Molecular dynamics (MD) simulations are key computational methods for understanding physical and chemical processes, as well as predicting the thermodynamic and kinetic properties of molecular systems. Despite their widespread use, traditional MD approaches suffer from several limitations, such as sampling inefficiency, inaccuracies in force fields, and challenges related to non-scalable time and spatial scales. Recent advancements in machine learning (ML) and artificial intelligence (AI) offer new opportunities to overcome these obstacles. Beginning with molecular representations, this chapter reviews the application of various ML techniques (supervised, unsupervised, and reinforcement learning) in addressing long-standing challenges in MD. In particular, we highlight the development of machine learning-based force fields, which offer a powerful solution to enhance both accuracy and efficiency in MD simulations. AI-driven enhanced sampling methods are also emphasised, ranging from the construction of more effective collective variables to generative algorithms for efficient density estimation of the Boltzmann distribution. Furthermore, we discuss a novel perspective on the deep integration of AI and MD, emphasising their synergy at both the software and hardware levels. As this interdisciplinary field continues to evolve, it presents not only significant challenges but also a wealth of opportunities for further exploration.