Warrens, Plazas and the Edge of Legibility

Long-time reader and astute commenter, Xianhang (Hang) Zhang wrote a very interesting post a couple of weeks ago on his blog: The Evaporative Cooling Effect. It is part of a fascinating series he is doing on social software. The post explores a phenomenon that is very close to the status illegibility phenomenon I explored two weeks ago, and in fact draws inspiration from the same Groucho Marx/Lake Wobegon observations that I started with.

Evaporative cooling is basically the effect of the highest status people in a group leaving, lowering the average status of those left behind.

What I found fascinating though, was Hang’s suggestion for how to combat the effect (and thereby stabilize groups). In my post, I proposed that status illegibility helps create the stability. Hang brings in another dimension, which is illegibility in the group’s environment/context.

In particular, in social software (or physical environments for that matter), smarter-than-average early adopters often leave when the “unwashed masses” start to jump on the bandwagon, devaluing the social cachet. Hang proposes that one of the best ways to combat this is to build (or rather catalyze the evolution of) “warren” architectures instead of “plaza” architectures. Here are the pictures that pair of evocative terms produces in my head. You might imagine something else:

Warrens vs. Plazas

A warren is a social environment where no participant can see beyond their little corner of a larger maze. Warrens emerge through people personalizing and customizing their individual environments with some degree of emergent collaboration. A plaza is an environment where you can easily get to a global/big picture view of the whole thing. Plazas are created by central planners who believe they know what’s best for everyone. The terms are very evocative, and should remind you of the idea of legibility in physical environments that we talked about recently, in my post A Big Little Idea Called Legibility. In fact, it wouldn’t be a gross oversimplification to say that warrens and plazas differ primarily in their legibility. There are many subtleties of course.

The warren/plaza concept also sheds new light on one of my oldest posts, Harry Potter and the Cuaron Slam. In that (rather murky) post, I argued that not The Prisoner of Azkaban was the best Harry Potter book, and that Alfonso Cuaron’s movie version surpassed the book. I can now summarize that whole post very briefly with the warren/plaza concept. The entire Harry Potter series is of poor literary quality because most of the books are very plaza-like overly-legible books. There is none of the atmosphere of mystery you get with warren-like books (the Lord of the Rings for instance). The overarching central-planning map overwhelms narratives and character arcs. The 3rd book is the only warren-like book.

Two pictures from that old post get at a different aspect of warrens and plazas. I offered an analogy between the book/movie context and plot and robot path planning algorithms, which come in two basic varieties: bird’s-eye-view map navigation, and the more difficult worm’s eye view navigation where the robot only knows what it can see and remember from a ground view (the main algorithm for doing this is called SLAM, hence the bad pun in the title). The second picture illustrates how SLAM-style navigation works: you gradually build up a map of your environment by remembering what you see in your field of view as you navigate.

Plaza path planning for robots

Warren navigation for robots

Again, the warren/plaza concept is useful.  Plaza navigation is easier, but less powerful, since it requires global information (which is only easily available with plazas). Warren navigation is much harder but much more powerful, since it does not require global information.  In warren navigation, learning is a necessary feature, since you cannot plot a shortest path  a priori. You need to explore and stumble and build up a map while groping towards the goal.

In the movie version, Cuaron’s camera work is literally warren like. Most of the shots are ‘follow the action’ ground-level shots. You feel like a mouse following the action rather than an eagle.

Hang offers Facebook and Quora as examples of highly “warren” like social sites. He also asserts that social sites that fail at a particular scale due to evaporative cooling are typically plazas.

The Edge of Legibility

I think there is a very fascinating line of thought here. I haven’t yet worked out the details, but here’s a potentially powerful conjecture that suggests where to go: social systems that thrive and grow are on the edge of legibility.

All kinds of legibility: status legibility, environmental legibility and probably a couple of other kinds. If they become too legible, they fail in the Seeing Like a State mode or through evaporative cooling.  If they become too illegible, they fail by ossifying, with highly ritualized and sacramental cultures, with a great deal of environmental irrationality and very few entries/exits.

The idea isn’t particularly original. In complex systems, an idea that was very popular in the early 90s (to the point of being a fad) was that rich, growing complex systems were on “the edge of chaos” (a state called self-organized criticality, SOC). Around the same time, in biology, Santa Fe theorist Stuart Kauffman proposed that biological systems self-organize in ways that keep them on the edge of criticality with respect to key chemical autocatalysis reactions (so for example, multicellular organisms evolved because the chemical soup inside a too-large single-celled creature would become supercritical and spiral out of control, while too-small cells would lack the critical concentrations of key molecules to get to autocatalysis at all).

This line of research isn’t dead; in fact it is steering towards the group dynamics stuff we are talking about. In the 2000s, a lot of complexity theorists got interested in what is known as the El Farol Bar Problem (and a related idea called the Minority Game). This line of work explores phenomena that are very strongly related to the Groucho Marx and Lake Wobegon effects. Surprisingly, the right approach to these problems seems to involve methods from statistical physics.

Resurrecting Self-Organized Criticality?

What’s new in what we are discussing here is the idea that complex systems that are also recursively self-aware in some ordinary sense (i.e. we are not talking “meaning of consciousness” here, but merely the ability to act on the basis of models of yourself, your group, and your environment; this can be programmed in simple AIs today) will differ from the old 90s style SOC systems in a critical way: their behavior will also be driven by the legibility of the situation to the various self-aware agents, and self-aware collectives of those agents, and so on recursively outward to the whole dimly self-aware global system.

So there may be an interesting model of social systems (both descriptive and prescriptive) lurking underneath this qualitative discussion: a combination of growth rates, “cell division” rates that keep a system on the edge of legibility that allow baby plazas to gradually morph into adult warrens, with the level of legibility for all actors itself driving the system’s evolution. Call it “Legibility-Modulated Ontogenic Evolution of Self-Organized Critical Systems.”

That sounds like the title of a paper I’d like to write. Pity I have no time any more to indulge in such things.

I am pulling ideas from a lot of different places here, so if you are interested in following this train of thought,  follow this Edge of Legibility trail starting with a review of the Self-Organized Criticality idea. Warning: this is a total geek-fest, and if you aren’t interested in the complexity theory side to many of the things I talk about on this blog, you should probably ignore this. But you should know that there is a radically different way of thinking about all the stuff I talk about in the Gervais Principle series and other posts on this blog, and that the other way is actually more natural for me. It just isn’t as much fun to read.

Another Personal Note

A couple of readers found the personal note I inserted into my last post interesting, so I’ll throw another one in here. Back before I went over to the dark side and actually did hands on technical work, the mathematical aspect of this kind of stuff was my main gig. During the early part of my grad school career, I was fascinated by the complexity theory work, which the mainstream of my field (systems and control theory) regarded as soft, and slightly disreputable interdisciplinary TV science (Santa Fe has a far bigger reputation in pop science than in mainstream academia).

Eventually the fascination led me to steer my PhD direction away from classical aerospace systems/control problems and towards complex systems. But I was uninspired by the main focus of the complexity theory world (things like sand piles, very simple agents like ants, traffic jams and so forth).  For both pragmatic reasons (you can’t manufacture a defensible PhD thesis in aerospace engineering by studying sand piles) and personal reasons, I studied things like formation flight and teaming algorithms for autonomous aircraft and spacecraft (very much the bandwagon research topic in the late 90s/early 00s). For my postdoc, between 2004 and 2006, I went even deeper into such stuff and did a lot of work on mental models and how individual decision makers cooperating or competing in a battlefield might act together given different views of the same world.

It was fascinating stuff, and I had a great deal of fun doing it. It was also stuff that was positioned perfectly to fall right through the cracks between the fields of AI, operations research, cognitive science and control theory. I even threw in bits and pieces of linguistics and the philosophy of language. And “falling through the cracks” is what happened. Interdisciplinary research sounds sexy to people outside the world of professional research, but it is an extremely risky thing to do, especially early in your career. This despite the fact that funding agencies clamor for interdisciplinary work. The publish or perish equation gets massively loaded on the perish side, and I perished.

The majority of researchers manage to dress up their work as “interdisciplinary” while still staying close to the lower-risk core of their disciplinary fields, while a minority take the risk of actually being interdisciplinary and do things so stunning that things work out despite the risk (these are the true pioneers who help reshape the academic landscape). Universities scramble to create positions for these pioneers, and they deserve it.

I wasn’t smart enough to pretend and dress up conservative work as interdiscplinary, and not the sort of genius who could actually make it work. I did good work, but I’ll be the first to admit it wasn’t powerful enough to reshape the landscape in ways that could create room for me.

My interests didn’t align with any established funding sources or publishing channels, so one fine day, after mulling over a pile of faculty position rejection letters and a few adjunct/second postdoc offers that would have had me in a holding pattern over the tenure landing track for a couple more years, I decided to call it quits. I nursed my academia sour grapes for a few weeks, got drunk, got over myself, and headed into industry. This was 2006.

It was probably the best decision of my life. When I look at my peers who are now in faculty positions I realize that I could never do what they do so well. They actually enjoy academic publishing and navigating the warrens of the funding and publishing ecosystems, rather than merely enduring the game as I did.  In a way, the free-form medium of blogging is my home territory, where I can basically do what I like without worrying about publish-or-perish pressures. I have no idea how I deluded myself for so long that I was actually cut out for the academic life.

I do regret that the blogging medium is not friendly to more mathematical exploration though (turning the sorts of ideas I explore in this blog into equivalent mathematical problems would take 100x more time and effort than just writing about them qualitatively).  Maybe someday I’ll be able to retire rich and early, scrape the rust off my math and programming skills, and get back in the game on my own terms.

Anyway, here endeth the geek fest. We’ll get back to your regularly scheduled programming next week.

Get Ribbonfarm in your inbox

Get new post updates by email

New post updates are sent out once a week

About Venkatesh Rao

Venkat is the founder and editor-in-chief of ribbonfarm. Follow him on Twitter


  1. I want to emphasize that I’m emphatically NOT saying Warrens = good & Plazas = bad. I’m glad Venkat managed to tease that out of the post. Rather, Warrens and Plazas both contain different tradeoffs. I’m intrigued with this idea of tying in system structure with SOC. It blends in nicely with my advocacy of how the software has to grow over time to accommodate the needs of the evolving community (ie: you can’t just figure out how the software should look for the successful community and put a new community into it and expect it to flourish.)

  2. Venkat: In this, you focus almost primarily on perception (what can I see & know about the world). For me, the more interesting problem is interaction (how can I affect the world and, more importantly, how can the world affect me).

    The dominant paradigm currently is the “more is better” paradigm. More pageviews, more likes, more comments are universally better than less. This paradigm is right for certain types of social gestures (noticeably: commercial ones) and they are built based on plazas.

    Take gossip for example. Gossip is more valuable for me to tell you precisely BECAUSE I haven’t told anyone else. Gossip is suited to warren like structures.

    • I think the two cannot easily be separated. How you interact is based on how you perceive. Gossip for instance is only a meaningful concept if the social graph is somewhat illegible, creating an uncertainty about who knows what (because you don’t know who knows who and at what level of trust, so you can’t perfectly predict how information will spread).

      What’s even more interesting is how the perception-interaction loop rewires the environment itself.


  3. I guess to follow up on my last comment on the way Facebook wants a McKinsey model for acquiring engineering talent, there appears to be a correlation between offering a culture highly conducive to evaporation and an organization’s innovative capacity through courting highly intelligent high-status individuals.

    I also don’t think the plaza is the right model for structure since all the participants, while formally aranged, are themselves indistinguishable in status from any other.

  4. By the way, you also used this explanation (social systems just on the edge of chaos) in the post explaining why teams of 12 people seem to get the “magic number” treatment… Maybe not directly related, but I think it deserves a mention in this context.

  5. Venkat says: A plaza is an environment where you can easily get to a global/big picture view of the whole thing. Plazas are created by central planners who believe they know what’s best for everyone.

    Where on earth do you get that? As you describe them, both plazas and warrens are fictive spaces that flow from the design imperatives (perhaps unwitting) of the software designers. No doubt they have implications for human interaction, but I doubt that they are anything so simple as the (sigh) false duality you offer. Plazas don’t really exist in nature (though plains and clearings do), and a real warren is more bewildering and dangerous in nature than the exciting, edge-of-legibility description you offer.

    So maybe plazas (the metaphorical kind created by software designers) are designed not to channel everyone in some coercive direction but to provide transparency, or immediacy, or anonymity, or equality, etc. And even if plazas are oppressive (gasp!) in their simplicity, there are reasons people flock to them (and later scatter, as with Chatroulette), of which I’ll offer one: understanding the rules and/or dynamics of a medium and demonstrating mastery of them is a not insignificant salve to those who would rather fall in line that have to recreate the wheel at every turn. If this were not true, it wouldn’t be so easy to observe the high frequency of imitative and by-the-numbers behaviors exhibited by so many suckers lured by the familiar siren song “Be unique, just like the rest of us!”

    • Lemme qualify that… they can arise that way, though that’s not the only way. Cost (plazas are cheaper) and age (evolutionary things start out more plaza like) are other ways.

      And you misunderstood the edge concept: I am proposing that the edge condition is somewhere between plaza and warren. A true warren would be too illegible and a perfect plaza too legible.

      I wouldn’t consider chatroulette a plaza though. But never mind that.

      And I am allowed 3 false dualities/week by law.


      • Venkat says: And you misunderstood the edge concept: I am proposing that the edge condition is somewhere between plaza and warren. A true warren would be too illegible and a perfect plaza too legible.

        Maybe, but I don’t think I misunderstand. More likely I simply wrote too little to explore the issue fully. (Probably wouldn’t realize it if in fact I did misunderstand, now would I?)

        You are saying that a sweet spot exists somewhere on the wilderness side of the continuum, though not at the extreme. Fair enough. But you deride the more civilized side, whereas I believe more people find a zone of comfort there, perhaps even more toward the extreme. But this objection adopts your dualism, which I think simplifies human interactions too much.

        Just to satisfy my curiosity, if Chatroulette isn’t an example of a plaza, how would you describe it?

        • Actually there is evidence that people find the less civilized side more comforting. There is research by Rachel and Stephen Kaplan on what visual images humans find most compelling, and they tend to be something they call ‘complex but legible’ but the examples are what I consider ‘edge of legibility.’ I’ll blog about this at some point. Their book on the topic is out of print unfortunately. But it is based on showing people pictures of stuff and tracking self-reports of anxiety/comfort. This is in the spirit of John Muir’s observation that humans _need_ the wilderness. On the other side, as James Scott’s book shows, plazas are often unusable spaces (as in Brazilia and Chandigarh). The evolutionary explanation offered by some (I forget the ref) is that edge-of-legibility environments offer a good balance of being able to see as a predator, and remain hidden, as prey. Plazas allow you to see more, but also be seen more.

          And I guess my view of plazas may come across as derision, but really it is more about my assessment that they are more brittle/unstable than warrens for a variety of reasons. Or in other words, it takes more energy/work to keep them going. Whether that’s a good or bad thing is a different matter. I believe, like Hang, that there are simply different tradeoffs.

          Chatroulette: it is _conceptually_ legible as far as the explicit design elements go, but not literally so, since you get matched with 1 person at a time, randomly. So that’s more warren-like since there may be emergent dynamics that you cannot get a view of. Twitter is the same way. Conceptually legible because it is so simple, but you can still only ever see a tiny corner of it. So it is not easy to see the emergent dynamics besides what Twitter wants to show you (like Trending Topics).

          BTW, just checked out your site. Thanks for the ‘Collapse’ pointer. Added it to my End of the World Trail.

          • Christopher Alexander’s “Nature of Order” thinking might have some value here, in terms of humans’ innate appreciation of structures that have “strong centers” and other patterns.

  6. You appear to be working out your metaphor and its implication, which is superior I think to merely defending it. Since it is still taking shape, I’ll offer a few more comments.

    Intricate visual images may be more pleasing to the eye and therefore more desirable, but your post is about behavioral dynamics in virtual spaces. When confronted with too many options, many experience a sort of paralysis arising from too much choice. I also think Erich Fromm’s Escape from Freedom deserves some consideration. We can agree to disagree where people are most comfortable, stimulated, or effective. It doesn’t boil down to one category. The need for the wilderness vs. the unusability of the plaza sounds to me suitable for actually spaces in meat world but not so much virtual spaces on the Internet, which is the domain of your metaphor.

    Further, you are hedging in the argument that the openness and legibility of the plaza leads to too complex and/or too voluminous behaviors within its space, making it ultimately illegible and discomfiting. Similarly, the unpredictability of the warren construed as a safe haven with fewer choices leads to the assertion of a double paradox: the warren and the plaza are actually their opposites. So Chatroulette and Twitter are immediately understood, readily legible, but their secondary unpredictability and/or volume make them illegible. Can you really have it both ways just to assert the supposed superiority of the warren?

  7. Am pecking away, pondering on the edge of whatever is legible to me in this post:

    -Given the brain’s need for pattern forming and detection, and the fact that we cannot meaningfully discuss a concept that is not even an unformed blur in our mind, there cannot be a true warren that we can conceive of?

    -Aren’t the mini-paths we discover (or favorite places) a kind of personalized plaza existing within a larger warren (or maybe too large a plaza that is not yet apprehended by one)?

    -How come the word ‘fractal’ did not come up somewhere here? Somehow it seems related. Like too-legible warrens become sub-modules of a larger forming warren that might evolve or be forced into a new plaza that would then hopefully grow fuzzier to…