Research

Paper

AI LLM March 04, 2026

Causality Elicitation from Large Language Models

Authors

Takashi Kameyama, Masahiro Kato, Yasuko Hio, Yasushi Takano, Naoto Minakawa

Abstract

Large language models (LLMs) are trained on enormous amounts of data and encode knowledge in their parameters. We propose a pipeline to elicit causal relationships from LLMs. Specifically, (i) we sample many documents from LLMs on a given topic, (ii) we extract an event list from from each document, (iii) we group events that appear across documents into canonical events, (iv) we construct a binary indicator vector for each document over canonical events, and (v) we estimate candidate causal graphs using causal discovery methods. Our approach does not guarantee real-world causality. Rather, it provides a framework for presenting the set of causal hypotheses that LLMs can plausibly assume, as an inspectable set of variables and candidate graphs.

Metadata

arXiv ID: 2603.04276
Provider: ARXIV
Primary Category: cs.LG
Published: 2026-03-04
Fetched: 2026-03-05 06:06

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.04276v1</id>\n    <title>Causality Elicitation from Large Language Models</title>\n    <updated>2026-03-04T16:58:10Z</updated>\n    <link href='https://arxiv.org/abs/2603.04276v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.04276v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Large language models (LLMs) are trained on enormous amounts of data and encode knowledge in their parameters. We propose a pipeline to elicit causal relationships from LLMs. Specifically, (i) we sample many documents from LLMs on a given topic, (ii) we extract an event list from from each document, (iii) we group events that appear across documents into canonical events, (iv) we construct a binary indicator vector for each document over canonical events, and (v) we estimate candidate causal graphs using causal discovery methods. Our approach does not guarantee real-world causality. Rather, it provides a framework for presenting the set of causal hypotheses that LLMs can plausibly assume, as an inspectable set of variables and candidate graphs.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='econ.EM'/>\n    <published>2026-03-04T16:58:10Z</published>\n    <arxiv:primary_category term='cs.LG'/>\n    <author>\n      <name>Takashi Kameyama</name>\n    </author>\n    <author>\n      <name>Masahiro Kato</name>\n    </author>\n    <author>\n      <name>Yasuko Hio</name>\n    </author>\n    <author>\n      <name>Yasushi Takano</name>\n    </author>\n    <author>\n      <name>Naoto Minakawa</name>\n    </author>\n  </entry>"
}