Degenerate Design

A study published in Nature yesterday found that Beijing’s trees produce only about 10% of the city’s emissions of volatile organic compounds, but account for around 52% of the chemical activity that turns those compounds into ground-level ozone.

»The point of design may be to make being wrong survivable«

The city’s willows and poplars were planted decades ago because they grow quickly and tolerate the local climate. About 35% of Beijing’s trees release large amounts of isoprene, a gas that can contribute to ozone formation. And the hotter it gets, the more active this chemistry becomes: at 35°C, the researchers found, the vegetation’s contribution was about seven times greater than at 20°C.¹

Planting those trees wasn’t a mistake. They were a sensible response to the information available at the time. The problem was that a property already known to science had not yet become consequential at the scale of urban planning.

Some things have no name until they matter. We can only measure what we know exists, and only investigate what we have reason to suspect matters. There is always a shadow beyond that: things we don’t know we don’t know.

The moment a design enters the world, its conditions begin to change. Its environment and relationships shift, and properties that once seemed irrelevant can acquire unforeseen consequences. We normally respond by gathering more information: measure more, monitor more, build better models. But if a property is genuinely unknown, we don’t know to measure it.

Perhaps the answer isn’t more measurement, but more multiplicity.

Biology provides a remarkable example. Evolution has no model of the future. It cannot anticipate a new environment, yet organisms can survive radical changes because they contain multiple, overlapping ways of functioning.

Neuroscientists Gerald Edelman and Joseph Gally called one of these properties »degeneracy«: structurally different components can perform similar functions, while the same component can contribute to different functions under different conditions. Unlike redundancy, degeneracy gives a system multiple ways to remain functional as circumstances change.² Evolution doesn’t know which possibilities will matter. It preserves variation, and the environment reveals what becomes useful.

AI researchers Kenneth Stanley and Joel Lehman make a related argument about search. When we optimise towards a predefined goal, we privilege what we already know how to value. But the most important discovery may be something whose value was invisible when the search began. An objective function can become a constraint on discovery itself. Multiplicity protects us in two directions: against failures we didn’t anticipate, and against possibilities we didn’t recognise.³

The implication for design is simple. We tend to think good design removes unnecessary things. Perhaps good long-term design sometimes creates things whose only apparent purpose is to preserve options for the future.

An ecology does this naturally. It contains multiplicity, variation, and relationships whose significance can change as conditions change. A city can be understood in the same way: as an ecology of overlapping capabilities.

The Beijing trees were not dangerous because someone failed to measure them. They became dangerous because a property that had once seemed irrelevant acquired consequences at a different scale and under different conditions. And because the city had replicated the same solution so extensively, that hidden property became a city-wide one.

The point of design may be to make being wrong survivable. We can’t measure what we don’t know exists. We can’t optimise for properties whose consequences haven’t yet appeared. Perhaps that’s what degenerate design means: not a more complete model, but a system built with enough different ways of functioning to survive being wrong.

References