August 22, 2011
It's a bug's life
Anyone who can be replaced by a machine deserves to be.
A noteworthy popular intellectual trend in recent years might be called “How Everything Works, In Spite of Itself.” Roughly, the trajectory can be described by James Gleick’s Chaos, which appeared in 1988; M. Mitchell Waldrop’s Complexity in 1992; and Steven Johnson’s Emergence, debuting in 2001. On the even more popular side, one can glance at Gladwell’s Tipping Point and Surowiecki’s Wisdom of Crowds, although more serious readers ought to be referred to Stuart Kauffman’s The Origins of Order. What unites these works – or rather, the trend that these books represent – is a perennial desire to see our world defined in terms of simple rules that, once intuited, reveal themselves as pervasive and universal. What are the consequences of this point of view, as we attempt to better understand societies and urbanism?
In a very real sense, this desire for heuristic happiness can be drawn straight back to the Enlightenment, if not even earlier. One can imagine Kepler experiencing equal parts delight and relief when his (only three, and very simple) laws of planetary motion persisted in their universality; or Newton’s, when he was able to derive these laws from the inverse square law of gravity. Whew! Kind of a shame to have to leave those Platonic solids behind, but there is something to be said for simplicity.
The principles derived by scientists working in the fields of chaos and complexity offer similar mercies. The desired outcome is more or less as follows: create a game of as few rules as possible, that in turn creates outcomes that are intricate, beautiful and pleasingly lifelike. Computer-assisted simulations such as Tim Conway’s Game of Life and Mitchell Resnick’s StarLogo have catalyzed the demonstration of how lifelike patterns evolve from simple rules. These simulations not only provide legitimate insights into real world processes, but also speak to us in a titillating fashion, inviting us to observe and name the resulting shapes generated by generations of cellular automata interacting with one another.
Resnick’s StarLogo, for example, quite accurately models simple organisms such as slime mold. Slime mold’s behaviour was long mysterious: first observed as a large, collective organism, it would seem to coalesce out of nowhere, but in fact it lived a good portion of its lifecycle as millions of unrelated single-celled organisms. In fact, slime mold unites into the larger, more easily observed sporangia form when food becomes scarce, thus beginning the process of seeding the next generation. Scientists had previously focused their research on identifying the “organizer” cells that would be responsible for guiding the formation of the sporangia but it eventually became indisputable that all cells were identical. The breakthrough was the realization that each cell emitted a signaling chemical known as acrasin, enticing any slime mold cells in the vicinity to join up with the emitting cell. StarLogo allows the armchair biologist to recreate this behaviour, and further play with it by altering the length of time the acrasin persists.
That these kinds of simple models provide compelling answers to the eerily efficient, wholly distributed emergent behaviors of slime molds, ants and other colonial organisms, tempts us to go much, much farther. We look at blobs wavering in silico on a screen and cannot but help imagine how not unlike they are to the forming and reforming blobs that we see on a rotting log, underneath a microscope, or at a rock concert. It is a step beyond anthropomorphism, since anthropomorphism is the personification of plants, animals or natural phenomena. We may in fact call this biomorphism, since we are ascribing living qualities to phenomena that most emphatically are not living (I only cook up this neologism with extreme reluctance, and with apologies to the short-lived art movement of the same name). Added to this is the conclusion that some biological attributes are scale-free* and we are invited to submit to the wondrously benign and generative powers of nature; this is a powerful recipe for pop-intellectual seduction. Well, perhaps not – since you can usually count on the game theorists to rain on our parade.
No one who has played Prisoner’s Dilemma in its myriad forms would accuse game theorists of being optimistic about the human condition. Accordingly, über-game theorist Thomas Schelling’s 1969 paper Models of Segregation outlines a game scenario where a randomly generated 2-dimensional matrix of two types of cells rearranges itself, much like Conway’s and Resnick’s games, using simple rules. That is, each cell acts on its preference for the identity of each of its immediate neighbors. If too many cells are different from the cell in question, it moves to a different position on the matrix to satisfy this preference. This calculation is repeated for every cell in the matrix, which concludes one generation. We can immediately guess that, given a high enough requirement for likeness, a noticeably segregated environment will be created after a sufficient but finite number of rounds. The surprising bit is the fact that, even if the preference is mild, the results are decisive (see Presh Talwalkar’s results of running a simulation of 2500 rounds, where the preference rule states that 30% of bordering neighbors ought to be of the same variety as the cell in question). Lest we jump to the conclusion that all segregation is bad, Schelling’s game provides the foundation to understand why, according to Paul Krugman, similar businesses congregate close to one another; it may also be somewhat more creatively used as an apologia for why everyone’s a little bit racist.
However, that is not all there is to it. Chris Snijders followed up on 40 years of research following Schelling’s initial work and summarized his findings at the recent conference on Game Theory and Society’ (15m40s ff). In discussing Schelling’s intellectual inheritors, it is interesting to see what has been omitted as well as what has been elaborated. Significantly, while segregation occurs even when neighbors express mild preferences, the assumption is that all preferences are acted on locally. When preferences are globally stated, however, the tendency to segregate over mild preferences disappears (for example, by applying preferences to the total number of clusters, or specifying the maximum diameter of connection between neighbors). Quite simply, it is the size of the board that matters, since a small board will render as indistinguishable the differences between acting globally and acting locally. One wonders what the impact of this would be on modeling a city of millions, versus a village of hundreds or even a town of thousands.
Furthermore, Snijders finds that most follow-up work to Schelling explores the end state and not the process. That is, a large proportion of the papers are preoccupied with demonstrating their pedigree by ascertaining whether their methods’ findings converge with Schelling’s or not. In turn, the concern with outcome and not process sheds increasingly less light on actual (that is, ecological or sociological) reality, and results in a sort of intellectual bubble. In fact, Snijders’s own work in applying Schelling’s game theory to the highly competitive Dutch construction industry does not unearth significant correlations to the emergent behaviours that Schelling identified in such a succinct way. While this should not detract from Schelling’s findings in the abstract, defining success as the similarity to which outcomes hew to Schelling’s original work leads us further and further away from the reality that we might be trying to describe, let alone act on.
Even without performing a review of the literature, though, we can raise a number of concerns with pursuing Schelling’s line of thinking to its reductio ad absurdum. While it is wondrous to see “lifelike” behaviour in these models, do we ourselves really adjust our preferences so rigorously, or even so locally? When are we ever in a position where we are surrounded by always exactly 8 neighbors? (New Yorkers who ride the subway might be the only people who experience this on a regular basis, but the crowding makes acting on one’s preferences inadvisable, if not impossible.) One also ought to consider to what extent we operate under the same simple set of rules that run the simulation and give it its shape and elegance. Preferences in a global situation may in fact be much more deterministic. We may very well decide where to live based on our proximity to work or family, and not whether we enjoy our immediate neighbors’ company. Furthermore, as telecommunication technologies continue to mediate our experience of work and play, I suspect we will see an ever-increasing de-emphasis on place.
These are obviously very literal interpretations of Schelling. More importantly, we ought to be quite cautious in considering the unalloyed goodness of self-organizing behaviours. In The Self-Organizing Economy, Paul Krugman writes:
“An economy with a strong business cycle exhibits more temporal self-organization than an economy that grows smoothly, but most of us would rather live in the latter. A city whose racially integrated communities unravel, producing huge segregated domains, becomes more spatially organized, but not better, in the process. Self-organization is something we observe and try to understand, not necessarily something we want.” (Krugman, pp5-6)
On the one hand, a recognition of emergence provides us with the reminder, never too often restated, that we are and continue to be part of nature and never can be separate from it. On the other hand, we are also asked to make the kinds of choices that set us apart from nature, since it is also our task to design culture.
Thus the more practical question asks if we really can use Schelling’s and other emergent models to help predict or design for better outcomes. For example, you could grow a city using the rules to Conway’s Game of Life, but why would you want to? What would it teach you? Would it help you to better design the urban experience for the people who are actually supposed to live there, or is it the intellectual equivalent of watching a Chia Pet grow?
As a more concrete example, what do we as designers, commentators or planners really need to know about emergence in order to design a better park, or can we use the decidedly more low-brow technique of desire paths, first articulated by Gaston Bachelard in The Poetics of Space in 1958? Hopefully Bachelard, in his capacity as a poet as well as a philosopher and critic, recognized that the desire path is not necessarily one of greatest efficiency. Ants create the most efficient pathways, since the emergent behaviour they possess does not give them any other choice. We may marvel at this, and learn through them how to create better pathways, but we are also able to have those design choices serve purposes that go beyond mere utility. Designers also have a responsibility to recognize that, remarkably, the world is not a 2x2 matrix, and that issues such as topography, solar exposure and shelter from rain (or snow) play a serious role in the successful experience of space. Newton’s classical mechanics tells us only what rockets we should not build (ie, those which will not work), but does not tell us what kinds of rockets we ought to build; similarly, relying on a sexy computer program to generate optimal solutions using virtual ants threatens us with the possibility that our designers will expect us to act as exactly that.
Citing the slime mold once again, Nathan Myhrvold notes the boundaries of the analogy, and what may lie beyond it. While emergence gives rise to truly amazing structures and feats of survival, what it does not teach us is that “creating an organ requires a deep subjugation of interests. The heart has to trust that the brain will run things and that the reproductive system will look after its long-term interests.” It is these forms of specialization that constitute the possibility of rapid adaptation. So when our cities are confronted with future systemic disruptions that are imminently obvious to any one individual, the emergence model only leaves us stuck with waiting for the stimulus, then picking up and moving to another square on the board. To a significant degree, biological systems based on emergence require stability: recall how poorly bee colonies have adapted to disruptive conditions. Myhrvold writes:
“Not all problems can be fixed after they have become severe. If we wait until all the cheap oil is gone before developing alternatives, the resulting economic disruption will be traumatic. And if we wait until global warming is intolerable before making serious cutbacks in emissions of long-lived greenhouse gases, the physics of the atmosphere virtually guarantees that all our efforts will then be too late.”
It should be noted that acrasin, the signaling chemical used by slime molds to attract each other, “was descriptively named after Acrasia from Edmund Spenser's Faerie Queene, who seduced men against their will and then transformed them into beasts. Acrasia is itself a play on the Greek akrasia that describes loss of free will.” We can argue about which species is more successful than others based on various criteria, but if we are to continue being successful – or being around – we ought to begin acting our biological age.
* This fascination with scale-free phenomena and how they correlate across not just living organisms but also with cities-as-organisms and across cities, regardless of size, wealth or geography, is another remarkably shiny marble and deserving of its own post. However, for the moment I refer the interested reader to Geoffrey West’s recent TED Talk on the subject. Without critiquing the work itself I would still want to ask the question of what practical advice does this give us. Indeed, if we are to go by West’s conclusions, our prospects are dire indeed.
Posted by Misha Lepetic at 12:35 AM | Permalink