The chimps in a cage metaphor is a great introduction to a problem that exists in all software development. I call it the Walls of Assumptions. When we write s

The chimps in a cage metaphor is a great introduction to a problem that exists in all software development. I call it the Walls of Assumptions. When we write software, we answer three questions: "What?", "How?", and "Why?". We write out the answers to "What?" and "How?" explicitly as data and source code. The last answer, alas, can never be written; at least, not explicitly. When we are good programmers, we do our best to write the answer Why implicitly. We write documentation, tutorials, examples, etc. These construct a picture whose negative space looks similar enough to live in Why's place. No matter what, the question "Why?" is always answered. How can this be, if that answer is never written? It is encoded into the entropy of the very act of writing. When we write software, we must make decisions. There are many ways a problem could be solved: choose only one solution. A chosen solution is what I call an "Assumption". It is assumed that the solution you chose will be the best fit for your program: that it is the answer your users need, or at least that it will be good enough for them to accomplish what they want. Inevitably, our Assumptions will be wrong. Users will bring unique problems that your Assumption isn't compatible with. While you hoped your Assumption would be a bridge, it is instead a Wall. The Walls of Assumptions in every program define a unique maze that every software user must traverse to meet their goals. Monolithic design cultivates a walled garden, where an efficient maze may fail entirely to lead the user to their goal. Modular design cultivates an ecosystem of compatible mazes that, while less efficient, can be restructured to reach more goals. --- The eternal hype around Natural Language Processing and Artificial Intelligence is readily explained with this metaphor. The most powerful feature of Natural Language is Ambiguity. Ambiguity allows us to encode more than one answer into data, which means we actually can write the answer to Why; we just can't read it computationally. Artificial Intelligence hinges on the ability for decision to be encoded into software. I'm not talking about logical branches here: I'm talking about the ability to fully postpone the answering of Why from time-of-writing to runtime. --- For the last year or two, I've been chewing on a potential solution to this problem that I call the Story Empathizer. So far, the idea is too abstract; but I still think it has potential.