The Roughness

Three findings from different fields, all making the same uncomfortable claim.

The first is about heat. A team at UMass Amherst, working with collaborators from MIT, Stanford, and Oak Ridge National Laboratory, found that polymer composites made with defective fillers — graphite oxide riddled with structural flaws — conduct heat 160% better than composites made with perfect fillers. The pristine graphite has intrinsic thermal conductivity of 293 W/mK. The defective version: 66 W/mK. The worse material produces the better composite.

The mechanism is the surface. Defective fillers have rough, uneven surfaces. That roughness prevents polymer chains from packing tightly at the interface — the boundary where filler meets matrix. The loose arrangement improves vibrational coupling, allowing heat to cross the boundary more efficiently. For decades, the field pursued ever-more-perfect fillers. The perfection was the problem. It created smooth surfaces that polymer chains could pack against too neatly, forming a tight boundary layer that blocked exactly the thermal transport the filler was supposed to enable.

The interface is where the work happens. And the interface needs roughness to work.

The second is about language. An international team led by Michael Pleyer published a reassessment in Trends in Cognitive Sciences of Hockett’s design features — a 65-year-old framework that defined human language through properties like arbitrariness, duality of patterning, and displacement. The framework has been the foundation of linguistics courses for generations. It’s elegant, clean, and increasingly wrong.

Modern research shows language is saturated with iconicity — form resembling meaning. Not just onomatopoeia like “buzz” and “crash,” but sound-symbolic patterns across entire vocabularies, stretched pronunciation that physically embodies duration (“slooooow”), gesture and facial expression integrated into the grammar itself, and emoji carrying genuine communicative function in written text. Features once thought uniquely human — productivity, displacement, recursive structure — appear to varying degrees in animal communication. The clean boundary between human language and everything else dissolves under empirical pressure.

Hockett’s framework treated these messy features as noise — departures from the “real” system. The revision treats them as the system itself. Language isn’t a clean digital code that got contaminated with analog messiness. It’s a multimodal, socially embedded, dynamically evolving system where the messiness enables learning, communication, and evolution. The gesture that accompanies speech isn’t decoration. It’s grammar that your hands are doing.

The third is about vaccines. Mathematical epidemiology has long modeled vaccine coverage as a population-level average: if 95% of people are vaccinated, herd immunity holds. But modeling work on heterogeneous vaccine coverage reveals something uncomfortable. When vaccine protection is distributed unevenly across a population — some people well-protected, others less so — the epidemic dynamics change dramatically. One model showed that highly heterogeneous vaccine protection produced a 35 percentage-point reduction in outbreak size compared to homogeneous protection at the same mean level.

Uniform coverage, the thing public health campaigns aim for, can be worse than uneven coverage at containing an outbreak. The mechanism: heterogeneity creates patches of high resistance that function as natural firebreaks. Disease burns through low-coverage pockets but can’t bridge the high-coverage zones between them. Uniform moderate coverage provides consistent but penetrable resistance everywhere — the disease can find paths through the entire population. The rough landscape of uneven immunity contains the fire better than the smooth landscape of uniform partial protection.

The bridge. In all three cases, roughness — the uneven, the imperfect, the messy — outperforms smoothness.

The smooth filler creates a neat boundary that blocks heat. The rough filler creates a disordered boundary that conducts it. The clean linguistic framework categorizes neatly but misses how language actually works. The messy multimodal reality — gesture, iconicity, social embedding — is the system, not noise on top of it. The uniform vaccine coverage creates a penetrable surface. The patchy, uneven coverage creates a landscape that contains spread.

The instinct in each field was the same: smooth things out. Polish the filler. Purify the framework. Equalize the coverage. And in each case, the smoothing removed the mechanism.

This is worth sitting with because the engineering instinct — in materials, in linguistics, in public health — is to simplify. Reduce variance. Remove irregularity. Optimize toward uniformity. These aren’t bad instincts. But they share a blind spot: they assume the variance is waste. That the irregularity is a problem to be solved rather than a feature to be understood.

Roughness isn’t always good. A rough approximation is still worse than an exact answer. A rough surface on a bearing destroys efficiency. The claim isn’t that imperfection is universally superior. The claim is narrower and more interesting: at interfaces — the boundary between two different things — roughness often does work that smoothness can’t. Heat crosses a polymer-filler boundary through vibrational coupling that needs loose packing. Language crosses the boundary between minds through multimodal signals that a clean code can’t carry. Disease crosses the boundary between populations through paths that uniform resistance can’t block but patchy resistance can.

The roughness is the grip. Smooth it out, and things stop catching.