Science advances through the generation and validation of new ideas, yet the scale of modern research makes it difficult to assess reproducibility and identify promising directions. This talk presents AI approaches for studying the science of science: first, methods for detecting convergence toward objective scientific truths across the literature; second, techniques for evaluating the novelty and potential impact of emerging research. I will conclude with an outlook on prototyping agentic systems operating over knowledge graphs, aiming to assist with automated hypothesis generation. Together, these approaches illustrate how AI can augment scientific practice by providing new tools for assessing robustness, prioritizing research, and guiding discovery.