With the rapid growth of the IoT, ensuring efficient and scalable multiple access (MA) mechanisms is crucial to support billions of interconnected devices. Traditional MA schemes face challenges in handling massive device density, heterogeneous traffic, and dynamic network conditions. These conventional approaches struggle with scalability, inefficient resource utilization, and energy consumption, particularly in large-scale IoT deployments. This chapter explores how AI-driven MA techniques can overcome these limitations by introducing intelligence into communication protocols. It introduces a conceptual shift from static MA techniques to dynamic, learning-based approaches, where IoT devices autonomously optimize their access strategies using AI. The chapter presents two key AI-enabled MA techniques: (1) Multi-Agent Deep Reinforcement Learning (MADRL) for Autonomous Channel Access (AutoCA), which replaces traditional CSMA/CA-based protocols by allowing IoT devices to learn and adapt access policies, thereby improving throughput, reducing collisions, and ensuring fairness. (2) Over-the-Air Computation (AirComp) for Federated Learning (FL), an innovative MA method tailored for AI-driven IoT applications, which aggregates data directly over the air, reducing communication latency and bandwidth usage while maintaining privacy-preserving distributed learning. Through extensive numerical evaluations and case studies, this chapter demonstrates the superior performance of AI-enhanced MA techniques, showcasing their ability to reduce packet loss, improve energy efficiency, and enhance network adaptability. These advancements pave the way for intelligent, scalable, and energy-efficient IoT networks, transforming how IoT devices communicate in next-generation wireless environments.