> Rabkin's Dictum: If you don't understand something, it's because you aren't aware of its context. Building off of a discussion with [[Andy Matuschak]] at [[Edge Esmeralda]] about his talk [How Might We Learn](https://andymatuschak.org/hmwl/): There's an apparent tension between in-context, just-in-time learning, and working one's way through a structured, abstract curriculum. The product sketches in How Might We Learn bridge this gap by imagining an AI (or more broadly, a [[ubiquitous computing]] system) that has access to both the world of a learner's desires & capacities, & the world of the discipline, tradition, industry they're operating in. So at every moment, following along with a learner, it can find the [[next right action]], the necessary context, the next step in the [[skill progression]], which might be asking a question, pointing to the relevant portion of a text, generating a dynamic representation, providing an interactive exercise, ... so that the learner progresses. There's also an apparent gap between tacit knowledge--e.g. how does an improvisational dancer know what move to make next?--& legible, transmissible knowledge. Tacit knowledge is called tacit because it's difficult to express. But we can also imagine our system observing that dancer and finding a [[next right action]] to support them not as a learner, but as a teacher. It can use the same capacity to synthesize context to help express & represent tacit knowledge. it might start with direct data capture: - video of dance movements - measurements from physiological sensors and continue through interviewing with dynamically generated questions, prompts, provocations: - ask the dancer to tell stories about their own subjective experience of choosing/not choosing what comes next - take a video, pause it at a certain moment, synthesize alternative video or images or continuations from that moment, ask questions about your intuitive reaction to them And all of this can feed a machine learning system that can tease out the hidden patterns and make them legible--in other words, to translate from an unknown language into a known one. Last year, [[Robin Sloan]] [asked](https://www.robinsloan.com/lab/phase-change/): "What could I do with a universal function — a tool for turning just about any X into just about any Y with plain language instructions?" He was asking in the context of software systems across the web -- but now we can ask this question again with X and Y as bodies of knowledge, whether they reside in computers, humans, or anything else. ___ Related: - Andy suggests: [An Easier Method for Extracting Tacit Knowledge](https://commoncog.com/an-easier-method-for-extracting-tacit-knowledge/)