71. Riff on Kata-Style Learning 🔗

I once attended a talk about “coding kata.” I don’t think the idea ever took off. It works even worse with hardware.

Something about engineering seems to defy learning through abstract formal exercises of growing combinatorial complexity. Like music scales or kata. Wonder why.

This may even be the dividing line between artisan craft and engineering. Unlike woodworking or other traditional proto-engineering crafts, the modern maker world is engineering, just at low-cost amateur level. It lacks the “katafibiality” of skills.

My suspicions: specialization of both labor and objects kills katafiability. Music has 12 semitones, martial arts have perhaps a dozen basic movements. My beginner Arduino electronics kit has like 50 components, all with idiosyncratic features to learn. My dremel has 28.

And this beginner level shit, barely scratches the surface. I’m now vaguely oriented in a world of 100s of electronics parts, and a seasoned electronics maker has a vocabulary of 1000s I think. Same with mechanical (think types of drill bits, fasteners…).

And these universes have very little formal structure. The theory of circuits structures this universe of artifacts far less than the mathematics of harmonic ratios structures music or geometry and kinematic structure martial arts.

It’s a bit like CISC over RISC (complex vs reduced instruction set computing, 2 architectural approaches in semiconductors).

The thing is, the larger universe is not “spanned” in a simple way by the combinatorial space around a small set. At best you get 80-20.

As in 80% of the variety is composite constructs from 20%. For example the L298N is a chip that contains too full H-bridge circuits, which you could also wire up with hookup wires and individual transistors and diodes on a breadboard. It’s not a “note” but a “scale”

…BUT

… 20% of the parts (probably more?) are NOT like that. They introduce new physics into the game. New “notes.”

For eg. thermistor introduces the physics of resistance+temperature. When you first make a thermistor circuit you have to add a new primitive “note” to vocabulary.

Every new component or tool has like a 20% chance of pushing you into new physics.

There’s that line “physics is the only science, everything else is stamp collecting.” Well, engineering is stamp collecting IN physics.

This is why, though most of my education is in formal foundations in math/physics, I think overindexing on that misses the point. It totally elides the part of engineering that is stamp collecting. You need the basic conceptual literacy to read/parse the stamps, but it’s stamps.

This is also why toys like this (I’ve been seeing this ad for weeks) leave me meh. If you’re struggling with concepts so much you need this, you’re in trouble. This removes what’s actually important about working with things. This is a math toy, not an engineering toy. tweet^[1]

In engineering school, what this toy is apparently trying to teach is conveyed through (for eg) the differential equations for RLC circuits and spring-mass-damper systems being similar. It’s a math class, not a shop class. It’s basic literacy needed to do the stamp collecting.

I think many toys make this mistake. They try to teach basic theoretical concepts that you’ll eventually learn with math/physics classes. Cute but redundant. But you get an illusion of learning stamp-collecting side that you can’t learn in a math/physics course, when you’re not.

In a way all the maker stuff I’m playing with and posting about right now is an attempt to kinda set aside the math/physics foundations, and re-establish hands-on stuff on independent empirical/phenomenological “stamp collecting” foundations.

The math/physics is still important, but as a cost of doing business. It’s an entry-level boundary condition that separates engineering from vocational technician skills.

That divide btw is an interesting one. Somehow industrialization created a vocational layer mostly sealed off from engineering. Welders, machinists, electronics assembly people, plumbers,… a vast universe of technicians who can get away with very limited math/physics.

Industrial vocations are partly descended from artisan trades, partly, artificial constructs like video games. You may intuit physics of wood if you do woodworking, but there’s no chance a minimum wage circuit assembler can intuit semiconductor physics with no textbooks.

Yeah, good point. There is “infrastructure fluency” in a few key tools (like a soldering iron) but in general, fluency is neither a thing in engineering, nor particularly central. You expect work to feel awkward and non-fluent half the time. tweet^[2]

Another good point. Advanced math starts to feel like extremely discriminating stamp collecting. Once you get past foundations and basic skills, all the action is in the “rare stamp” theorems that are both true and important in the ocean of unimportant trivial truths. tweet^[3]

For example, early in linear algebra/control theory you learn about the Cayley-Hamilton theorem, an important “stamp.” A non-trivial AND important truth as in lots of important stuff depends on it. Like a breadboard or a drill in making. en.wikipedia.org/wiki/Cayley–Ha…

One of the reasons I didn’t get far in math is that though I was good at the raw skills like manipulating trigonometric identities or differentiation or Laplace transform mechanics, I never developed a “taste” for how to wield the important and charismatic theorems/equations.

Things like the CHT or say the Brouwer fixed-point theorem are like temperamental tools with strong “personalities” (think soldering iron) that give you superpowers once you master them. I’m citing these because they are among the few I did gain some literacy in.

By contrast, I gave up on topology after the first grad course because it is just full of such things. Topology is the “fantastic beasts and where to find them” part of math. Really advanced stamp collecting. If I understand correctly, Grothendieck was a sort of Newt Scamander.

I’m possibly doing an injustice to music and martial arts here, but I think they’re closer to Euclidean geometry than engineering. As in, the combinatorial space covers most of it, and what isn’t covered is more subconsciously/tacitly learnable. tweet^[4]

While there’s stuff beyond the combinatorial space of scales/ragas etc, it it’s mostly tacit. Things like instrument-specific timbre and elements like microtones (raga musicians in particular go on and on about that and entire schools rest on microtonal aesthetic theories)…

Fair enough… I imagine you might suddenly run into a particular weird “new physics” effect while experimenting with a particular combination of formal scale/instrument/microtonal flourishes. Often those are what I personally like best in music. New emergent effects. tweet^[5]

I can believe it. I have a suspicion that kata is bullshit work. Practicing basic combinations/drill (kihon) and sparring (kumite) make sense, but kata never made sense to me. I gave up karate after a year in grade school because they only did kata and it bored me to tears. tweet^[6]

Come to think of it, this may also be why I never pursued music. My tabla teacher had me doing kata-like drills for years and I never got to accompaniment performance (the equivalent of sparring) while some friends in more musical families were learning accompaniment from day 1.

Maybe conclusion from this thread is that kata/etude like things don’t just fail in engineering, but they don’t even really work where they are common by tradition. Probably overstating the case, but why not. Null hypothesis: all kata-like things are bullshit work. Any defenders?

I’m tempted to connect kata like things to pattern language type ideas and call bullshit on that too, since they’re clearly cousins. But I don’t want to be murdered by the Christopher Alexander mafia 😆

Friction is real action, flow is self-indulgence? 🤔 tweet^[7]

The problem here is that there’s no real value to repeating the exercise. It’s a one-time thing, and doing it again or even on a new board doesn’t do much. The essence of kata is repetition. I don’t think anybody would do 100s or 1000s of led-blink or hello-world programs. tweet^[8]