Modern deep learning and AI-assisted code generation are lowering the technical barrier to building mathematical models of biological systems, enabling researchers with domain expertise to create semi-mechanistic, hybrid models that combine prior knowledge with learned structure. However, the predictive success of these grey-box models does not guarantee that the inferred mechanisms truly reflect the underlying biological processes, as the learned components can mask errors in the mechanistic part. To address this ambiguity, the paper proposes using causal inference to identify which causal relationships are discernible from data and knowledge, and ensemble analyses to quantify uncertainty across models that fit the data equally well, thereby clarifying what is supported, what remains uncertain, and which experiments could resolve the remaining questions.
arXiv:2608. 12036v1 Announce Type: new Abstract: AI models have achieved remarkable success across diverse domains, yet the mechanisms underlying their capabilities and the risks they may pose remain poorly understood.
arXiv:2607. 22811v1 Announce Type: cross Abstract: Hybrid mechanistic/data-driven models, which combine first-principles with learned components, are increasingly used in process engineering and scientific machine learning.
arXiv:2606. 02632v1 Announce Type: cross Abstract: Modern Machine Learning (ML) and Artificial Intelligence (AI) models, especially large language models (LLMs), are increasingly used to generate scientific hypotheses and mechanistic explanations from observational data.
arXiv:2608. 09696v1 Announce Type: new Abstract: Predicting the answer to interventional ``what if'' questions --- the outcome of an action never taken --- requires a \emph{mechanistic}, causal model, not a curve fit; and learning such a model requires \emph{experiments}, because passive data leaves its mechanisms unidentified.
The paper surveys mechanistic interpretability as a means to understand and control neural networks for AI safety. It reviews foundational concepts such as features, polysemanticity, transformer computation, and circuits, and discusses methods for feature extraction, circuit discovery, causal verification, and training-data attribution. The authors propose a unified evaluation framework, outline six research orientations, and explore how mechanistic insights can aid safety through knowledge editing, steering, and monitoring while acknowledging limitations and open problems.
The paper surveys mechanistic interpretability for AI safety, exploring how studying internal representations and computations of neural networks can help evaluate, predict, and control their behavior. It reviews foundational concepts such as features, polysemanticity, transformer computation, and circuits, and surveys methods for feature extraction, circuit discovery, causal verification, and training-data attribution. The authors propose a unified framework for evaluating mechanistic claims, outline six research orientations, and discuss how mechanistic understanding can aid safety through knowledge editing, steering, and monitoring, while also highlighting limitations and open problems.
Predicting the answer to interventional ``what if'' questions --- the outcome of an action never taken --- requires a \emph{mechanistic}, causal model, not a curve fit; and learning such a model requires \emph{experiments}, because passive data leaves its mechanisms unidentified. Experiments are expensive, so the central problem is \emph{data efficiency}.
arXiv:2607. 18777v1 Announce Type: new Abstract: Evaluating machine learning in scientific domains requires separating correct predictions from correct reasons under realistic distribution shifts.
arXiv:2607. 07316v1 Announce Type: new Abstract: This article offers a comprehensive overview of mechanistic interpretability, an emerging field that seeks to reverse-engineer the internal algorithms of modern neural networks.
arXiv:2608.30165v1 Announce Type: cross Abstract: Scientific progress depends not only on finding solutions, but on learning the rules that explain why they work and using that understanding to desig...
arXiv:2608. 07214v1 Announce Type: cross Abstract: Modern AI is no longer a single model but an ecosystem: classical ML predictors, deep and multimodal models, large language models, and agents, each trained and tuned over different data sources and each producing outputs at scale that become inputs to the others.
arXiv:2610.19810v1 Announce Type: cross Abstract: Large language models are increasingly competent at leveraging scientific knowledge and existing computational tools to execute complex workflows. At...