Finding effective treatments for cancer is fundamentally a high-dimensional probabilistic planning, search, and optimization problem, characterized by thousands of molecular subtypes, tens of thousands of plausible drug regimens, and a dearth of high quality data on clinical responses. These explosive combinatorics and the lack of data are overwhelming traditional approaches to clinical development, including the latest Bayesian adaptive trial designs and big data analytics. However, this is precisely the kind of knowledge-based planning and search problem that AI is uniquely designed to solve. Many organizations are developing relevant AI-based applications, from decision support and treatment planning to treatment validation and real time response monitoring. We will review these developments and describe a bold industry-wide initiative to integrate them for the immediate benefit of all cancer patients and their physicians, while radically improving the efficiency of clinical research and drug development.