# Transformers Use Causal World Models in Maze-Solving Tasks

arXiv: https://arxiv.org/abs/2412.11867 · HTML: https://mlsanity.com/paper/2412.11867/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- Placed in World models for language agents, one of 11 directions in World models.
- Builds on 20 of its 28 references that are in the map.
- First posted on arXiv 16 Dec 2024.

## Placed in

- World models / [World models for language agents](https://mlsanity.com/wm/world-models-for-language-agents/)
- Mechanistic interpretability / [Circuits and mechanistic analysis](https://mlsanity.com/mechinterp/circuits-and-mechanistic-analysis/)

## Builds on (most cited first, 20 of 20)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [Sparse Autoencoders Find Highly Interpretable Features in Language Models](https://mlsanity.com/paper/2309.08600/) (2023)
- [Understanding intermediate layers using linear classifier probes](https://mlsanity.com/paper/1610.01644/) (2016)
- [Toy Models of Superposition](https://mlsanity.com/paper/2209.10652/) (2022)
- [Eliciting Latent Predictions from Transformers with the Tuned Lens](https://mlsanity.com/paper/2303.08112/) (2023)
- [Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task](https://mlsanity.com/paper/2210.13382/) (2022)
- [Does Circuit Analysis Interpretability Scale? Evidence from Multiple Choice Capabilities in Chinchilla](https://mlsanity.com/paper/2307.09458/) (2023)
- [Toward Transparent AI: A Survey on Interpreting the Inner Structures of Deep Neural Networks](https://mlsanity.com/paper/2207.13243/) (2022)
- [Acquisition of Chess Knowledge in AlphaZero](https://mlsanity.com/paper/2111.09259/) (2021)
- [Future Lens: Anticipating Subsequent Tokens from a Single Hidden State](https://mlsanity.com/paper/2311.04897/) (2023)
- [Evaluating Cognitive Maps and Planning in Large Language Models with CogEval](https://mlsanity.com/paper/2309.15129/) (2023)
- [Emergent World Models and Latent Variable Estimation in Chess-Playing Language Models](https://mlsanity.com/paper/2403.15498/) (2024)
- [Measuring Progress in Dictionary Learning for Language Model Interpretability with Board Game Models](https://mlsanity.com/paper/2408.00113/) (2024)
- [A Mechanistic Analysis of a Transformer Trained on a Symbolic Multi-Step Reasoning Task](https://mlsanity.com/paper/2402.11917/) (2024)
- [Evidence of Learned Look-Ahead in a Chess-Playing Neural Network](https://mlsanity.com/paper/2406.00877/) (2024)
- [A Configurable Library for Generating and Manipulating Maze Datasets](https://mlsanity.com/paper/2309.10498/) (2023)
- [Dictionary Learning Improves Patch-Free Circuit Discovery in Mechanistic Interpretability: A Case Study on Othello-GPT](https://mlsanity.com/paper/2402.12201/) (2024)
- [A Philosophical Introduction to Language Models - Part II: The Way Forward](https://mlsanity.com/paper/2405.03207/) (2024)
- [Sparse Relational Reasoning with Object-Centric Representations](https://mlsanity.com/paper/2207.07512/) (2022)

## Built on by (most cited first, 13 of 13)

- [What Has a Foundation Model Found? Using Inductive Bias to Probe for World Models](https://mlsanity.com/paper/2507.06952/) (2025)
- [Jacobian Sparse Autoencoders: Sparsify Computations, Not Just Activations](https://mlsanity.com/paper/2502.18147/) (2025)
- [Outcome-Based RL Provably Leads Transformers to Reason, but Only With the Right Data](https://mlsanity.com/paper/2601.15158/) (2026)
- [The Observer Effect in World Models: Invasive Adaptation Corrupts Latent Physics](https://mlsanity.com/paper/2602.12218/) (2026)
- [From Next Token Prediction to (STRIPS) World Models](https://mlsanity.com/paper/2509.13389/) (2025)
- [Do LLMs Build Spatial World Models? Evidence from Grid-World Maze Tasks](https://mlsanity.com/paper/2604.10690/) (2026)
- [Hierarchical Experimentalist Agents](https://mlsanity.com/paper/2606.29315/) (2026)
- [Interaction Locality in Hierarchical Recursive Reasoning](https://mlsanity.com/paper/2605.20784/) (2026)
- [What if Othello-Playing Language Models Could See?](https://mlsanity.com/paper/2507.14520/) (2025)
- [Causal Masking on Spatial Data: An Information-Theoretic Case for Learning Spatial Datasets with Unimodal Language Models](https://mlsanity.com/paper/2510.27009/) (2025)
- [VERSE: Visual Embedding Reduction and Space Exploration. Clustering-Guided Insights for Training Data Enhancement in Visually-Rich Document Understanding](https://mlsanity.com/paper/2601.05125/) (2026)
- [A Co-Evolutionary Theory of Human-AI Coexistence: Mutualism, Governance, and Dynamics in Complex Societies](https://mlsanity.com/paper/2604.22227/) (2026)
- [World Modeling in Transformers](https://mlsanity.com/paper/2609.21748/) (2026)

## Abstract (arXiv)

Recent studies in interpretability have explored the inner workings of transformer models trained on tasks across various domains, often discovering that these networks naturally develop highly structured representations. When such representations comprehensively reflect the task domain's structure, they are commonly referred to as "World Models" (WMs). In this work, we identify WMs in transformers trained on maze-solving tasks. By using Sparse Autoencoders (SAEs) and analyzing attention patterns, we examine the construction of WMs and demonstrate consistency between SAE feature-based and circuit-based analyses. By subsequently intervening on isolated features to confirm their causal role, we find that it is easier to activate features than to suppress them. Furthermore, we find that models can reason about mazes involving more simultaneously active features than they encountered during training; however, when these same mazes (with greater numbers of connections) are provided to models via input tokens instead, the models fail. Finally, we demonstrate that positional encoding schemes appear to influence how World Models are structured within the model's residual stream.