Paper
Interpretability for Turing Machines
We show that susceptibilities, an interpretability technique developed for neural networks, can identify the presence of algorithmic structure in Turing machines by probing the local loss landscape of a learning problem for noisy Turing machines introduced by Murfet and Troiani (arXiv:2504.08075). We prove that symmetries and path separation in the algorithm implemented by a Turing machine induce permutation symmetri · Authors: Billy Snikkers, Rumi Salazar, Daniel Murfet, Will Troiani · Published 2026-09-04 · arXiv:2609.04661
- Published
- 2026-09-04
- Authors
- 4
- arXiv
- 2609.04661
- Category
- cs.LG
Abstract
arXiv:2609.04661
Billy Snikkers, Rumi Salazar, Daniel Murfet, Will Troiani
We show that susceptibilities, an interpretability technique developed for neural networks, can identify the presence of algorithmic structure in Turing machines by probing the local loss landscape of a learning problem for noisy Turing machines introduced by Murfet and Troiani (arXiv:2504.08075). We prove that symmetries and path separation in the algorithm implemented by a Turing machine induce permutation symmetries and low-rank blocks in its susceptibility matrix. We study this empirically on a set of deterministic finite automata (DFAs) and demonstrate that algorithmic features can be recovered by principal component analysis and clustering methods in susceptibility space.