by Giles Strong

For the past six or so years, I’ve practised a martial art called Shorinji Kempo. Like many other arts, it incorporates several philosophies and concepts. One of these, The Three Teachings of Ken, concerns one’s progression in learning the various techniques. Simply put, it describes three stages of mastery: shu – learn & copy, ha – adjust & adapt, ri – master & break free.

In the first stage, one merely focuses on replicating the teachers’ movements exactly. By doing so, one can learn the core principals of a technique, and what it aims to achieve. Having gained an understanding of the technique, one can move to the second stage, and begin to adjust the technique. This doesn’t mean completely reinventing it, but adapting it to better suit oneself. An obvious example might be finding easier ways to apply it, given the difference in physical build between one’s self and one’s opponent.

By the third stage, one might have discovered little tricks and adjustments that increase the overall efficiency of the technique and give it its own unique flavour. These little adaptations go on to be copied by other students and shared. It’s quite usual for a teacher to initially show how they do a technique, and then show some of the ways they’ve seen other people perform it.

I’ve practised in several countries now, and find it very interesting how the same techniques can be applied in different ways, according to how the practitioners have been taught and developed. As I wrote for a grading examination last year, it is like learning to cook a meal: first one starts by following a recipe exactly; eventually one learns to adjust quantities, substitute ingredients, add different flavours, and take inspiration from other recipes; and after some time, the meal is no longer “Fearnley-Whittingstall’s vegetable soup”, but one’s own take on what vegetable soup is. Provided it is still a soup, offers similar or greater levels of nutrition, and hasn’t vastly increased in cost or production time, the evolution can be regarded as a success.

In physics, and indeed in life in general, one is required to use many different skills. I try to bear this philosophy of mastery in mind when trying to learn them: I find examples or people to copy, such as coding tutorials, inspiring music and musicians, lectures and seminars, and existing work in papers. I then work on reproducing them, or taking ideas and inspiration from them. As I grow to learn more about the particular skill, I adapt what I learn to better suit my needs and style, and eventually incorporate them into my own repertoire of techniques.

Another example to consider is literacy: you are able to speak and listen because you copied some system of translating concepts into sounds, and vice versa. Similarly you are able to read and write because you copied a system of relating concepts to lines or sets of pixels. However, your way of writing and speaking will be different to other people’s, even within the same language. This is because you eventually stopped merely copying and began adapting your choice of words, manner of speaking, and writing style. We can even see some of these differences, thanks to a recent post by Greg, in which he visualises each of this blog’s authors’ repertoires of vocabulary, and how they have evolved over the past year.

Of course, acquiring a skill takes time, and often it is required to perform the skill prior to fully mastering it: we’d have died of starvation long ago if we’d waited to learn to cook the perfect meal before eating. And even if we did not, the perfect meal doesn’t exist; an excellent meal can always be better.

No, acquiring a skill does not mean aiming to become perfect at it, it means to be willing to put in the work necessary to improve it over days, years, decades, to learn from mistakes and successes, and to substitute old concepts for better ones. It is useful to consider the phrase zen zen shu gaku – soak up, soak up, acquire, learn. The brain is like a sponge, and sponges don’t soak up water instantly, they absorb it slowly over time. (And like a sponge, the brain (being ~78% water), will likely leak said water if subjected to sufficient pressure, but that’s kind of gross to think about and doesn’t really aid our simile).

This method of copy, adapt, master, however, focuses almost entirely on the growth of the person, and not on the growth of the skill itself. Given its original intention, as a progression of mastery, this is sufficient; whilst Shorinji Kempo aims to adapt to meet the current needs of society, the physical techniques and core philosophies are fixed. Divergences from these are not within the art.

Similarly, the growth of language relies on mutual understanding between speakers and a critical acceptance of new words or grammatical formalisms; it is perhaps better modelled as something that emerges from society, rather than something which can be driven by individuals.

Other skills and fields, however, are less bounded and more amenable to change, such as analysis techniques, art, scientific knowledge, and music. Aside from strokes of genius, the development of these is likely to come from one’s ingenuity and inspirations based on mastery of the skill and others closely related, and so on the work of others who made earlier contributions to the field.

I’m not familiar with studies and theories of learning, but I cannot think of another means of acquiring a skill aside from starting from scratch. Even then, however, to even know that the skill exists, one must surely have seen prior examples which would lead to an inherent bias in one’s understanding, and so to copying.

Perhaps an alternative method might be just copying and adapting, without ever understand the skill, and so never achieving mastery. This method can be seen in computer-generated artwork. Here, algorithms learn to create new artwork by being trained to reproduce existing artwork. Their styles can even be adapted by importing the styles of other paintings into the piece. Some systems can do this so well that humans cannot accurately tell the difference between ‘human art’ and ‘machine art’. At least until AGI, though, I would argue that these algorithms never understand what art is, and instead learn very well the process required to produce it, and so never reach mastery.

Whilst I’d dismissed the possibility for humans to learn a skill from scratch, for machines, the idea is more plausible. For skills with some set aim, a set of rules, and a feedback mechanism, the system can learn by itself how best to achieve the goal. This field of unsupervised or reinforcement learning requires iterating through many strategies, evolving them according to the results, and possibly developing them by playing against itself. Something that would take a human far too long to do, but for a computer is possible.

Early last year such a system, AlphaGo, beat Lee Sedol, one of the best players of Go. The game is ancient and like human players, AlphaGo had been initially trained by following the games of previous players. Last month a newer version, AlphaGo Zero, was introduced. This version foregoes initial training on past games, and truly learns the skill from scratch. After three days it could beat the version that had bested Lee, and after 40 outperformed the prior best version. It’s possible that the strategies and techniques that these systems develop could be studied and adopted by humans, and so could be used as a means to develop both the skill, and one’s mastery of it, simultaneously.

Well, this post rambled on for a lot longer than I’d intended. Guess I still have a long way to go in developing my blogging skills, but thanks for surviving to the end! Hopefully it’s been of some interest. I’d like to know whether you are aware of or use any other methods of learning skills. Let me know in the comments. Cheers.