When I was a kid, my parents splurged, and bought the (at the time, rather expensive) Encyclopedia Britannica.  My father had wanted to own it himself as a child, though the purchase was clearly mostly for my benefit, as the heavy volumes lived in my room, taking up the entire bottom rows of my three bookshelves.  And, frankly, I loved it.  As anyone who, in the years since, has fallen down a Wikipedia rabbit-hole can attest, it’s remarkably easy to get engrossed in random encyclopedia entries, whether on how the Tower of Pisa got its famous lean, or on the mechanics required for a snake to swallow and digest animals (including whole people) wildly larger than itself.

The Britannica was divided into three parts: the Macropaedia, which had long-form articles on major subjects; the Micropaedia, which had shorter entries on a far wider array of topics; and, my favorite, the Propaedia, a single, relatively slim volume.

Let’s say you wanted to construct an encyclopedia.  Where would you start?  How would you decide which topics to include?  Britannica’s answer was the Propaedia, a taxonomy of all the world’s knowledge.  Like the phylogenic tree of life, it started from a single root, and sub-divided endlessly: ‘Matter & Energy’ split to ‘The Universe’ then to ‘Galaxies & Stars’ to ‘Extragalactic Radio Sources’ to ‘Quasars.’  ‘History of Mankind’ broke into ‘The Modern World’ then to ‘Western Europe 1500-1789′ to “The European States’ to ‘France’ to ‘The Age of Louis XIV.’  I spent almost as much time poring over that single volume as all the others combined. I loved the structure, the organization.  I loved the way it linked across seemingly disparate fields and bodies of knowledge. I intuited, even at a young age, that the better my framework for the big picture, the more easily I would be able to understand, retain, and connect all the details.

These days, I’ve been thinking about the Propaedia a lot, as I’ve been spending much of my time working on the algorithm for Composite.  The idea is simple: if you’re a professional athlete, or a movie star prepping for a role, you have an excellent, experienced, and educated coach who designs a workout plan, a nutrition plan, a set of lifestyle changes, etc., all tailored specifically to you and your goals.  And those plans, plus accountability to the coach to help you actually stick to them, tend to yield extremely impressive results, as a trip to the ballpark or cinema illustrates.

The rest of us, however, just go for a run, copy workouts from fitness magazines, or hire trainers at local gyms whose primary credentials usually include having played D3 football and being great at yelling “it’s all you, bro, it’s all you!”  None of which, perhaps unsurprisingly, work quite as well.  So, fundamentally, Composite is about leveraging the power of AI (as well as technology in general, plus recent research in sport science and behavioral medicine) to let everyone get the pro athlete/movie star custom treatment—and results.

Composite’s algorithm is a neural network, so it will evolve over time, continuously improving its prescriptions as it learns from members’ results (as measured on things like blood panels, body composition, and benchmark workout times).  But, to set the algorithm up, we had a sort of Catch-22: you can’t train a neural net without a ton of data, but because there’s never previously been a use for that kind of data in the fitness world, it doesn’t really exist (at least not in a central, digital way).

To get around the chicken and egg problem, we’ve had to do some heavy lifting, initially building the algorithm using GOFAI (“good ol’ fashioned AI”), setting the symbol weights entirely by hand.  Doing that, in turn, has meant coming up with taxonomy after taxonomy after taxonomy.  We’ve had to reduce all possible, beneficial exercises, all possible nutritional approaches, all training periodization structures, all healthful lifestyle changes, all stretches and mobilizations and pre-hab movements, etc., into meaningfully structured trees.

Which, on the one hand, is kind of bananas.  But, on the other, has been absolutely the best part of the job.  All of those taxonomies have gone through a crazy number of iterations already, and I still regularly jump out of bed at 3 AM to scribble down an epiphany that sends at least one of them straight back to the drawing board.

Each time I work on those taxonomies, I think back to the Propaedia.  And I’m still not sure: did all that time perusing it change my brain and how I look at the world?  Or did it just so perfectly fit the way I already saw things, and gave me a master class in structured thinking, done rigorously and at scale?

Either way, that book was the best thing I could have had sitting on my bedroom shelf.  And I wonder if my own future kids, skipping around Wikipedia, but unable to hold in their own hands a single, unifying big picture, won’t be missing something beautiful and important as a result.