Yoshua Bengio and his team have published a detailed paper on the application of AI in scientific research, covering aspects such as data collection and organization, learning meaningful scientific data representations, generating scientific hypotheses using AI, and AI-driven experiments and simulations. The paper also discusses the core challenges AI faces in crossing disciplinary boundaries, including data normalization, distribution shifts, model interpretability, and computational resource requirements. It proposes several solutions, such as employing geometric priors, self-supervised learning, and language modeling techniques. The authors of the paper include three Chinese first authors and Yoshua Bengio, and it has been published in a Nature review article.