I'm a mathematician who scripts everything I can. I've been thinking about mind-mapping for decades.<p>My breaking point: I have DropBox folders for 1770 math papers, that I periodically read on computers and tablets. The significance of most of these papers is lost to me. I want to be able to browse this collection in a "supermarket shelf" topology. Most of creatitivity is not planned; planning tends to stifle creativity. Rather, a frenetic mind is fortunate enough, after months of fruitless wanderings, to encounter the right set of ideas in close enough proximity to notice. For example, a PCB protoboard has the correct 0.1" spacing to guide a router jig for making a pasta guitar. Who knew?!<p>Ideally one traverses a private web site with links to every asset. The associations need to be both manual and automated, logical and accidental. For example, time of creation or visits is critical data for determining proximity. Looking at each PDF in plain text to attempt to infer proximity would be a great application of deep learning tools, leveraging existing proximities as training data.<p>On a larger scale I have repeatedly urged the American Mathemetical Society to open up its MathSciNet cash cow to be a premier playground for machine learning. Unlike computer science, mathematics does have a detailed index of all published research, through the efforts of many volunteer reviewers. However, math is crippled by many archaic beliefs: That ideas are organized by fields such as number theory, and there is a single global worldview specifying this organization. That the value of MathSciNet is its hand-curated organization. I believe that we're crippling a generation of mathematical progress by not allowing mathematics to be at the forefront of mind mapping efforts, with many competing voices/tools describing its organization and facilitating idea browsing. Autocomplete is still at the "pager" stage of development, with "smart phones" yet to come. We could all benefit from the cloud autocompleting each of our mind-maps. Mathematics is a relatively small, constrained domain where shared mind-mapping and autocomplete could first flourish.<p>I've separately used a text file as scratch paper many days for the last fifteen years, saving every file. I can retrieve much useful information through full text search. I am reminded however of Don Knuth's adage "Premature optimization is the root of all evil," and the smaller optimization problem of how to best organize physical receipts. Excessive organization is wasted effort and a form of neurosis. I scan lots of documents but not every receipt; they all go in boxes dated by quarter, vacuum packed after a year. When I really need a receipt I can find it, but the effort to file has been properly balanced with the effort to retrieve. Same with text files; full text search works. It would be a mistake for me to put more effort into the structure of these files. Rather, through scripts and AI tools, I can evolve an indexing exoskeleton around these existing unstructured raw materials.<p>I need mind-mapping to manage assets, not to create a private web site of my thoughts. And there cannot be a user/programmer dichotomy; we each need to script additional associations into our mind maps. That belongs in exoskeleton, not cluttering up primary text.