On the Hardness of Computing Counterfactual and Semi-factual Explanations in XAI

Authors: André Artelt, Martin Olsen, Kevin Tierney

TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We provide an overview of the computational complexity results in the literature for generating these explanations, finding that in many cases, generating explanations is computationally hard. We strengthen the argument for this considerably by further contributing our own inapproximability results showing that not only are explanations often hard to generate, but under certain assumptions, they are also hard to approximate. We discuss the implications of these complexity results for the XAI community and for policymakers seeking to regulate explanations in AI.
Researcher Affiliation Academia André Artelt EMAIL Faculty of Technology Bielefeld University, Germany Martin Olsen EMAIL Department of Business Development and Technology Aarhus University, Denmark Kevin Tierney EMAIL Department of Business Decisions and Analytics University of Vienna, Austria
Pseudocode No The paper discusses computational problems and their complexity, including proofs by reduction (e.g., 3-SAT reduction in Appendix A.2). However, it does not present any structured pseudocode or algorithm blocks for any method or procedure.
Open Source Code No The paper does not contain any explicit statement about releasing source code for the described methodology, nor does it provide any links to a code repository. The Open Review link is for the peer review process.
Open Datasets No The paper is theoretical, focusing on computational complexity and inapproximability results. It does not conduct experiments on any specific datasets, thus there is no mention of datasets being publicly available or access information for them.
Dataset Splits No The paper is theoretical and does not involve empirical experiments or dataset evaluation, therefore no information regarding training/test/validation dataset splits is provided.
Hardware Specification No The paper is theoretical and focuses on computational complexity. It does not describe any experimental setup or the specific hardware used to run experiments.
Software Dependencies No The paper is theoretical, analyzing computational complexity. It does not mention any specific software dependencies or versions (e.g., programming languages, libraries, or solvers with version numbers) required for experimental reproduction.
Experiment Setup No The paper is theoretical, presenting computational complexity results and proofs. It does not include details on experimental setup such as hyperparameters, model initialization, or training schedules.