Abstract: Integrating energy and digital revolutions is pivotal for green, low-carbon transformations and achieving "dual-carbon" goals. With breakthroughs in generative large-scale models, artificial intelligence (AI) has evolved from traditional perceptual discrimination to generative decision-making. This enhances solutions for cognitive reliability, interpretability of scheduling decisions, and scenario adaptability in power system construction. Focused on electric power artificial intelligence (EPAI), this paper proposes a roadmap comprising basic technology, data intelligence, perceptual, cognitive, and decision-making intelligence. It details each direction's application status, technological maturity, and current EPAI status. It analyzes EPAI application difficulties in the power sector by addressing challenges like interpretability and security. Proposing core technological directions, it anticipates their application effectiveness in new power system construction, aiming to empower AI technology in energy and power businesses.