Thread by @StatesWarring

What’s the difference between studying and learning?

This has come up a few times lately on ancient China twitter, so strap yourselves in, put on some lo-fi beats, and let’s do this.

Tldr: learning hurts, studying doesn't. If you’re memorizing someone else’s observations, you’re studying. If you’re lying bleeding in a ditch thinking “well, I probably won’t do that again”, you’re learning.

If you want more, however, read on.

Confucius was big on studying. It’s in the very first lines of the Analects:

He also wasn’t keen on the more empirical, ditch-based forms of learning (the translation of the relevant section on Ctexts is pretty bad, so here’s a different one):

https://twitter.com/Solzi_Sez/status/1346562621332328448

This wasn’t just a personal preference. Confucianism adopted a fake-it-til-you-make-it approach to understanding, and the preference for study over experience-based learning emerges naturally from that.

The great sages of old truly understood the nature of the universe, and you probably can’t match that. What you can do, however, is copy their actions as recorded in the ancient texts.

Maybe by copying you’ll actually come to understand why they did these things, but even if you don’t, you’ll still achieve a better outcome than if you’d just followed your own preferences.

Taoism disagrees. From their perspective, Confucians, in their focus on written sources as a gateway to truth, were looking at the finger rather than the moon. They believed that real understanding came from direct contact with the world.

Taoists didn’t reject the concept writing altogether (obviously, or we wouldn’t have their books), but they emphasized that written information should be treated as a guidebook rather than a textbook (a metaphor I've used before, but I'm lazy, so sue me).

In other words, the Tao Te Ching or the Han Feizi are intended to be carried with you on your travels in case you need pointers, not as taken as objects of study in themselves. (Something that modern academia tends to neglect. Ho hum.)

For Taoists, it was the underlying reality that was important, not the literary description of it. Hence the idea of zuowang (sitting and forgetting). The goal was to forget words and get at the truth by experiencing reality unmediated by others’ conceptual categories.

The texts tend to focus on people who were exemplary when it came to achieving this state of flow, for obvious reasons. Nevertheless, it was something that you had to work towards, and you worked towards it by trying, getting it wrong, and then doing better as a result.

The underlying idea was that one way or another the universe is going to show you how to exist within it, and whether the lessons go easy or hard is up to you.

If you succeed, great. If you fail, that's also great, because either you'll do better next time or you'll die and improve the gene pool. The system's set up so that micro-failures improve macro-performance, so it's all good, fam.

As an individual existing in such a system, your goal is thus to get into a receptive frame of mind, from which you can experience the learning process as painlessly as possible.

In other words, learning is a gradient descent function: en.wikipedia.org/wiki/Gradient_…

This is why AIs are so much better at this than individual humans. They can fling themselves at a problem a million times without getting bored or discouraged or dying, and they’re entirely unbothered by the preconceptions that humans have to zuowang over.

https://twitter.com/vkrakovna/status/980786258883612672

Humans only start to exhibit the same properties in large groups. A million idiots trying to solve a problem will usually do a far better job than one smart guy, simply because the fact that they can try more approaches and fail at a lesser cost means that they learn faster.

And this is where it gets interesting. Legalists realized that you can reverse-engineer the processes described in Taoist texts.

You don’t just need to rely on people gradually learning from nature which berries are edible or how to cut up an ox or whatever. You can *change the gradient yourself* to make them learn what you want them to learn.

Sure, you can give people a book of exemplary acts of the ancient sages and hope that they’re inspired to imitate it, but it’s much quicker and more effective to change the incentives to which they are subject and allow them to learn via the consequences that they experience.

If you lose a body part every time you head in direction X, people are probably going to stop going there before very long. If you find out that you get $20 every time you go in direction Y, it's pretty much a no-brainer.

This has a parallel in machine learning too: to change the results of a reinforcement learning model, you don’t change the algorithms, you change the reward function to which they are subject. en.wikipedia.org/wiki/Reinforce…

Originally these ideas were deployed with the aim of getting people to do things that are not necessarily fun, but which will benefit them in the long run.

Farming is boring as all get out and gross to boot, but if you want to eat someone’s going to have to do some. If you want your farms to be productive, then, you need to make sure the rewards for doing it (or the punishments for not doing it) outweigh the downsides.

But these methods are just tools, and you don’t *have* to use them to such virtuous ends. We see this at the end of the Li Si chapter of the Shiji (sorry, I don’t think there’s an English translation of this online).

Li Si has been imprisoned by Zhao Gao. Gao knows that Si has a good chance of convincing the Emperor of his innocence if given a hearing, so before this can happen he sends his own henchmen in as if they were official investigators.

Every time Si gives a "wrong" (i.e. truthful) answer, he gets a beating. The result is that by the time the real investigators arrive he is refusing to say anything, and gets executed.

In one of the big ironic moments that Sima Qian loved so much, Gao effectively trains him to go against his own interests in exactly the same way that Si trained the Qin populace to follow theirs.