Open Problems in Mechanistic Interpretability

Lee Sharkey, Bilal Chughtai, Joshua Batson, Jack Lindsey, Jeff Wu +24 more

1/27/2025

cs.LG

No official code found. Best implementation: bgsbgsbgs42/Investigating-Superposition-and-Spurious-Correlations-in-Small-Transformer-Models (0 stars, last commit 2025-06-13).

Abstract

Mechanistic interpretability aims to understand the computational mechanisms underlying neural networks' capabilities in order to accomplish concrete scientific and engineering goals. Progress in this field thus promises to provide greater assurance over AI system behavior and shed light on exciting scientific questions about the nature of intelligence. Despite recent progress toward these goals, there are many open problems in the field that require solutions before many scientific and practical benefits can be realized: Our methods require both conceptual and practical improvements to reveal deeper insights; we must figure out how best to apply our methods in pursuit of specific goals; and the field must grapple with socio-technical challenges that influence and are influenced by our work. This forward-facing review discusses the current frontier of mechanistic interpretability and the open problems that the field may benefit from prioritizing.

View on arXiv View PDF

Code Implementations(1)

bgsbgsbgs42/Investigating-Superposition-and-Spurious-Correlations-in-Small-Transformer-ModelsCommunity61%

Exploring the open problems in mechanistic interpretability of (1) the phenomenon of superposition in small transformer models, and (2) the emergence and encoding of spurious correlations in simple neural networks.

00Python, Jupyter NotebookFeb 8, 20251 years ago

Similar Papers

Mechanistic Interpretability Needs Philosophy

91% match2025

Unboxing the Black Box: Mechanistic Interpretability for Algorithmic Understanding of Neural Networks

90% match2025

A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models

88% match1 repo2412024

Validating Mechanistic Interpretations: An Axiomatic Approach

87% match1 repo22024

Position: An Inner Interpretability Framework for AI Inspired by Lessons from Cognitive Neuroscience

86% match2024

A Mathematical Philosophy of Explanations in Mechanistic Interpretability -- The Strange Science Part I.i

86% match2025

Cite this paper

@article{sharkey2025open,
  title  = {Open Problems in Mechanistic Interpretability},
  author = {Lee Sharkey and Bilal Chughtai and Joshua Batson and Jack Lindsey and Jeff Wu and Lucius Bushnaq and Nicholas Goldowsky-Dill and Stefan Heimersheim and Alejandro Ortega and Joseph Bloom and Stella Biderman and Adria Garriga-Alonso and Arthur Conmy and Neel Nanda and Jessica Rumbelow and Martin Wattenberg and Nandi Schoots and Joseph Miller and Eric J. Michaud and Stephen Casper and Max Tegmark and William Saunders and David Bau and Eric Todd and Atticus Geiger and Mor Geva and Jesse Hoogland and Daniel Murfet and Tom McGrath},
  year   = {2025},
  eprint = {2501.16496},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2501.16496v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

[![Code of Paper](https://codeofpaper.com/badge/2501.16496.svg)](https://codeofpaper.com/paper/2501.16496)