal-Edge Classifiers
从因果合理性到因果可靠性：评估大语言模型作为校准后的直接因果边分类器
Large language models (LLMs) are increasingly used to provide prior causal knowledge for structural causal discovery, yet whether their direct-edge judgments and confidence can be trusted remains unclear. We systematically evaluate 12 instruction-tuned open-weight models across six benchmark causal graphs, five prompting strategies, and four confidence sources: verbalized, logit-based, cross-prompt agreement, and cross-model agreement.
大语言模型（LLMs）正越来越多地被用于为结构因果发现提供先验因果知识，但其对直接因果边的判断及其置信度是否值得信赖尚不明确。我们系统地评估了 12 个经过指令微调的开源权重模型，涵盖了六个基准因果图、五种提示策略以及四种置信度来源：口头表达置信度、基于 Logit 的置信度、跨提示一致性以及跨模型一致性。
Under our language-only pairwise protocol, our evaluation yields three key findings. (i) LLM-based causal judgments are strongly recall-dominant: models predict overly dense graphs with many false-positive edges, while prompting mainly shifts the precision-recall trade-off rather than resolving overprediction. Gains from model scale diminish on the largest graphs and do not eliminate miscalibration.
在我们的纯语言成对协议下，评估得出了三个关键发现。（i）基于 LLM 的因果判断具有强烈的召回导向：模型倾向于预测过于稠密的图，包含许多假阳性边；而提示策略主要是在改变精确率与召回率之间的权衡，而非解决过度预测问题。模型规模带来的增益在大型图上会减弱，且无法消除校准偏差。
(ii) LLMs often capture causal relatedness without reliably identifying directness or orientation. Relative to published reference graphs, models misclassify 40.0% of indirect and 36.0% of reversed non-edges as direct edges, versus 28.2% of other non-edges. Moreover, 80.8% and 84.6% of these false positives receive verbalized confidence of at least 80%, revealing substantial overconfidence in structurally incorrect predictions.
（ii）LLMs 往往能捕捉到因果相关性，却无法可靠地识别因果的直接性或方向性。相对于已发表的参考图，模型将 40.0% 的间接边和 36.0% 的反向非边错误地分类为直接边，而其他非边的误判率为 28.2%。此外，这些假阳性结果中，有 80.8% 和 84.6% 的口头表达置信度至少达到 80%，这揭示了模型在结构性错误预测上存在严重的过度自信。
(iii) Conventional confidence estimates are unreliable, whereas agreement offers a more promising signal. Logit-based confidence frequently collapses near 1.0 regardless of correctness, while cross-prompt and cross-model agreement achieve better mean calibration and discrimination, though their advantages are not statistically significant after Holm correction. A benchmark-familiarity audit further identifies potential familiarity in five model-dataset pairs, all involving AsiaM. Overall, our results suggest LLMs are better viewed as sources of externally validated soft causal priors than as direct evidence of causal structure.
（iii）传统的置信度估计不可靠，而一致性指标提供了更有前景的信号。基于 Logit 的置信度往往无论正确与否都趋近于 1.0，而跨提示和跨模型的一致性在平均校准和区分度上表现更好，尽管在 Holm 校正后，这些优势在统计学上并不显著。基准熟悉度审计进一步识别出五个模型-数据集对中存在潜在的熟悉度问题，且均涉及 AsiaM 数据集。总体而言，我们的研究结果表明，将 LLMs 视为经外部验证的软因果先验来源，比将其视为因果结构的直接证据更为合适。