Luis Lopez

Machine learning (ML) models are increasingly discussed in the philosophy of science in terms of idealization and even by analogy with toy models. I argue that this framing conflates idealization with approximation. Once the relevant ML target is specified as relations among features represented in data---relative to a target system---the claim that ML models are dissimilar to their targets becomes too coarse. Scientific ML models can and often must be approximately similar to their targets in the relevant respects. Approximation failure therefore provides the more direct standard of assessment, and ML models do not function as toy models.

This chapter explores the relevance of ‘symphoria,’ a concept from organic chemistry, to a general theory of computational implementation. It suggests that symphoria, “the bringing together of reactants in the proper spatial relationship,” can be generalized to help explain how and in virtue of what diverse physical systems, such as DNA-based neural networks and photonic neuromorphic systems, carry out computations. The chapter discusses how this concept might integrate with existing accounts of physical computation, offering a new perspective that preserves both extensional adequacy and the naturalistic desideratum of a good theory of implementation.

This paper explores the development and evaluation of physics-specific large-scale AI models, which we refer to as large physics models (LPMs). These models, based on foundation models such as large language models (LLMs) are tailored to address the unique demands of physics research. LPMs can function independently or as part of an integrated framework. This framework can incorporate specialized tools, including symbolic reasoning modules for mathematical manipulations, frameworks to analyse specific experimental and simulated data, and mechanisms for synthesizing insights from physical theories and scientific literature. We begin by examining whether the physics community should actively develop and refine dedicated models, rather than relying solely on commercial LLMs. We then outline how LPMs can be realized through interdisciplinary collaboration among experts in physics, computer science, and philosophy of science. To integrate these models effectively, we identify three key pillars: Development, Evaluation, and Philosophical Reflection. Development focuses on constructing models capable of processing physics texts, mathematical formulations, and diverse physical data. Evaluation assesses accuracy and reliability through testing and benchmarking. Finally, Philosophical Reflection encompasses the analysis of broader implications of LLMs in physics, including their potential to generate new scientific understanding and what novel collaboration dynamics might arise in research. Inspired by the organizational structure of experimental collaborations in particle physics, we propose a similarly interdisciplinary and collaborative approach to building and refining large physics models. This roadmap provides specific objectives, defines pathways to achieve them, and identifies challenges that must be addressed to realise physics-specific large scale AI models.