pith. sign in

arxiv: 2605.02318 · v1 · submitted 2026-05-04 · 💻 cs.AI · cs.CE· cs.LG· stat.ML

Can Causal Discovery Algorithms Help in Generating Legal Arguments?

Soham Wasmatkar , Subinay Adhikary , Rakshit Rohan , Shouvik Kumar Guha , Saptarshi Pyne , Kripabandhu Ghosh This is my paper

Pith reviewed 2026-05-08 19:35 UTC · model grok-4.3

classification 💻 cs.AI cs.CEcs.LGstat.ML
keywords causallegalalgorithmsdiscoveryargumentsassaultbeenphysical
0
0 comments X

The pith

Causal discovery applied to annotated homicide cases identifies probabilistic links between legal concepts that can support generation of legal arguments.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The authors took 150 real homicide cases and marked them with whether certain legal ideas like physical assault or property dispute were involved. They then used computer programs designed to find cause-and-effect links in data to see how these legal ideas relate to each other. For example, they found that if there's no physical assault, it's certain that the homicide wasn't due to a property dispute. This suggests the algorithms can spot useful patterns for building arguments in court cases. The work is an initial exploration into bringing causal AI methods into the legal field. The probabilities attached to these relationships give a measure of how reliable the argument might be. This work is an initial attempt to bring advanced AI techniques from causality into legal practice, where decisions often depend on understanding what leads to what. By automating the discovery of these links, it might help in generating arguments more systematically. However, the success depends on how well the algorithms work with the specific nature of legal data, which is often narrative and interpretive rather than purely numerical.

Core claim

some of the causal relationships help generate viable legal arguments, e.g., if one could establish that a physical assault has not taken place during a homicide, it should be a sufficient condition (with probability 1) to establish that the homicide has not been committed due to a property-related dispute.

Load-bearing premise

That the manual annotations of legal concepts in the 150 cases are accurate and complete, and that the causal discovery algorithms applied to this small observational dataset recover true causal relationships rather than spurious correlations or artifacts of annotation.

Figures

Figures reproduced from arXiv: 2605.02318 by Kripabandhu Ghosh, Rakshit Rohan, Saptarshi Pyne, Shouvik Kumar Guha, Soham Wasmatkar, Subinay Adhikary.

Figure 1
Figure 1. Figure 1: A Hypothetical Causal Graph between Three Variables–a Defendant being a Relative (R) of the victim, an heir (H) to the victim’s wealth, and found to be guilty (G). Each of these nodes are assumed to be binary variables, e.g., a defendant can either be a relative (R) of the victim or not (!R). Each directed edge represents a cause-and-effect relationship between two nodes, e.g., the edge R → H indicates tha… view at source ↗
Figure 2
Figure 2. Figure 2: Examples of How Text Spans Within or Across Sentences are Annotated with Legal Concepts. 150 are randomly selected for annotation. These case files are annotated with all of the (7+10) = 17 legal concepts by three legal experts using the legal annotation tool LeDA [13] ( view at source ↗
Figure 3
Figure 3. Figure 3: An Example of How the Inter-Annotator Agreement (IAA) Score is Calculated view at source ↗
Figure 4
Figure 4. Figure 4: The Causal Discovery Algorithm-Aided Workflow for Automated Generation of Legal Arguments. A curated legal corpus of 150 homicide cases are utilized to identify 17 legal concepts. Subsequently, each case is annotated with a legal concept only if the concept is applicable to the case. Based on the presence (0) or absence (1) of the 17 legal concepts in 150 homicide cases, a 150-by-17 binary data matrix is p… view at source ↗
Figure 5
Figure 5. Figure 5: The Consensus Causal Graph. The nodes represent the legal concepts. N1: witness testimony, N2: prosecutorial delay or inability, N3: testimony challenged, N4: riot, N5: death sentence, N6: life imprisonment, N7: homicide murder, N8: evidence inconsistency, N9: expert witness testimony, N10: political rivalry, N11: physical assault, N12: evidence insufficient, N13: investigation agency, N14: revenge, N15: p… view at source ↗
read the original abstract

In 2011, Judea Pearl received the Turing Award, considered the Nobel Prize in Computing, for fundamental contributions to artificial intelligence through the development of a calculus for probabilistic and causal reasoning. It includes pioneering the development of causal discovery algorithms. These computer algorithms can analyze large multivariate datasets and automatically discover the causal relationships among the constituent variables. They have been widely used in many critical fields such as medicine and economics to support decisions. However, to our knowledge, they have not been leveraged in law. This paper attempts to alleviate this gap by investigating whether causal discovery algorithms can be leveraged for automated generation of legal arguments. To that end, a novel legal dataset is prepared by identifying 17 legal concepts, such as physical assault and property dispute. A curated collection of 150 homicide cases are annotated with these concepts, e.g., a case is annotated with physical assault only if a physical assault had been reported in that case. Subsequently, a selected set of widely-used causal discovery algorithms is applied to the annotated dataset to discover the causal relationships between the legal concepts. Additionally, the degrees of belief associated with the discovered relationships are quantified in mathematical probabilities. It is shown that some of the causal relationships help generate viable legal arguments, e.g., if one could establish that a physical assault has not taken place during a homicide, it should be a sufficient condition (with probability 1) to establish that the homicide has not been committed due to a property-related dispute. Thus, this paper shows that causal discovery algorithms can be helpful in generating legal arguments, opening up avenues for promising future endeavors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption that causal discovery algorithms recover meaningful causal structure from binary annotations of legal concepts in case narratives, plus the assumption that those annotations faithfully capture the underlying facts.

axioms (2)
  • domain assumption Causal discovery algorithms can identify causal relationships from observational data under standard assumptions such as faithfulness and no hidden confounders.
    Implicit when applying the algorithms to the annotated dataset without additional controls.
  • domain assumption The 17 legal concepts were annotated accurately and consistently across the 150 cases.
    Required for the input data to support any downstream causal claims.

pith-pipeline@v0.9.0 · 5616 in / 1508 out tokens · 91607 ms · 2026-05-08T19:35:59.473353+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

  • Foundation/Cost (no overlap) Cost.FunctionalEquation.washburn_uniqueness_aczel unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    Six widely-used causal discovery algorithms, namely, PC, GES, GRaSP, BOSS, LiNGAM, and ANM, are selected based on their methodological diversity.

  • No overlap with reality_from_one_distinction, 8-tick period, D=3 forcing, or constants chain. Foundation.RealityFromDistinction.reality_from_one_distinction unclear
    ?
    unclear

    Relation between the paper passage and the cited Recognition theorem.

    Based on the presence (0) or absence (1) of the 17 legal concepts in 150 homicide cases, a 150-by-17 binary data matrix is prepared.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

  1. [1]

    Zanga, E

    A. Zanga, E. Ozkirimli, F. Stella, A survey on causal discovery: theory and practice, International Journal of Approximate Reasoning 151 (2022) 101–129

    work page 2022

  2. [2]

    C. K. Assaad, E. Devijver, E. Gaussier, Survey and evaluation of causal discovery methods for time series, Journal of Artificial Intelligence Research 73 (2022) 767–819

    work page 2022

  3. [3]

    X. Shen, S. Ma, P. Vemuri, G. Simon, Challenges and opportunities with causal discovery algorithms: application to alzheimer’s pathophysiology, Scientific reports 10 (2020) 2975

    work page 2020

  4. [4]

    J. J. Anker, E. Kummerfeld, A. Rix, S. J. Burwell, M. G. Kushner, Causal network modeling of the determinants of drinking behavior in comorbid alcohol use and anxiety disorder, Alcoholism: clinical and experimental research 43 (2019) 91–97

    work page 2019

  5. [5]

    P. M. Addo, C. Manibialoa, F. McIsaac, Exploring nonlinearity on the co2 emissions, economic production and energy use nexus: A causal discovery approach, Energy Reports 7 (2021) 6196–6204

    work page 2021

  6. [6]

    Liepin, a, T

    R. Liepin, a, T. de Lima, E. Lorini, G. Pisano, G. Sartor, A causal model checker for legal cases, in: Proceedings of the Twentieth International Conference on Artificial Intelligence and Law, 2025, pp. 248–257

    work page 2025

  7. [7]

    A. P. Dawid, F. Dotto, M. Graves, J. B. Kadane, J. Mortera, G. Robertson, J. Q. Smith, A. L. Wilson, A comparison of graphical methods using the case of the murder of meredith kercher as an example, Law, Probability and Risk 24 (2025) mgaf002

    work page 2025

  8. [8]

    Dahlman, E

    C. Dahlman, E. Kolflaath, Causal models versus reason models in bayesian networks for legal evidence, Synthese 200 (2022) 477

    work page 2022

  9. [9]

    Liefgreen, D

    A. Liefgreen, D. Lagnado, The role of causal models in evaluating simple and complex legal explanations, in: Proceedings of the Annual Meeting of the Cognitive Science Society, volume 43, 2021

    work page 2021

  10. [10]

    Liepin, a, G

    R. Liepin, a, G. Sartor, A. Wyner, Arguing about causes in law: a semi-formal framework for causal arguments, Artificial intelligence and law 28 (2020) 69–89

    work page 2020

  11. [11]

    Liepina, G

    R. Liepina, G. Sartor, A. Wyner, Causal models of legal cases, in: International Workshop on AI Approaches to the Complexity of Legal Systems, Springer, 2015, pp. 172–186

    work page 2015

  12. [12]

    Adhikary, P

    S. Adhikary, P. Sen, D. Roy, K. Ghosh, A case study for automated attribute extraction from legal documents using large language models, Artificial Intelligence and Law (2024) 1–22

    work page 2024

  13. [13]

    Adhikary, D

    S. Adhikary, D. Roy, D. Ganguly, S. Kumar Guha, K. Ghosh, Leda: a system for legal data annotation, Frontiers in Artificial Intelligence and Applications (2023) 370–367

    work page 2023

  14. [14]

    A. Z. Wyner, W. Peters, D. Katz, A case study on legal case annotation., in: JURIX, 2013, pp. 165–174

    work page 2013

  15. [15]

    Spirtes, C

    P. Spirtes, C. N. Glymour, R. Scheines, Causation, prediction, and search, MIT press, 2000

    work page 2000

  16. [16]

    D. M. Chickering, Optimal structure identification with greedy search, Journal of machine learning research 3 (2002) 507–554

    work page 2002

  17. [17]

    W.-Y. Lam, B. Andrews, J. Ramsey, Greedy relaxations of the sparsest permutation algorithm, in: Uncertainty in Artificial Intelligence, PMLR, 2022, pp. 1052–1062

    work page 2022

  18. [18]

    Andrews, J

    B. Andrews, J. Ramsey, R. Sanchez Romero, J. Camchong, E. Kummerfeld, Fast scalable and accurate discovery of dags using the best order score search and grow shrink trees, Advances in neural information processing systems 36 (2023) 63945–63956

    work page 2023

  19. [19]

    Shimizu, P

    S. Shimizu, P. O. Hoyer, A. Hyvärinen, A. Kerminen, M. Jordan, A linear non-gaussian acyclic model for causal discovery., Journal of Machine Learning Research 7 (2006)

    work page 2006

  20. [20]

    Hoyer, D

    P. Hoyer, D. Janzing, J. M. Mooij, J. Peters, B. Schölkopf, Nonlinear causal discovery with additive noise models, Advances in neural information processing systems 21 (2008)

    work page 2008

  21. [21]

    F. X. Diebold, Elements of forecasting, South-Western College Pub. Cincinnati, OH, USA, 1998

    work page 1998

  22. [22]

    Lee Rodgers, W

    J. Lee Rodgers, W. A. Nicewander, Thirteen ways to look at the correlation coefficient, The American Statistician 42 (1988) 59–66

    work page 1988

  23. [23]

    Pearl, D

    J. Pearl, D. Mackenzie, The book of why: the new science of cause and effect, Basic books, 2018

    work page 2018

  24. [24]

    B. A. Spellman, A. Kincannon, The relation between counterfactual (but for) and causal reasoning: Experimental findings and implicaitons for jurors’ decisions, Law and Contemp. Probs. 64 (2001) 241

    work page 2001

  25. [25]

    S. J. Russell, Judea pearl, 2011. URL: https://amturing.acm.org/award_winners/pearl_2658896.cfm

    work page 2011

  26. [26]

    Y. Yu, L. Hou, X. Liu, S. Wu, H. Li, F. Xue, A novel constraint-based structure learning algorithm using marginal causal prior knowledge, Scientific Reports 14 (2024) 19279

    work page 2024