Machine learning methods are transforming the ways that anyone can interact with map collections. Using the results of a computer vision pipeline to understand past places at scale comes with new challenges and opportunities. The Machines Reading Maps project has worked with partners at the Library of Congress, the National Library of Scotland, the British Library, and also the David Rumsey Map Collection to create datasets of the text on large collections of digitized maps. This talk explores the results of the Machines Reading Maps project to date including our work in progress to, on the one hand, visualize and analyze the results for historical research, and, on the other, use text on maps data to improve map collection discovery.