This talk explores the emerging role of artificial intelligence in scientific discovery. It first highlights the importance of explainability in AI for Science, discussing how interpretable AI approaches can improve model transparency, identify data biases and failure modes, incorporate domain knowledge, and facilitate scientific hypothesis generation. This talk introduces different categories of explainable AI methods, including knowledge-infused, knowledge-verified, knowledge-based, and knowledge-agnostic approaches, and discusses how they can enhance the reliability and scientific value of AI-driven discoveries. It also presents examples of applying explainable AI to scientific problems, including nano–bio interface research, to demonstrate how AI can move beyond prediction toward mechanistic understanding. This talk further explores the transition from large language models to AI agents for science, highlighting their potential as collaborative research partners that can support scientific reasoning, workflow automation, and discovery. It concludes by emphasizing that while AI can accelerate the process of answering scientific questions, human researchers remain essential for identifying meaningful questions and driving scientific innovation.