Digital Twin AI: Opportunities and Challenges from Large Language Models to World Models

TL;DR

This work reframes digital twins as evolving AI-enabled systems, proposing a four-stage lifecycle—modeling the physical twin, mirroring it into a digital twin, intervening via predictive and diagnostic AI, and achieving autonomous management through large language models and intelligent agents. It synthesizes physics-based modeling, data-driven learning, and generative AI (including world models) to enable proactive, reasoning-capable twins across eleven domains, from healthcare to smart cities and aerospace. The paper highlights cross-cutting challenges in scalability, explainability, and trust, and offers directions toward responsible AI-driven DTs, including integration of physics-informed neural networks, neural operators, and multimodal foundation models. By linking methodological advances to practical applications, it underscores the potential for DTs to become trustworthy, self-improving cognitive systems that reason, communicate, and generate creative scenario analyses. The synthesis also explores quantum computing as a potential accelerator for select DT workloads, outlining how quantum modules could complement classical DT pipelines in the future.

Abstract

Digital twins, as precise digital representations of physical systems, have evolved from passive simulation tools into intelligent and autonomous entities through the integration of artificial intelligence technologies. This paper presents a unified four-stage framework that systematically characterizes AI integration across the digital twin lifecycle, spanning modeling, mirroring, intervention, and autonomous management. By synthesizing existing technologies and practices, we distill a unified four-stage framework that systematically characterizes how AI methodologies are embedded across the digital twin lifecycle: (1) modeling the physical twin through physics-based and physics-informed AI approaches, (2) mirroring the physical system into a digital twin with real-time synchronization, (3) intervening in the physical twin through predictive modeling, anomaly detection, and optimization strategies, and (4) achieving autonomous management through large language models, foundation models, and intelligent agents. We analyze the synergy between physics-based modeling and data-driven learning, highlighting the shift from traditional numerical solvers to physics-informed and foundation models for physical systems. Furthermore, we examine how generative AI technologies, including large language models and generative world models, transform digital twins into proactive and self-improving cognitive systems capable of reasoning, communication, and creative scenario generation. Through a cross-domain review spanning eleven application domains, including healthcare, aerospace, smart manufacturing, robotics, and smart cities, we identify common challenges related to scalability, explainability, and trustworthiness, and outline directions for responsible AI-driven digital twin systems.

Figures (4)

• Figure : AI-driven digital twin framework and application landscape. The four-stage lifecycle conceptualizes digital twins as evolving intelligent systems: First, describing the world (physical twin) via physics-informed AI and observational data. Second, mirroring the world into synchronized digital simulators (digital twin) through generative AI. Third, intervening in the world with predictive AI for forecasting, diagnosis, and optimization. Ultimately, achieving autonomous management of the world via agentic AI powered by large language models and foundation models. This conceptual framework generalizes across a wide range of application domains.
• Figure : Physics-Based Methods and AI Systems.(a) Workflow of Physics-Informed Neural Networks (PINNs), integrating prior knowledge into the learning pipeline through regularization and domain constraints. (b) Architecture of PINNs incorporating data, PDEs, and boundary/initial conditions into a unified loss function. (c) Deep Operator Network (DeepONet) structure, modeling nonlinear operators via separate branch and trunk networks. (d) Fourier Neural Operator (FNO) framework, leveraging Fourier transforms for efficient learning of solution operators in PDE problems.
• Figure : Generative AI models. (a) The framework of Generative Adversarial Networks (GANs). (b) The framework of Denoising diffusion probabilistic models (DDPMs). (c) The framework of Neural Radiance Field (NeRF). (d) The framework of 3D Gaussian Splatting.
• Figure : Patient-centric digital twin framework for healthcare. Multi-source patient data, including real-time physiological signals, electronic health records, clinical measurements, laboratory results, genetic and drug data, are continuously ingested into a digital twin platform through data intake and monitoring. The digital twin enables disease modeling and analytics as well as treatment simulation and personalization, generating actionable insights that support clinical decision-making and intervention. By closing the loop between data, modeling, and intervention, the framework facilitates virtual clinical workflows, operational trials and drug discovery, and accelerated research, ultimately improving patient care, reducing costs, and enabling personalized medicine.