Socratic Fishing in Lake Quora

Allow me to introduce you to Seb Paquet, an expert Socratic fisherman on Lake Quora.  He is particularly adept at baiting the hook just right to catch fish of the species Wannabis Oracularis, to which I belong. He is entirely to blame for getting me addicted to Quora in the last month or so (you can follow me here). For those who haven’t yet heard of it, Quora is a booming Q&A site. It just might be the next big social media site to cross the chasm and go mainstream. It is certainly booming right now, and is the darling of tech watchers. But unlike other recent Valley favorites like FourSquare (narrow appeal) and Groupon (for shopaholics), Quora might well become as fundamental to the Web as Google, Facebook or Twitter. Everybody asks and answers questions after all.

If you think the Q&A market is a tired and played-out ancillary market (lazy schoolkids looking for help cheating on homework on Yahoo Answers, tedious transactional Q&A on LinkedIn, let-me-Google-that-for-you sites), you’d be wrong. Quora has demonstrated that Q&A is a viable fundamental market, not a bolt-on ancillary to other markets like social networking or asymmetric messaging. Hang Zhang first helped me appreciate the very subtle social design lurking underneath the apparently simple architecture of Quora, and Seb Paquet, through his baiting, has provided me, over the last month or so, with a crash course in the dynamics of Q&A. Initially, I thought Quora was a fad, that owed its initial meteoric growth to the pedigree of its founders and early backers. I even unfairly labeled it in my head as “Valley mutual admiration society,” but I have now become a convert.

The DNA of Q&A

Think about Q&A for a moment. Until Quora came along and forced me to think about it, I’d assumed that it was a derivative type of transaction, rather than a fundamental kind. Q&A, I figured, was just a conversational dynamic within any community, and didn’t need specialized treatment. Unlike general socializing (parties, games) and quick bump-into conversations (watercoolers), Q&A is not specially associated with any particular part of our physical social world. So unlike Facebook or Twitter, there are no obvious source metaphors to draw from. Q&A happens everywhere, and nowhere in particular.

Support for Q&A, I figured, was a commodity “cost of doing business” feature on any community site where status is based in part on authority, and there is a conversational dynamic underlying the system.

To understand the Quora phenomenon, you have to think about the phenomenology of Q&A. Consider the different types we encounter in everyday life:

1. Commodity factual questions (“let me Google that for you” or LMGTFY)
2. Test/Interview questions
3. Learning questions
4. Socratic questions
5. Procedural questions (“do we vote now or later?”)
6. Transactional questions (such as asking for directions, or asking about the price of an item in a shop)
7. Privileged data questions (such as asking about the details of an insider deal)
8. Permission questions (Can I/May I?)
9. Ascriptive authority questions (based on formal signs of authoritative knowledge, such as degrees: “Doctor, what’s this rash on my arm?”)
10. Situational authority questions (a news anchor asking an on-site reporter a question)
11. Opinion questions
12. Solicitous questions (“Are you comfy?”)
13. Diagnostic questions
14. Trick questions
15. Entrapment questions
16. Questions people wish others would ask them (faux-FAQs on personal blogs)
17. Insight questions that motivate research (I consider myself pretty good at this kind and have blogged about it)
18. Debate questions that trigger a conversation between people with comparable, but non-identical relevant knowledge bases
19. Flattery questions that are designed to give the person being asked an opportunity to show off
20. Insult questions that are designed to put down the person being asked

I am sure there are many more varieties, but here’s one interesting feature: in Q&A, the relative status of questioner and answerer depends on the type of question, and the answer and response that actually follow. Very interesting sorts of status conflict can emerge when the two parties make different status assumptions, as in the classic kind of bad-faith entrapment question:

Jane Doe: So, what do you think about the situation in Burma?

John Doe: Well… [launches into a 1-minute, speculative mini-lecture based on what he knows]

Jane Doe: You have trouble admitting when you don’t know something, don’t you?

(here, Jane wasn’t really interested in Burma; she was interested in tempting John into a moment of mini-pedantry so she could make fun of it; both assumed status superiority to start, but Jane won the contest.)

Q&A and Status Fluidity

This fluidity of relative status in Q&A is what makes it a fundamental kind of messaging transaction. Status shifts strongly, but locally, during the process of asking and answering questions. Which makes sense, since status shifts generally occur in response to truly new information being injected, and Q&A models are optimized to draw new information in.

By contrast, neither Facebook or Twitter is designed for that.  Your status on those services is dominated by the accumulated, average status of your recent past, and no one action can move that much.  Your current status just gets reinforced over time. It takes a tweet or wall post of extreme stupidity to damage your credibility on Twitter or Facebook. And you cannot build Q&A effectively into either because the accumulated status would be eroded by the acid effects of unbridled Q&A, unless moderated deliberately.

You can see this deliberate moderation (achieved via self-policing) on LinkedIn, where Q&A is muted, since everybody is a little afraid of having a stain on what is effectively their resume (resumes are the main professional status object).

New information can and does enter these systems, but it is strongly filtered by a confirmation bias, either by individuals or groups.  Your LinkedIn resume-profile is going to be self-serving. Your Facebook world is going to socially filter for “friendly.” Your Twitter world lets in new information, but cannot really process it with any depth, only pass it along as shortened URLs.

Which means that Q&A is a fundamental interaction, marked by high, but localized status volatility, and the dominance of current, situational status over aggregate, accumulated status.

Just as Twitter carved out a fundamental role for itself by making one sort of asymmetry explicit (relationship asymmetry), Q&A formats make another type of asymmetry explicit (situational status asymmetry based on questioner/answerer roles). In Q&A, unlike in general conversations, you are your situational role.

In Q&A therefore, each question induces a local universe of status on its own. If you have a lot of credibility from past answers, and a huge following, it still won’t excuse a stupid answer or prevent a smarter upstart from winning the “best answer” game. Q&A is fundamentally a more temporally and socially-localized game than other conversation types. The followers/following ratio, which is so  important on Twitter, does not matter as much on Quora (and for that reason, I predict that celebrities will have much less influence, apparent or real, on Quora than on Twitter, when they discover it).

The Genealogy of the Conversational Web

Q&A is one of the oldest time-suck activities on the Web. It predates the Web and even the Internet actually. It was a core part of the dial-up BBS (bulletin board) systems that used to exist independently of the ‘Net in the eighties. Usenet was the form the idea took on the Internet initially (Usenet has since wound its way, via a tortuous path through Deja, into Google Groups). Q&A was a big part of the conversational activity in those old media.

The overall undifferentiated idea of “bulletin board” has morphed today into an entire universe of multiple conversation models:

• Blog-comment models
• Reblogging and backtracking models (blog-to-blog relays and conversations)
• Comment aggregation (Disqus)
• Email groups
• Wall-to-wall chatter on Facebook
• Hashtag based transient conversations on Twitter
• Web forums (generally bolted on to a community that does other things as well)
• News aggregation/discussion sites.
• General announcement lists/boards

The basic Web forum is easiest to recognize as the lineal descendant of the original world of BBSes, but the BBS is actually the ancestor of nearly all conversation models on the Web.

The one conversation model that was clearly not present in the BBS world was wikis. Wikis are, in my opinion, the only fundamentally new conversation model on the Internet that arose independently of the BBS/Usenet family tree. The basic metaphor for wikis is the blackboard as a collective, harmonized, canonical, erasable, editable, conversational memory. By contrast the bulletin board is more of a cacophony, where the only source of privileged status for a piece of content is recency of creation. The two metaphors are conceptually distinct, and one cannot be derived from the other, which is why wikis had to be invented separately.

Conversation models are fundamental to the Web, and are more fundamental than either publishing or relationship models (the bases of Web 1.0 and Web 2.0 respectively), since specific types of conversations, in specific venues, are what produce content and help you acquire and retain relationships. The taxonomy/genealogy of Web technology is at heart a taxonomy/genealogy of conversation types, with blackboards and bulletin boards as the two original ancestors (perhaps there are others; I can’t think of any, unless you include the basic payment transaction as a financial “conversation”).

You could think of the history of the conversational Web as a series of failed and successful attempts to take a particular conversational dynamic that existed in an undifferentiated medium like Usenet, making it explicit, and adding features relevant to that dynamic. So relationship maintenance conversations were made explicit by social networks. Speech-on-a-stage+discussion session models became blogs. Watercooler conversations turned into Twitter and news aggregation. All were nascent in the original BBS models.

But somehow, nobody got Q&A truly right until now. Sure, there were specialized ones like Stack Overflow, but really, Quora is the first serious candidate. What makes it work?

RTFM, LMGTFY, FAQ

Think back to the single most annoying element of the BBS family tree: newbies joining and asking stupid questions or attempting to restart discussions that old-timers have beaten to death multiple times.

The real-time-evolving BBS model was and remains just plain bad at handling this. And all its descendants have this flaw as well.  Think of a physical cork bulletin board: all the space gradually gets used up, and rude people start to tack new notices on top of old ones. At some point, somebody cleans up by ripping out old notices. Reverse chronology and archiving/trashing is implied in the very physical structure.

And nobody likes to go hunt in the archives, no matter how good the search, or how well organized and categorized the content. If you see a system where there live people and dead (or “archived”) information, your instinct is to ask the live people rather than go digging in the graveyard. In fact, I’ll propose that as a law:

The First Law of Information Discovery: In a social information system containing both identifiable experts and well-organized and searchable information archives, newcomers will always choose to ask the experts over searching the archives, unless the cost of appearing stupid/lazy/rude is too high.

And it is the right instinct, since seasoned old-timers will not only be able to point you to the right piece of dead content faster than search, but will also be most able to answer questions that haven’t been asked, or interpret distant relatives of existing questions.  The “cost of appearing stupid/lazy/rude” is much higher in a tightly-knit community where everybody knows each other and you want to build up a reputation over time. If you are just asking a drive-by question, as a visitor in a community where you don’t intend to stay, you’ll have much less concern about appearing stupid/lazy/rude (security forums are full of people asking stupid questions when they discover a virus problem, and those communities are among the nicest in how they handle the newbies; they don’t tolerate lazy/rude though).

Of course, another reason people ask stupid/lazy/rude questions is because they are too stupid/lazy/rude, relative to the community. More on that later.

The main reaction to the First Law is the well-known RTFM phenomenon: impatient and crotchety old timers telling annoying newbies to go RTFM (“read the friendly manual” is the usual polite expansion of the acronym). FAQs (bottom-up manuals, really) were invented as a band-aid for situations where there was no canonical manual anchoring the discussion theme. LMGTFY is a cousin of RTFM: discovery help delivered with a dose of sarcasm.

So how do you solve this problem?

The Blackboard and the Bulletin Board Have a Baby

Here’s my theory of why it took so long for somebody to get Q&A right (or right enough to merit serious discussion): Q&A fundamentally needs a blend of wiki and BBS elements. And so far, with the exception of Google Wave, the two families have not intermarried effectively. There are other pieces to the puzzle, but this is the main one. Two relatively isolated gene pools had to mix.

This is also why there is a basic problem with creating a “true FAQ” system. People ask the same or similar questions in different ways, so how do you count “frequency?”

There are two obvious elements to solving this problem: redirection and question editing. You should be able to redirect a new “Anyone know the capital of Alaska?” to an old “What’s the capital of Alaska?” and keep editing old questions to reflect the most frequent form/search phrase, thereby giving it its final canonical form.

Any given question is actually what is technically known as a small world in the overall “question graph.” At the core of each question’s small world are basically interchangeable questions, with more peripheral members having subtly different emphases that make them true variants. Each question’s small world is connected to related questions via weak links.

And no, this isn’t 20/20 hindsight on my part. I personally, and several people I know, have had this insight independently. Any decent engineer with a basic understanding of data structures and graph theory will think of this within a few hours of playing around with a BBS system or creating a faux-FAQ for a product.

The trick is executing on that insight right, and getting the key psychological pieces right, and finishing the solution to the point that it solves the RTFM/LMGTFY/FAQ problem that BBS descendants struggle with.

In terms of social psychology, this means that questions need to be owned by the community, but answers by the individual. Quora gets this dynamic almost perfectly right.  The names of questioners are not highlighted in any way. Questions can be redirected or edited without the permission of the original person asking the question, but edits to answers come to you as “suggestions.”

They’ve even managed to have their cake and eat it too: individual answers are owned by their authors, but there is a room for an “answer summary” that may or may not get used depending on the level of dissent in the answer set, and the structural friendliness of the answers to easy synthesis. List generation questions such as “best books of 2010?” can obviously be summarized in true wisdom-of-the-crowd ways, while ideological questions are harder.

You also need a mechanism to correct for the First Law and send newbies to existing questions without annoying old timers, and Quora gets this almost perfectly right as well (and this is an idea I hadn’t thought of or heard of before). Rather than search as the starting point for the experience, they stuck with asking a question as the starting point (i.e., they decided not to fight the First Law directly). But they drive users to search anyway, via a pretty robust autocomplete mechanism, and an integrated cue that takes you to explicit search. So they head off newbies asking stupid repeat questions as early as technically and psychologically possible. I’ll be surprised if this can be simplified further:

There are many more interesting design features to Quora (I could do an entire post on the voting and following models), but this is merely good engineering. The fundamental innovation lies in the marriage of wiki and BBS models and working through the implications correctly, and in a social-psychologically sound way.

Can Quora Grow Past the Good Old Days?

About 10 years ago, I was heavily involved in an early Web forum at Sulekha.com. I was there on Day 1, became one of the opinion leaders, and eventually joined the company and managed the forums for a while. It was on the huge, noisy fishmarket end of the spectrum, as far as these things go. I’ll admit it was fun playing dictator for a while.

There I noticed a dynamic which has since become a widely-recognized characteristic of any conversation-based system: the early opinion leaders get increasingly dismayed by the quality of new recruits and new conversations (the stupid/lazy/rude threshold getting gradually lowered). At some point, after a bunch of hand-wringing “Good Old Days” conversations among themselves, they leave. The phenomenon has a name. This is the evaporative cooling effect that Hang Zhang wrote about. The main symptom is the “Good Old Days” conversational trend. Social systems may or may not survive the mass defections. Some communities happily adopt a lower-status identity, while others are hollowed out and die.

Hang also proposed that one of the main ways to combat evaporative cooling is to use an architecture he labeled “warren” as opposed to the alternative “plaza” model. I riffed on the Warren-Plaza theme a few weeks ago, and threw in one of my own favorite memes, legibility, and in particular, status legibility.

If you don’t have time to read all those posts, here is the 30-second version. Warrens are like rabbit warrens: low global visibility, highly personalized corners where small groups can aggregate/self-segregate. Plazas are like grand public town squares with lots of global visibility. Warrens are maze-like: it is hard to break out of your current corner, and it is easy to get lost. You couldn’t get lost in a plaza even if you wanted to.

In other words, the metaphoric-spatial structure of a warren is (deliberately designed to be) somewhat illegible, and maintains that calibrated level of legibility as it grows, thereby retaining a “small town” feel even as it grows into a metropolis. The illegibility also applies to social status. You can easily see relative, global status when everybody is crowding into a big public plaza, trying to get a glimpse of the King. Much harder when the entire group is broken up into multiple small groups.

“Warrenizing” a site undeniably slows evaporative cooling, by allowing groups to self-segregate. On Sulekha, we had a warren of individual topic boards, and a firehose plaza that consolidated the entire discussion stream at the landing page. When flame wars got too hot, we used to simply turn the plaza stream off, so people were forced to go directly to individual corners. Things would immediately get quieter. Groups would get sequestered, and therefore less likely to leave because of unpleasant activity elsewhere that they didn’t like.

But now that I have used Quora for a while, I am not sure the warren architecture and sufficient status illegibility will be enough. Apparently, I am a latecomer. There is already a lot of Good Old Days chatter on Quora.

In fact, I am leaning right now towards the conclusion that if they don’t make some basic changes,  Quora will die, rather than go downmarket, through evaporative cooling.  One reason is that though the architecture is a warren architecture, one corner is heavily overpopulated (the techie/entrepreneurial crowd).  Unfortunately, the theme that this crowd loves is, I believe, at its limit of “warrenizability.” You cannot subdivide the topics within it in more fine-grained ways, to sequester the crowds more finely. Every user’s Q&A interests have a resolution limit. I am interested in “lean startups.” I don’t want to further confine myself to “East Coast lean startups for consumer content businesses.” Topic-based (interest graph) warrenization is at its limit on Quora. Social-network-based warrenization is also at its limit (it’s just inherited from Facebook and overloaded).

To use an ontogenic metaphor, Quora is a victim of unbalanced growth. It’s like a baby with a fully-developed, brawny and hairy right arm. It is flexing that bicep, and intimidating the rest of the body so much, growth is being arrested. You can’t amputate the arm. You can’t ignore it. So what can you do?

Warrenizing Time

Quora is faced with two problems: calming the limit-of-differentiation entrepreneurial corner, and getting other corners to catch up. The trick to both, I think, is tuning the blackboard-bulletin-board knob more firmly towards “blackboard.”

The baby needs more wiki genes.

The problem here is that the core experience architecture of Quora is built around what is known as a stream: a real-time scrolling reverse chronological flow (pioneered in its “live” form by Facebook and Twitter, but dating back to BBSes). Highly personalized and customized no doubt, but still basically a ticker, zoomed in strongly on “now.” This creates a strong bias towards the bulletin-board end of the architecture (by emphasizing “now”). By contrast, wiki models are typically timeless. You get no time cues when you land on a Wikipedia page, usually.

My colleagues at PARC have built a tool called WikiDashboard that allows wikis to bolt on a more explicit temporal dimension. Quora needs the reverse technology: something that mutes the excessive temporality (to be precise, excessively real-time temporal structure) and emphasizes longer time scales and timelessness.

One way to think of this is that Quora has successfully warrenized with respect to the interest graph and the social graph. But temporally, you still have a plaza. One single, universal time scrolling back at a fixed rate to the dawn of time.

Think about time in relation to questions.

• There are timeless questions, and there are current questions, and there are questions with life-cycles of every size in between.
• There are questions that pop up frequently, infrequently, or clustered around a particular date.
• Questions may be related to each other very strongly, but be too far apart in time (wouldn’t you want to automatically situate discussions of the 2nd Iraq war near discussions of the 1st Iraq war based on time?)
• Some questions evolve to their canonical form very quickly, because lots of people ask it all the time, while others take much longer.
• Some questions are large “small worlds” (lots of variants), while others are tiny worlds (few variants). Question-and-Search works better on the latter. Question-and-Search only works well on the former after the community has invested significant effort in building redirects, and editing existing questions.

Here, I suspect the Quora team has merely taken the path of least resistance for a first iteration, since the basic stream is the most popular architecture right now. There are many obvious attacks on the “warrenize time” problem (an all-time-great questions tab, a rewind button that allows newbies to start consuming a Quora theme from when it gained momentum, rather than jumping into its current state, a way to easily find “most answered” or “most variants” corners of the question graph, graphics that show the lifecycle of a question or set of questions, a logarithmic receding horizon as opposed to linear). Again, all of these ideas are sort of obvious to anyone who’s thought even briefly about temporal organization (okay, maybe I’ve thought about it a more than most, since I am writing a book about it).

But I bet there are social psychological conundrums (like the First Law) lurking here on the time dimension too. It won’t be an easy path.

But I hope Quora makes it. In a way, this is the fundamental social technology that is a most natural fit for me personally. I am not a natural Facebook or Twitter fish. I might be a natural Quora fish.