Causal Abstraction: A Theoretical Foundation for Mechanistic Interpretability
Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel pith:HR6OHQBWrecord.jsonopen to challenge →
read the original abstract
Causal abstraction provides a theoretical foundation for mechanistic interpretability, the field concerned with providing intelligible algorithms that are faithful simplifications of the known, but opaque low-level details of black box AI models. Our contributions are (1) generalizing the theory of causal abstraction from mechanism replacement (i.e., hard and soft interventions) to arbitrary mechanism transformation (i.e., functionals from old mechanisms to new mechanisms), (2) providing a flexible, yet precise formalization for the core concepts of polysemantic neurons, the linear representation hypothesis, modular features, and graded faithfulness, and (3) unifying a variety of mechanistic interpretability methods in the common language of causal abstraction, namely, activation and path patching, causal mediation analysis, causal scrubbing, causal tracing, circuit analysis, concept erasure, sparse autoencoders, differential binary masking, distributed alignment search, and steering.
This paper has not been read by Pith yet.
Forward citations
Cited by 18 Pith papers
-
Localizing Model Behavior with Path Patching
Path patching provides a method to express and quantitatively test hypotheses that neural network behaviors are localized to sets of paths.
-
Interpreting Neural Combinatorial Optimization via Evolving Programmatic Bottlenecks
EPB distills NCO models into evolving program portfolios via LLM-driven textual-numerical optimization, matching original performance while exposing stage-dependent heuristic-like behavior.
-
From Mechanistic to Compositional Interpretability
Compositional interpretability defines explanations as commuting syntactic-semantic mapping pairs grounded in compositionality and minimum description length, with compressive refinement and a parsimony theorem guaran...
-
From Mechanistic to Compositional Interpretability
The paper introduces compositional interpretability as a category-theoretic framework that casts mechanistic explanations as commuting syntactic-semantic mappings optimized under faithfulness and complexity constraint...
-
Bucketing the Good Apples: A Method for Diagnosing and Improving Causal Abstraction
A four-step recipe partitions the input space using interchange intervention behavior to diagnose where causal abstractions hold and to guide improvements, demonstrated by recovering a full hypothesis from scratch in ...
-
Safety from Honesty in a Disinterested AI Predictor
A disinterested Bayesian Predictor trained on contextualized statements has low probability of producing harmful agency because dangerous behaviors require rare coordinated underestimation of harm with no training sig...
-
Steering Vision-Language Models with Joint Sparse Autoencoders
JSAE jointly factorizes pooled vision and language activations in VLMs into aligned interpretable features, revealing layer-dependent asymmetry in additive steering versus suppression on three models.
-
XtrAIn: Training-Guided Occlusion for Feature Attribution
XtrAIn shifts occlusion from input space to parameter space along the training trajectory to produce cleaner feature attributions than standard methods.
-
When Behavioral Safety Evaluation Fails: A Representation-Level Perspective
Behavioral safety metrics for LLMs are insufficient because models can maintain safe outputs while remaining vulnerable to latent-space interventions, as shown via dissociated models and the new Latent Vulnerability Score.
-
Probing Minimalist Phase Structure in LLMs: What Universal Dependencies Cannot Represent
Structural probes on UD-invariant wh-movement stimuli reveal phase-count gradients and phase-internal cohesion effects in 12-13 of 13 LLMs, indicating syntactic abstractions beyond UD annotations.
-
From Weight Perturbation to Feature Attribution for Explaining Fully Connected Neural Networks
XWP and XWP_c are novel attribution methods for FCNNs that estimate feature importance by perturbing attached weights to avoid added bias and out-of-distribution issues in occlusion approaches.
-
Correcting Influence: Unboxing LLM Outputs with Orthogonal Latent Spaces
A latent mediation framework with sparse autoencoders enables non-additive token-level influence attribution in LLMs by learning orthogonal features and back-propagating attributions.
-
Sparse Feature Circuits: Discovering and Editing Interpretable Causal Graphs in Language Models
Sparse feature circuits are introduced as interpretable causal subnetworks in language models, supporting unsupervised discovery of thousands of circuits and a method called SHIFT to improve classifier generalization ...
-
Linear Representations of Sentiment in Large Language Models
Sentiment is represented as a single linear direction in LLM activation space that is causally relevant across tasks and is summarized at punctuation and names in addition to charged words.
-
Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces
Reasoning in large output spaces proceeds via shortlisting then fine-grained reasoning; this characterization enables a mechanistic distillation strategy that outperforms standard distillation.
-
Causal Tongue-Tie: LLMs Can Encode Causal Direction, But Their Yes/No Outputs Fail to Express
LLMs encode causal direction internally via probes but revert to commonsense in Yes/No outputs on anti-commonsense items, showing output accuracy alone does not measure causal understanding.
-
Temporal Preference Concepts and their Functions in a Large Language Model
Causal localization via attribution and patching identifies a temporal preference subgraph in mid-to-upper layers of Qwen3-4B-Instruct-2507, with time-horizon geometry in the residual stream and initial evidence for s...
-
LLMs Should Not Yet Be Credited with Decision Explanation
LLMs support decision prediction and rationale generation but lack evidence for genuine decision explanation, requiring stricter standards to avoid over-crediting.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.