On the Loss and Preservation of Knowledge

post by Samo Burja · 2018-03-08T18:40:00.737Z · score: 110 (35 votes) · LW · GW · 20 comments


  Traditions of Knowledge
    Signs of traditions of knowledge
    A Cautionary Note
  Live Traditions
    Features of living traditions
  Dead Traditions
    Problems related to transferring a body of knowledge
    Problems related to creating an organization

[Crossposted from my Medium blog]

Let’s say you are designing a research program, and you’re realizing that the topic you’re hoping to understand is too big to cover in your lifetime. How do you make sure that people continue your work after you’re gone? Let’s say you are trying to understand what Aristotle would think about artificial intelligence. Should you spend time reading and trying to understand Aristotle’s works, or can you talk to modern Aristotelian scholars and defer to their opinion? How can you make this decision? Both of these situations require an understanding of traditions of knowledge — in particular, an understanding of whether a tradition of knowledge has been successfully or unsuccessfully transmitted. But first: what is a tradition of knowledge?

Traditions of Knowledge

A tradition of knowledge is a body of a knowledge that has been successfully successively worked on. It is useful to classify traditions of knowledge into three types: living, dead, and lost traditions.

• A living tradition of knowledge is a tradition in which the body of knowledge has been successfully transferred, i.e. passed on to people who comprehend it (e.g. cryptography). Note that the content of the tradition’s body of knowledge does not have to be strictly or fully accurate for the tradition to be living; it merely needs to be passed on.

• A dead tradition of knowledge is a tradition in which the body of knowledge has been unsuccessfully transferred, i.e. its external forms, its trappings, have been transferred, but not the understanding of its body of knowledge (e.g. scholars who can recite Aristotle but can’t use arguments as he did; Buddhist monks who chant the instructions to meditation rather than doing meditation itself). Note that this means a tradition can be dead while people still read its texts.

• A lost tradition of knowledge is a tradition that has not been transferred at all (e.g. numerous schools during the Hundred Schools of Thought period in China; the theology of the Cathars, which is only preserved in the words of their critics). The people who had the knowledge died without leaving any successors or substantial record of their knowledge.

It can be difficult to distinguish between different traditions of knowledge. There are traditions within traditions, and there are traditions that become fellow travelers, in the sense that they are related to but merely adjacent to one another. There are also traditions that have a long history of arguing against each other.


It matters whether a tradition of knowledge is living or dead. This is obviously the case if you are starting a research program — you want the tradition you start to stay alive. Whether or not the Aristotelian tradition is dead also matters if you are trying to understand what Aristotle would have thought about artificial intelligence: it determines whether or not you can trust the “authorities” on Aristotle — if the tradition is dead, then their expertise will not be helpful to you. It also matters if a tradition of knowledge is lost: this will inform your understanding of what it is possible to know about that tradition. For this essay, we will focus on understanding how to distinguish between a living and a dead tradition. This can be tricky; it’s hard to trace traditions of knowledge, so it’s also hard to notice when they die.


How can you tell whether a tradition of knowledge is living or dead? First, you have to be able to identify signs that indicate the existence of a tradition of knowledge. You have to be able to recognize signs that indicate the existence of a tradition at all, then determine whether those signs taken together indicate that the tradition is dead or that it is alive (the signs used to recognize the existence of a tradition are the same signs used to distinguish between living and dead traditions).

Signs that indicate the existence of a tradition of knowledge vary in the degree to which they indicate that a tradition is alive, that understanding has been passed on. A collection of signs that weakly or do not at all indicate continuity of understanding without any signs that strongly indicate continuity of understanding is a sign that the tradition under investigation is dead. Below are common signs.

Signs of traditions of knowledge

These are listed roughly in order from best to worst indicators of a living tradition:

• The production of a notable effect (e.g. powerful generals, well-balanced swords). It is possible for a notable effect to be produced without understanding, for example by following a set of instructions. In practice, though, the production of notable effects requires actual understanding because effective action is too complex to be captured in instructions.

• Shared methodology (even if not explicitly stated)

• Shared concepts (even if under a different name)

• Shared conceptual framework or theories

• Extension of the theory in the tradition (i.e. new ideas based on shared concepts)

• Master/apprentice relationships

• Explicit knowledge of specific arguments

• Shared terminology

• Accreditation (depends on quality of accreditation system)

• References to specific authors

• Familiarity with a person’s works

• Existence of a physical location where the tradition is ostensibly kept (e.g. a prestigious university)

A Cautionary Note

It’s important to remember that in order to trace traditions, you have to investigate the actual transfer of knowledge. This means that you can’t, for example, rely on the existence of a physical location where the tradition is supposedly kept to justify that the tradition is alive. There are many possible scenarios in which a tradition has died or been lost, and yet the physical location has been preserved. A useful way of determining whether a tradition of knowledge exists and is living is by investigating chains of master/apprentice relationships. When looking at the works of masters and apprentices, you can tell whether there are shared methods, concepts, ideas, and so forth. Furthermore, the existence of master-apprentice relationships at all is an indicator of a living tradition, because master-apprentice relationships are especially effective means of knowledge transfer (this is borne out by the historical record).

Live Traditions

What keeps a tradition of knowledge alive? First, let’s review our definition of a living tradition of knowledge: A living tradition of knowledge is a tradition in which the body of knowledge has been successfully transferred, i.e. passed on to people who comprehend it.

Features of living traditions

Apart from the transfer of the tradition’s knowledge itself, there are features that traditions can have that promote their survival. For example:

• Transfer of verification mechanisms, i.e. mechanisms to check the body of knowledge against reality

• Transfer of mechanisms to check the transferred body of knowledge against the original body of knowledge so as to correct errors in transmission

• Transfer of the generating principles of the body of knowledge (which allows people to verify, correct, and extend the theory), like theorizing techniques

• Explication of the generating principles of the body of knowledge and transfer of this explicit knowledge. This is different from transferring the generating principles themselves, which must be understood implicitly to be truly transferred.

• The production of masters, as opposed to mediocrities or even experts. Masters are more likely to be capable of preserving, extending, or reconstructing the tradition as necessary.

• Teachers that can reliably assess whether students understand the knowledge, to prevent the Counterfeit Understanding Problem, explained below

• An institution dedicated to keeping the tradition alive

• Institutional defenses against the takeover of the institution, e.g. a test or requirement for entry

Remember: traditions of knowledge are preserved intentionally. It’s hard to keep a tradition of knowledge alive.

Dead Traditions

The overwhelming odds are that traditions become lost or die. Decay is the default; entropy usually prevails. This can happen for many reasons, including:

The Problem of Counterfeit Understanding

Students of a tradition can appear to possess understanding of a tradition’s body of knowledge despite actually lacking it. This is counterfeit understanding. This can happen if students merely reproduce the teacher’s verbal behavior, are trying to guess the teacher’s password, or are simply cheating. This can also happen if teachers cannot correctly assess whether the students have achieved real understanding.

Some types of knowledge are particularly vulnerable to counterfeit understanding, such as knowledge about introspection, which is quite difficult to verify. Even types of knowledge that we might think are robust to counterfeit understanding may not be. Don’t make the mistake of thinking that institutions that produce material effects, for example, have an easier time transferring knowledge.

There are a number of sub-problems that exacerbate the problem of counterfeit understanding:

The Problem of Standardized Education

Standardized education is useful because, among other things, it is easily scalable, but standardized methods of education (e.g. standardized tests as a means of assessment rather than non-standardized evaluations by masters) tend to produce counterfeit understanding because education is too complex to be easily standardized.

The Problem of Purported Change of Purpose

Sometimes counterfeit understanding will be concealed by hiding the resulting loss of capacity as change of purpose. If a country has failed to keep the knowledge of how to make swords alive, for example, they might conceal it by saying, “We don’t need to make swords! The style of combat has changed to favor spears.”

The Difficulty of Recognizing Understanding

Being able to tell whether people have true or counterfeit knowledge is a difficult skill. Even a master in the tradition’s knowledge itself may lack this ability.

The Lack of Awareness of Implicit Models

People who don’t understand the distinction between implicit and explicit models, and who thus can’t or don’t transfer their implicit models, will fail to transfer the actual body of knowledge, unless the entire body of knowledge has been successfully made explicit, which is exceptionally difficult.

The Problem of Lost Generators

If the generating principles of a tradition’s body of knowledge are not transferred, then students of this tradition won’t be able to re-generate knowledge that has been lost (and the loss of some knowledge is practically unavoidable) or generate new knowledge that builds upon the tradition. Barring complete knowledge transfer by every generation, which is extremely difficult, this will result in the decay and eventual death of the tradition.

The Problem of Syncretism

Syncretism, or the amalgamation of different schools of thought, is a moderately negative sign that people may be failing to transfer a tradition of knowledge. While syncretism is fine if it is an upgrade to the tradition, it is often difficult to tell if it yields an upgrade. Syncretism indicates a dead tradition if: (1) people are trying to import something into a system that doesn’t make sense, (2) people are importing things because the original tradition stopped making sense to them, or (3) if the institution which has served to transmit the knowledge has been captured (see below).

The Problem of Creating a Single Point of Failure

Although creating an institution dedicated to transferring a tradition of knowledge is very useful, and is necessary to preserve a tradition in the long run, it can also be dangerous. By institutionalizing a tradition, you can also introduce single points of failure. The bad judgement of one teacher at an organization, for example, can yield a whole class of students whose thought is severely damaged.

The Problem of Institutional Capture

If an institution built to transfer a tradition of knowledge gains power or prestige, it will attract people who want to use the institution for other purposes than the preservation and development of the tradition. Once the institution is captured for the power it holds, and the goal of the organization is no longer to transfer the tradition, the body of knowledge can easily fail to be transferred. Some types of knowledge are extremely vulnerable to institutional takeover, e.g. traditions involving political theory, because every social theory is also an ideology.

There are various ways to defend a tradition from death by institutional capture. One way is simply to understand the tradition — it’s much easier to defend it if you understand it, because others can’t distort it while you’re unaware. Another way is to tie resources to the propagation of the tradition, e.g. by dedicating a grant to fund people who only work on certain texts. Implementing these defenses, however, is tricky. If you overdo the defense mechanisms, they may prevent the successful transfer of knowledge. You can imagine a grant tying people to a particular work being detrimental if actual understanding is achieved by reading a different work, and there is no financial incentive to read that work. On the other hand, if you underdo the defense mechanisms, and the institution is captured, the tradition will die just the same.


Comments sorted by top scores.

comment by Ben Pace (Benito) · 2018-03-08T20:08:50.679Z · score: 38 (8 votes) · LW(p) · GW(p)

(This comment is me trying to generate my own signs of successful traditions of knowledge, before reading most of the post. Is fairly rambly #BetterWrittenQuicklyThanNotWrittenAtAll)

My first thought on what signs to use when looking for the life of traditions, is to borrow from models of hiring (a company's hiring process is what determines whether the core insights of a company have been successfully shared to all its employees, so it seemed analogous). Some quick examples:

  • Does it depend on short interviews or extended trial work with current core members of the org? If the former, the company culture will be eroded quickly, and if the latter, much less likely.
  • Does the hiring process heavily use the contacts of the current employees? Using top recommendations from existing teammates can quickly zoom in on those with the most appropriate competencies and thinking patterns.

For intellectual traditions, I'd look to how much selection the practitioners of the art had on who followed them, and whether they were able to train them a great deal. For example, if you're trying to continue a tradition of building compilers a certain way at Stanford, if you're a professor there you'll get to choose your postdocs from a wide set of excellent PhD graduates. If you're able to work with a number of them first too, then you get a really strong sense of how they all think. An extra fun thing is that you're able (with such a technical field as CS) to pick technical problems that you give people when they apply, and that you swear them to secrecy on, so that nobody can learn them in advance and practice for them. Pick a question that's a good indicator of the core technical understanding, and you're off to the races.

Any process of passing on knowledge that doesn't have these attributes will be correspondingly worse.

Another thought on this is from HPMOR. As it says in the book, the difference between Minister for Magic, and Order of Merlin is that successive Ministers are voted in by the wizarding public, whereas the Order of Merlin is handed on by personal choice of the previous holder of the Order of Merlin at their death - and this is why the Minister of Magic is Cornelius Fudge, and the Order of Merlin is Albus Dumbledore. An unbroken line from Merlin himself. This is related to the students of Noble Prize Winners often being incredible themselves (something I've been told but never verified myself FYI).

So these are hiring analogies. I expect that most of Samo's examples will be natural to this framework. If not... then it's probably due to looking at different scales of the process. For example, you can guess if a company has a good hiring process if they're successful, so you can look back at great companies and suppose they're successful. Analogously, if a field (e.g. mathematics) consistently produces great insights over the course of a century, you can guess that the tradition was alive. And at the other end of the scale, you can probably look at the general rationality signs of the person doing the hiring, as opposed to the process itself.

Okay, that's my brainstorming.

goes and reads post


  • I notice I had thought some of the points, and dismissed them. For example having shared methodology, concepts or conceptual framework seemed to me too easy to goodhart [LW · GW] on, so I dismissed them.
    • However, (a) they are still bayesian evidence toward the tradition being passed on correctly, and (b) if there is little incentive for people to game the system, then they're perfectly good metrics.
  • I passingly mentioned judging output, and I agree with the OP that this is in fact the strongest signal. (I used a similar one myself here [LW · GW].)
  • Existence of a physical location did not even slightly enter my mind. But it totally fits the company framing - I've worked on remote teams before and they're terrible at producing quality work together and transferring insights to each other.

Overall I had an okay basic framework, but the post had a delightful amount of detail that I either hadn't noticed or hadn't made explicit, so I give myself a C+ grade. (And now I'm super excited for sequels to this post!)

comment by Samo Burja · 2018-03-08T22:15:40.821Z · score: 28 (6 votes) · LW(p) · GW(p)

Excellent exercise! It seems a good way to evaluate content, initially building out or explicating one's own models before comparing it to someone else's, I will remember to do this.

I'm happy to hear you've found the model I present on useful and interesting!

comment by Ben Pace (Benito) · 2018-03-08T23:21:32.216Z · score: 6 (1 votes) · LW(p) · GW(p)

Thanks. The important parts I neglected to do (due to time, will see if I can come back and finish in next few days) are:

  1. Figure out what explicitly to add to my thinking that would’ve given me an A+. What simple models was I missing?
  2. Ask if there was any way I could’ve figured it out from first principles in advance, without reading your post and making the update in part 1.
comment by Stuart_Armstrong · 2018-03-16T19:53:30.123Z · score: 29 (7 votes) · LW(p) · GW(p)

I'm finding this a bit vague, to be honest. More examples would be good. For instance, modern mathematics is a living tradition, certainly, but modern mathematics would not help if you wanted to answer "how would Archimedes consider the Riemann hypothesis?" Now, Archimedes's school certainly died, under a Roman sword, but a historian with some mathematical knowledge might be able to have a guess.

I think the post is conflating three things: a) Can anyone today say something correct about the origin of the tradition of knowledge, b) has the tradition of knowledge passed down the initial knowledge accurately, and c) is the tradition vibrant and productive today. Mathematics obeys c) but not b), and historians of mathematics can do a) about mathematics, but mathematics itself can't. In my opinion, Catholic ethics obeys a), b) somewhat, and not really c).

comment by Kaj_Sotala · 2018-03-14T09:10:20.121Z · score: 29 (6 votes) · LW(p) · GW(p)

Curated this post for:

  • Discussing an important topic right in the intersection of instrumental and epistemic rationality: figuring out which kinds of sources of knowledge we might be able to trust.
  • Including a thorough list of various considerations related to the question.

Something that could have made the post even better, would have been if it was a more engaging: a long listing of different considerations makes for somewhat dull reading, and I easily found myself starting to skim rather than actually paying attention to everything.

comment by Andrew_Clough2 · 2018-03-15T13:39:54.272Z · score: 21 (6 votes) · LW(p) · GW(p)

This really reminds me of some of what Kuhn wrote about in The Structure of Scientific Revolutions. One important additional way to identify living traditions that Kuhn notes would be that the texts used to instruct students are not the original works where ideas were first outlined but have been re-written to be clearer. The ability to say the same thing in different words is almost central to a tradition's life.

comment by Samo Burja · 2018-03-15T20:56:20.087Z · score: 14 (3 votes) · LW(p) · GW(p)

Indeed, he was one of the major influences on my views of society, epistemology and knowledge.

Mortimer Adler uses the same argument in How to Read a Book to argue for reading original texts, because they are known to have allowed for the transmission of relevant knowledge, textbooks sometimes succeed and sometimes fail at this.

comment by Evan_Gaensbauer · 2018-03-15T21:41:25.103Z · score: 18 (4 votes) · LW(p) · GW(p)

If access to learning mastery in an ongoing living tradition, or having the best chance of reviving a dead tradition if so-called experts of it can't be relied upon, requires access to a physical location, that doesn't seem like an insurmountable obstacle for most people. Elon Musk predicts one day it will be in the reach of the average middle-class person if they wanted to afford a ticket on the trip to Mars. If that's realistic, I think it's within the grasp of the average person to gain access to high-fidelity body of knowledge of a tradition. This includes physical locations. I can think of some common barriers.

  • The very best bodies of knowledge preserved at physical locations may not be accessible to everyone. Most people won't be able to get into the world's top universities where methods are passed down from one generation of award-winning scientists to the next. This goes for other types of institutions as well. Lots of people can teach themselves to code, but that doesn't mean just anyone can end up as an engineer on one of Google's most advanced or important projects.
  • People may be limited to their proclivities and inclinations. For example, somebody who has organic inclinations to the arts trying to make it in a technical field isn't as likely to succeed as they would be in the arts, and visa-versa.
  • In the most conventional sense access to educational institutions which preserve traditions of knowledge are heavily geared toward youth. Not just that it's odd to think of someone in their thirties being accepted into an undergraduate program at Harvard, but that so much education is designed to fit the mindset of younger student bodies. I think if we think of 'tradition of knowledge' more broadly, someone dedicated to gaining access to a high-quality body of knowledge at a physical location could do it well into being middle-aged. I don't know if that gives enough time for people to gain mastery, although they could become experts. If someone is middle-aged with enough spare time to dedicate themselves to pursuing mastery of a tradition of knowledge, I don't know how much of a predictor that would be someone isn't cut out to specialize in traditions of knowledge. By then a lot of people are into careers which de facto constitute at least expertise in a field, or are doing other things at that stage of life which occupy most of their flexible time. Not fitting a mold predisposed to pursue expertise/mastery in a tradition of knowledge doesn't mean people can't pursue all kinds of other valuable roles. If someone has had potential for so long while it remains unfulfilled it could be a strong indicator of unfitness to pursuing mastery in a tradition of knowledge. I expect at least expertise in a lot of traditions of knowledge is accessible to people who are middle-aged.

Adjusting for these considerations, it seems within reach for most people to gain access to a physical location which preserves a tradition of knowledge with enough fidelity to allow at least expertise to be gained in many different traditions.

comment by Samo Burja · 2018-03-17T01:38:29.383Z · score: 14 (3 votes) · LW(p) · GW(p)

This is an excellent analysis of a particular aspect. Firstly I do want to emphasize that this mindset is already very rare, few people reason from the existence of traditions of sound knowledge and begin thinking how to access them. The exercise for most remains, pardon the pun, academic.

The only thing I would add is less of an emphasis of universities and more on particular institutions such as branches of government or particular companies. This represents a kind of institution that, assuming a sound career or skill-set, those in middle age are better positioned to understand and make use of. Particular social circles can fulfill this function as well. Further those in middle age can and do easily gain access to postgraduate education of high quality but seem to do so less frequently. As a very practical example I could cite Robin Hansons reorientation towards the social sciences after a successful stem career.

comment by ChristianKl · 2018-03-10T06:25:03.715Z · score: 18 (4 votes) · LW(p) · GW(p)

A lot of traditions of knowledge die because they aren't useful and lose out to competitors and that's okay. Often useful bits get used by other traditions and the work of the founder of a tradition wasn't in vain.

comment by cousin_it · 2018-03-15T14:50:59.327Z · score: 9 (2 votes) · LW(p) · GW(p)

Yeah. I think curiosity is the main ingredient. If you're curious, you can restore a tradition from a book just fine, and probably add your own stuff in the process, as Renaissance scholars did.

That leaves the question of whether curiosity itself is culturally transmitted. But I think it's innate, children have it when parents don't.

comment by Chris_Leong · 2018-03-09T02:04:29.551Z · score: 14 (3 votes) · LW(p) · GW(p)

I feel that those post neglects the potential for traditions to be revived. Maybe the revival will be different from the original, but it very well may constitute an improvement as they can learn from subsequent reactions to the movement. Traditions also have a tendency to get locked into certain flaws and those who revive it have an opportunity to avoid at least the most obvious of those.

comment by Kaj_Sotala · 2018-03-09T21:19:11.424Z · score: 17 (4 votes) · LW(p) · GW(p)

One interesting example of traditions being revived might be the Historical European Martial Arts community: there is no unbroken lineage of teachers who would have kept the art of fighting with, say, 15th century weapons alive to this day. However, there are surviving manuals that were written by masters of that time, and communities have sprung up which have taken the manuals, started training according to them, and turned the whole thing back into an actual living martial art where people are once again figuring out the best ways to fight using these weapons.

comment by habryka (habryka4) · 2018-03-08T19:18:29.880Z · score: 11 (3 votes) · LW(p) · GW(p)

Promoted to frontpage.

(Also, this is probably my favorite post in this sequence so far, and I quite enjoyed it. Thank you for writing it!)

comment by habryka (habryka4) · 2019-11-29T21:11:11.751Z · score: 6 (3 votes) · LW(p) · GW(p)

I think of Samo's posts, this one was the one that stuck with me the most, probably because of my strong interest in intellectual institutions and how to build them. 

I've more broadly found Samo's worldview helpful in many situations, and found this post to be one of the best introductions to it. 

comment by vedrfolnir · 2018-03-09T21:58:55.027Z · score: 6 (2 votes) · LW(p) · GW(p)
Let’s say you are trying to understand what Aristotle would think about artificial intelligence. Should you spend time reading and trying to understand Aristotle’s works, or can you talk to modern Aristotelian scholars and defer to their opinion?

How much can a tradition tell you about the opinions of its founders?

To what extent do the various traditions of Christianity fit your definition of a living tradition? How much of the content of the most 'living' Christian traditions comes from Jesus, and to what extent does this content reflect Jesus' actual opinions or those of the early Christians, insofar as this can be known?

You can ask the same questions of any other tradition or genus of traditions. Confucianism, for example, or some school or other of Buddhism.

(Presumably, any difference between living traditions reflects some sort of change in at least one of them -- unless 'free variation' existed at an early stage, and different descendants converged on different resolutions of the variation.)

It's possible for traditions to be heavily modified upon contact with other traditions without syncretism taking place, and even with deliberate esoterogeny. I've read that Indonesian pagans, upon contact with Christianity and Islam, set about trying to reform their paganism along Abrahamic lines -- not with the aim of becoming more like Christianity and Islam, but with the aim of preserving their tradition and its distinctiveness from the Abrahamic religions, by reforming paganism into an equal of Christianity and Islam, a "serious religion". Of course, it just so happens that the models for "serious religion" are the Abrahamic religions.

Are there traditions that are unusually good at preserving their contents? How did these traditions preserve them? Of course, this requires having a definition of 'contents'. The Rig Veda had been around for a thousand years before writing came to India, and the earliest known manuscripts of it date to about fifteen hundred years after that (although there were almost certainly earlier manuscripts that decayed or were lost) -- but it was mostly unintelligible until modern historical linguistics, and the proposed interpretations were mostly off the mark. Even now, a lot of it is obscure. But the text is very well preserved.

comment by Swimmer963 · 2019-11-29T22:04:19.740Z · score: 4 (2 votes) · LW(p) · GW(p)

read for research on my EA/rationality novel project, pretty relevant, well laid out

comment by JohnGreer · 2018-03-27T09:30:22.684Z · score: 3 (1 votes) · LW(p) · GW(p)

For those that like this topic, the fiction book The First Fifteen Lives of Harry August is probably worth reading.

comment by leggi · 2020-01-04T05:49:48.205Z · score: 1 (1 votes) · LW(p) · GW(p)

I'm struggling with clarity in my head using the current labelling:

Living traditions - makes sense.

Dead traditions - Dead naturally brings me to 'dead and gone' - where as they are still being transmitted but have lost their meaning. I think corrupted, lost in translation ....

Lost traditions - works better for me when I've re-labelled dead traditions as corrupted (or another better word that hasn't jumped to me yet).

comment by Yannick_Muehlhaeuser · 2018-03-31T09:10:28.524Z · score: 0 (2 votes) · LW(p) · GW(p)

Interesting overall, but more examples would have been helpful.