Resartus: a social revision engine

Thanks to all UR readers who have returned for our fall term. No, it is not September yet, but the air is full of fog and brown apple moths, and my little daughter is at five months. I am delighted to report that she holds her own bottle. She also has a brutal, laserlike, almost creepy stare, and Mrs. Moldbug fervently denies that she in any way resembles Hitler. I suppose her hair is lighter.

And thanks again to those who contributed comments in the OL series. I still intend to edit, or at least select, the comments to construct a coherent thread of response. Or at least one as coherent as the essays themselves. I also will tie up the many loose ends left dangling. But not, as St. Augustine put it, just yet.

Many interesting things have happened over the last month. I will discuss these things, but again, not yet. The next UR post will appear not on the 21st but the 28th, and it will be called something like “US foreign policy from Hamilton to Herron to Holbrooke: two centuries of mendacious, counterproductive bungling.” But in the meantime, there’s always the War Nerd.

Today, though, we’re going to look a little more at the solution I proposed: Resartus. A couple of people have built systems roughly related to the proposal: Lex Libra at resartus.thinkernews.com, Daniel Nagy and Baldvin Kovacs at thiblo.com. Lex has also started a Google group, resartus. Please join this group if you’re interested in contributing.

I have registered resartus.org, but I have no intention of running the project. Since the golden rule of software is that who writes the code makes the rules, my suggestions are just that: suggestions. Following the old Roman design of the dual executive, I appoint Daniel and Lex the consuls of Resartus, with joint plenary power. It is their bag, man.

The consulate is permanent, or at least indefinite. Each consul is free to appoint his or her own successor at any time for any reason. I will assign the domain in accordance with the consuls’ wishes, no matter how corrupt, incompetent or tyrannical they may prove to be.

Let’s take a moment to meet these people. Dr. Nagy, of course, is a frequent commenter here at UR. His homepage is here. So is his PGP key. Daniel is that kind of guy.

As for Lex, I have no idea who he is. I have never met him. (I’ve never met Daniel, either, but at least we’ve Skyped.) Lex’s handle is only slightly less ominous than mine. And while I promised to protect his privacy, I can’t resist revealing this short biography, which may of course be entirely fraudulent:

I currently do coding and product design for a venture backed startup in Cambridge, MA. Like you, I was once a prototypical progressive. I attended high school at not Andover—MM and then majored in history at not Princeton (graduated 2006). During college I interned on Capitol Hill, volunteered on campaigns, worked in City Hall etc. Gradually, I observed firsthand the many pathologies of government. Midway through not Princeton I taught myself to code and dove into the tech startup world. It’s amazing how your perspective on politics changes when you start asking, “How can I start a business to solve this problem?” rather than “How could a government program solve this problem?” That change, along with numerous Navrozov moments, soured me on both the university and progressivism.

I think I first found your blog last February when someone submitted your post about the financial crisis to programming.reddit. I followed your links and started reading Rothbard and DeSoto, plus Hoppe, Leoni, Szabo, Sailer, Stefan Zweig, Grand Duke Alexander, etc.. It’s been a fascinating trip. So thank you!

You see the grade of young minds we’re corrupting here at UR. Quality over quantity. If Lex gives you any trouble, however, you can address him derisively as “Princeton.”

In any case, Resartus is theirs—to (a) see through, (b) screw up or (c) let die on the shelf. The idea is out there, and if they don’t do it I suspect someone else will. If you are not interested in working with Lex and Daniel, even if they are not interested in working with each other, DNS has no shortage of names. Feel free to fork. I will use this space to promote anyone who builds anything even vaguely inspired by the idea.

But I do want to use this week’s post to clarify my own “vision” of Resartus. After this, I will butt out. Since Daniel and Lex have irrevocable plenary power, they are of course free to ignore anything I say. They are also free to ignore anything you say. My impression, though, is that they are sensible people who know how to listen, and I expect they will listen to you as well as to me. Obviously, nothing like Resartus can happen without an enormous community contribution, in design and administration as well as mere content.

For simplicity, I’ll describe Resartus in the present unconditional tense, as though it actually existed and I, not Lex and Daniel, was its designer and administrator. Please remember that neither of these statements is true. The consuls may or may not use any of the ideas below. If you don’t like their decisions, please don’t complain to me.

Resartus is a social revision engine. Perhaps the easiest way to explain this is to explain what it is not. Here is an example of what Resartus is not: createdebate.com. (Thanks to Alan Gooding for pointing me to this site.)

Here is the first “debate” I saw when clicking there today. Its creators, who I suspect are young men very like Lex, and who have clearly put a lot of hard work into the venture, must be tearing their hair out over “Who is the badest Celebertiy?” and its like. But what can they do? Alea jacta est. The Onion, as so often, nails it.

A social revision engine is not a chat board. If it degenerates into a chat board, it is dead. The world has no shortage of chat boards. It may even have an excess. Resartus is designed to complement Wikipedia—a remarkably valuable and useful service, though untrustworthy in general and often malignantly deceptive on controversial issues. Think of it as Wikipedia for controversial material and (perhaps eventually) original research, and you won’t be too far off.

One of the conclusions this leads me to is that Resartus, at least as such, is not cut out to be a classic “Web 2.0” or YCombinator style startup, like the unfortunate createdebate.com. I.e.: it cannot be corporate. It needs to be transparent. It could have a corporate side and a transparent side, but it needs to separate them awfully well. Imagine how many people would have contributed content to La Wik if its domain had been, say, cyberpedia.com, and their articles had been sucked into a one-way database and surrounded by ads. Google may be able to get away with this, but you can’t.

(As a veteran of more than one Silicon Valley bubble, my feeling is that the Web 2.0 era is starting to feel a little played out, anyway. One large problem is that, with “cloud” services like Google App Engine, Amazon EC2, etc., hosting an application is starting to verge on the trivial. The lower the fixed cost, the more the community model outcompetes the capitalist model—no capital, no capitalism. Another problem—especially for Resartus—is that smart people don’t click on ads, at least not unless they’re actually searching for products and services. Over time, I expect transparent social networks to outcompete corporate ones. Maybe it’s time for Web 3.0.)

This is not to say that a buck cannot be made off the thing, if it succeeds. Look at the antics of Jimbo and his friends. But success is a prerequisite, and it’s my impression that resartus.org would outcompete resartus.com, even if the latter was not owned by some spammer. If you are full of piss and vinegar, please feel free to prove me wrong.

In any case: on to the product. It may not be commercial, but it remains a product. This means it deserves what, in the biz, we jokingly call a “PRD.” I will avoid elaborating on this acronym. Trust me, you don’t want to go there. Let’s do more or less the same thing, but make it fun.

A social revision engine exists to help you, the reader, make up your mind about a controversial issue without appealing to external authority. For example, Wikipedia’s policy suggests:

Material that has been vetted by the scholarly community is regarded as reliable; this means published in peer-reviewed sources, and reviewed and judged acceptable scholarship by the academic journals.

Material from mainstream news organizations is welcomed, particularly the high-quality end of the market, such as The Washington Post, The Times in Britain, and The Associated Press.

Here at UR, we refer to these fine institutions collectively as the Cathedral. Note that La Wik does not stoop to filling us in as to why we should believe the Cathedral. It is simply infallible, like the Vatican. Om mane padme hum. “Trust the computer. The computer is your friend.”

The process by which the “scholarly community” and the “mainstream news organizations” produce their reliable material is quite different from the process by which the Vatican produces its. The claim that the former is infallible—or even nearly infallible, or even fallible but eventually convergent toward the truth—is not one to be scoffed at. The Cathedral is a grand old edifice, a fabulous achievement of Western civilization. It is full of many fine people, many of whom do excellent work. As a whole, I don’t trust it at all and I think it needs to go. But this is just my own two cents. If you do trust the Cathedral, you have much less need for Resartus.

Need? Well, need is a strong word. You don’t need anything, besides oxygen and oat mush. But you should want to use Resartus. Or at least someone should. As with any product, we need to start by considering the users. No users, no traffic, no nothing.

Users of Resartus come in three categories: readers, writers, and developers. The developers are the people who build and administer the site—such as Daniel and Lex. Developers should not be writers; this is a conflict of interest.

Initially, everyone who visits Resartus will be a writer. The elusive “pure reader” will only arrive after a considerable degree of success. Nonetheless we discuss this individual, because every plan is a plan for success.

The Resartus reader—call her Janet—wants to make up her mind on some controversial issue. This issue is relevant (important enough to be worth Janet’s time). It is binary (it can be defined as a question whose answer is “yes” or “no.”) And it is disputed (plenty of people are strongly convinced of each side of the question). For example, “who is the badest Celebertiy?” is clearly disputed, but it is neither relevant nor binary.

Janet reads Resartus because she is confident that it provides her with the best available perspectives on both sides of this controversy. Moreover, these arguments need not be excavated from a comment thread, or even from two comment threads. They are structured and organized for Janet’s benefit.

I like the word trial, or (interchangeably) case, for a dispute on Resartus. The site is certainly not designed to be used for trials in the judicial sense of the word—at least, this is not a market requirement. However, a judicial proceeding is the epitome of relevant, binary and disputed. Janet is a juror in the case. For whatever reason—perhaps she just wants to know the truth—she intends to decide it. (This is her job, not Resartus’s. We produce no verdicts.)

In my opinion the rocket should be a considerable distance from the ground before anyone tries to use it to second-guess the actual judicial system. Think of the libel issues. But there are some exceptions already: I think Resartus would be a fine tool for exploring the cases of Bruce Ivins or Floyd Landis. (Indeed, the Trust But Verify blog has taken a very Resartus-like approach to the Landis case.) And do we really know who killed the Kennedys?

Every rocket, however, must start on the ground. With no flammable materials nearby. So I think the best subjects for initial Resartus trials are scientific and technical controversies, preferably ones which have not (such as global warming) experienced democratic polarization.

Technical trials are as far from the “badest Celebertiy” as we can imagine. I am especially fond of them because in many cases, I simply have no idea who is right. For me these cases include string theory, peak oil, and polywell fusion. Each of these can be phrased as a triable proposition: string theory is not a science, global petrochemical production is likely to decrease in the near future, Bussard’s polywell may be a viable solution for energy generation. These are not issues which Joe Sixpack has much of an opinion on, but they are certainly relevant and more or less binary, and you will certainly find strong views on both sides of each.

Social networks, in general, degrade over time. I am very conscious of this because my first social network was Usenet. Usenet in, say, 1990 had some excellent things going for it: (a) the average IQ of someone with a Usenet account was about 120, and (b) almost all accounts were administratively responsible. When this changed, Usenet was no longer viable. There is really nothing like the old Usenet today.

Resartus needs mechanisms to prevent such degradation. Many mechanisms. As many mechanisms as possible. The simplest one: start at the top. If you can resolve controversies in high-energy physics, you can deal with global warming. If you can deal with global warming, you can deal with the Russo-Georgian war. But if you start with the “badest Celebertiy,” or anything close, there is no hope.

It may be hard to round up a quorum of high-energy physicists. So another fine source of early trials is our old friend, the software industry. Emacs versus vi would be a fun trial. Or Python versus Ruby. Or even Linux versus Windows. The relevance here is debatable, but fun is often a good replacement for relevance.

But I am assuming a non-obvious design decision here. There is another way Resartus, at least at first, is different from createdebate.com: it hosts one main trial at a time. Think of Resartus as a hall of justice, with one courtroom. All cases are tried, sequentially of course, in that one room. Eventually the hall may expand and have several courtrooms, but not until necessary.

Why? Our goal is to create a critical mass of high-quality discussion. To do this, we need a critical mass of high-quality writers, preferably generalists who can dive into as many different subjects as possible. Especially at first, this is a very limited group. If we divert this group off into 74 different permanent ongoing arguments, we have no critical mass in any of them. Imagine creating Usenet all over again: how would you do it? You would start with one group, misc.general, then split it and split it again as the thing became unwieldy.

(Moreover, if we only have one courtroom, we have an excuse to put deadlines on our trials, which should focus the attention wonderfully. Two weeks, for example, should be enough to solidify the major points on just about anything.)

Of course, completed trials are not deleted. They are moved to the back burner. Discussion may not even need to be closed, although it probably should at first. Ideally, in a successful Resartus, a wide variety of trials are continually maintained and updated, so that Janet can get the dirt on whatever subject she wants to understand. For example, the ultimate use of Resartus might be a complete revision of modern history, with trials on every political controversy for the last 200 years. Who was right in the War of 1812? I really don’t know. And I would like to. But the rocket has to get off the ground first, and creating a community is like making gunpowder explode: it takes compression.

For scheduled trials, you need a scheduler. Initially, at least, this must be the developers. Lex and Daniel do not just write code; they decide what Resartus is going to focus on, when. They schedule, configure and administer the trials. They do not rely on random IP addresses to submit questions like “who is the badest Celebertiy?” Crowdsourcing has its limits.

Moreover, we have yet to answer the difficult problem: how to ensure that, from a world of random IP addresses, we somehow construct the strongest possible arguments on both sides. This is a mission-critical feature for Resartus. If Janet cannot be confident that she is seeing the best case on each side, Resartus is useless to her. She has no way of knowing that she is watching the string theory C-team go up against the loop quantum gravity A-team. She may know an ass-kicking when she sees one, but this kind of ass-kicking tells her nothing.

It may be possible to solve this problem with a karma system, like Slashdot’s, in which quality is determined entirely from peer votes. I am not a believer in democracy, but Slashdot has done a pretty decent job with their moderation system. So have Reddit and Hacker News. Certainly Resartus needs something of the kind. There are many problems with nondirected moderation, but one of the main ones is that people vote for content rather than quality. It goes without saying, or should, that plaintiff’s lawyers should not be voting defense lawyers up and down. Since the general Resartus approach is to separate the sides of a case—more on this shortly—we avoid this deadly pitfall.

There is another mechanism, however: human editors. I would like to think it’s possible to construct a quality filtering system which is entirely user-generated. But the only way to test this is to compare it to the work of a human, and a good one.

In other words: it may be possible to produce a flat, purely crowdsourced trial which still satisfies Janet’s needs. It is a goal. It may be possible to reach this goal. Or not. If so, it will take a lot of tuning and community-building.

Before this point, however, scheduling a trial involves securing the time of at least one editor. Following the judicial metaphor, the editor is like the attorney. (Attorneys in a normal judicial trial do not solicit input from the spectators, but there’s a first time for everything.) Editors are appointed by the developers for each trial. They may be experienced Resartus writers, or guest experts from outside the community.

For example, a trial of string theory is one thing. A trial of string theory in which the prosecution is edited by Peter Woit and Lee Smolin, and the defense is edited by Luboš Motl, is quite another. You might get the same results without the celebrities. But you might not.

In general, at least one side of a Resartus trial will be in some way revisionist—i.e., inconsistent with the wise and holy teachings of the Cathedral. The revisionist side is either attacking some canonical belief, or promoting some unconventional one. Either way, without an editor, we can expect the case to be hopelessly disorganized and mispresented. The opposite, canonical side has a much better chance of being able to get by with mere crowd moderation.

I’ve described the ingredients of a trial. Now let’s zoom in a little and take a closer look at the process itself. These details are more shaky than the broad strokes above, and I wouldn’t be surprised if Daniel and Lex just throw them all out.

Like a modern Western trial, a Resartus case is asymmetric. Again, the burden of argument falls on the revisionist, who needs to make a case that the canonical interpretation of reality is wrong. Few will bother, at least at first, using this tool to make a case for the conventional wisdom. By definition, it has no shortage of defenders.

Thus we can describe the revisionist editor (or editors) as a prosecutor (or prosecutors). If this reminds you pleasantly of Cicero and his ilk, it should. Although in the end it was abused and resulted in a few too many decapitations, the Roman idea that anyone could prosecute anyone for anything was one of great coolness. Certainly the Verreses of our day could use a Cicero or two. And in Resartus there are no verdicts, and certainly no proscriptions.

The prosecutor opens the trial by stating a brief, but explosive, conclusion. For example:

String theory is pathological science. Floyd Landis is a clean, upstanding young man. America would be a better place without [ethnic group]. Emacs is a more powerful editor than vi. Lee Harvey Oswald was a stooge of the Knights Templars.

To support this claim, the prosecutor then composes a statement. The statement should be short, but no shorter than necessary. It explains all the logic and facts necessary to understand the connection between Oswald and the Templars, or other revisionist argument.

The content of the trial is an annotation tree against the prosecutor’s statement. Annotations can be added by either prosecution or opposition. They may be local to some part of the statement, in which case they are marked (as unobtrusively as possible) much as footnotes. Or they may be global, with no such connection. Janet may want to follow local annotations, or she may just want to see a list of all annotations.

There are three classes of annotation: exhibits, queries, and objections.

An exhibit is a document that expresses some fact pertaining to the case. The document is to be taken on its own merit; there are no “reliable sources.” However, links to non-Resartus sites need to be archival quality: there should be a reasonable guarantee that the target of the URL will not change, that the URL is not and will not be firewalled, etc., etc. The developers maintain a list of archival-quality link targets. To exhibit a document from a non-archival source, copy it to Resartus.

A query is a sincere request for additional information or clarification. A rhetorical question is not a query. The difference between objections and queries is a matter of taste, of course. But taste matters, which is why we have the category.

An objection is an arbitrary counter-statement. Like the original statement, it is open-ended. It should be short and to the point.

The response to a query is a clarification, which can then be annotated as if it was part of the original statement. The response to an objection is a set of counter-annotations. Thus the annotation ping-pong continues recursively on down, until both sides are satisfied that their point has been made and their opponents are simply dense. At that point, they leave it to Janet.

One of the basic principles of Resartus is that stonewalling is not an effective defense. If Verres does not show up for the event, Cicero can still create a trial that is every bit as damning.

Protection against stonewalling is provided by auto-annotation. If Verres does not query or object to Cicero’s statement of his atrocious crimes, Cicero can add his own queries and objections, and answer them himself. Autoqueries (so familiar in the FAQ form) and autoobjections should be colored, labeled or otherwise distinguished, because like any rhetorical technique they can be abused. However, there is nothing more humiliating than discovering that one’s opponent has anticipated all of one’s objections.

The problem of producing this annotation tree is essentially a collaborative editing process. But unlike most collaborative editing processes, it is the product of two groups, not one. There is no reason to expect the prosecution and the opposition to be able to collaborate, or even engage in a civil conversation. They are enemies. Their aim is to humiliate and destroy each other. Verres’ head is on the block, as is Cicero’s reputation.

Within each side, we can expect great amity and civility to prevail. Both the prosecution and the opposition are teams. All are working for the same victory. The prosecution has the advantage of an editor, and probably for early trials the opposition should have one as well. (Otherwise, the task of prioritizing, editing and (perhaps hardest) unifying annotations must be left to good old voting.)

Prosecution and opposition also need separate discussion boards. Messages on these boards are not annotations. They will not be seen by Janet. They are for internal purposes only. To participate in these discussions, or to enter annotations, a Resartus user needs to pick a team. There is no crossing over. If you decide halfway through the trial that your side is actually wrong, dropping out is your only recourse. Adversarial discipline is essential.

If Janet really exists—i.e., once undecided readers actually show up—it may be desirable to have a third discussion board, for the undecided reader. Undecideds can share their questions and concerns, which will probably be scanned by both prosecution and opposition and raked into the annotation tree. Once they reach a conclusion, however, they must post it and drop out.

Why the separation? Because I have been reading Internet boards for (god help me) more than half my life, and I have never once seen a productive group discussion between two factions. At least, I have never seen such a discussion that couldn’t obviously have been improved by providing a private board for each faction, and a structured arena for them to explore the disagreement. And I also have never seen this technique applied, which means that it either (a) sucks and is stupid, or (b) is totally cool and will take over the world.

Which one is it? That’s for you to find out, if you’re interested. Hopefully someone is. In any case, I think I have laid out enough details to make it clear what I mean by “Resartus.” The actual Resartus is in the hands of Lex and Daniel. Hopefully it will be something cool.

The next UR post will appear on August 28, 2008.