Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
Decomposing Interventional Causality into Synergistic, Redundant, and Unique Components
Authors: Abel Jansma
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The formalism is then illustrated by decomposing the causal power in logic gates, cellular automata, chemical reaction networks, and a transformer language model. Our results reveal how the distribution of causal power can be context- and parameter-dependent. The decomposition provides new insights into complex systems by revealing how causal influences are shared and combined among multiple variables, with potential applications ranging from attribution of responsibility in legal or AI systems, to the analysis of biological networks or climate models. |
| Researcher Affiliation | Academia | Abel Jansma Dutch Institute for Emergent Phenomena, the Netherlands Institute for Logic, Language and Computation, University of Amsterdam Institute of Physics, University of Amsterdam EMAIL |
| Pseudocode | No | The paper describes a mathematical approach and illustrates it with examples but does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code to reproduce the figures in this section is available at [12]. [12] Abel Jansma. accompanying repository. https://github.com/AJnsm/causal Decomposition, 2025. |
| Open Datasets | No | The paper uses simulated systems (logic gates, cellular automata, chemical networks) and a pre-trained language model (distilbert-base-uncased-finetuned-sst-2-english) with custom-generated input sentences. It does not use or provide any publicly available datasets. |
| Dataset Splits | No | The paper does not use any pre-existing datasets that would require explicit training/test/validation splits. For cellular automata, it mentions 'random initialisation' and 'middle-1 initialisation' as ways to set up initial states, not dataset splits. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU, GPU models, memory amounts) used for running the experiments. The NeurIPS checklist confirms that the experiments are not compute-intensive and could run on any platform. |
| Software Dependencies | No | The paper mentions the use of the 'Transformers Python library [26]' and a specific language model 'distilbert-base-uncased-finetuned-sst-2-english [23]'. However, it does not specify version numbers for Python, the Transformers library, or any other software components. |
| Experiment Setup | Yes | Let the two inputs be independently drawn from a binomial distribution with P(X1 = 1) = P(X2 = 1) = p, and the output be their logical OR, AND, XOR, or COPY. To get empirical estimates, we simulated automata with 100 cells and periodic boundary conditions that evolved over 10k steps, where the first 500 states were discarded. Parameters are set to k1 = 10, k2 = 1, [X1] = [X2] = ϵ = 1. Let the baseline sentence be this movie is . Let an intervention correspond to appending a word to the baseline sentence, and define the causal effect of appending string A to be: CE(A; Y ) = Y ("this movie is" + A) Y ("this movie is") |