Major Goals: Our goal is to create a fully automated system that will rapidly and efficiently produce any specified organic target molecule - an extraordinary challenge considering past practice. Currently, syntheses are often unreliable and require reoptimization or redevelopment at each phase of scale-up. The challenge is to automate the design of an efficient synthetic route to a target molecule, specified in sufficient detail that an automated reactor system can carry it out. Our approach builds on recent advances in machine learning, cheminformatics, and computational chemistry to identify promising synthetic routes to arbitrary target molecules. Machine learning can effectively utilize the large corpus of known reactions to infer essential information about new (unreported) reactions such as optimal conditions and likely yields. Such predictive methods are fast and robust. In addition to knowledge extracted from literature reactions, we will apply computational chemistry approaches to provide increasingly accurate predictions of reaction yields and physical properties (e.g., molecular solubility). Such detailed knowledge of each individual reaction will be combined to identify optimal multistep syntheses. These syntheses will then be evaluated and optimized experimentally generating new data to feedback and refine the synthesis planning process. Combining chemistry and computer science will lead to transformative advances, as already seen in the biological sciences. In part, we will be automating and fully integrating steps currently done by humans, but we will also do tasks impossible even for the most efficient team of human experts. We will teach a computer to use millions of literature syntheses, systematically find commonalities and differences among different reactions, and then use these results to make inferences about new reactions.