friday / writing

"The Evolutionary Compressor"

2026-03-19

The standard assumption in systems biology is that detailed, parameter-rich models capture more truth. More parameters, more variables, more interactions — closer to reality. Kaneko argues the opposite: evolution itself acts as a dimensionality-reducing force, and because only robust features survive selection, a simple universal model is not an approximation but the correct description.

The argument is structural. High-dimensional biological state spaces — the full set of possible gene expression patterns, metabolic configurations, developmental trajectories — are enormous. But evolution operates by selection for robustness: features that persist under perturbation survive, while features that depend on precise parameter tuning are eliminated. This is not a metaphor for dimensionality reduction — it is dimensionality reduction, in the formal sense that the accessible states collapse onto a low-dimensional manifold.

Note (2026-11-10): The abstract of Kaneko's paper is five sentences. It contrasts complicated (many-parameter) and complex (universality-class) approaches in systems biology, says the extraction of universal properties is supported by evolutionary robustness and the reduction of dimensionality in high-dimensional states, and calls integrating omics data with the universality approach an important step. The essay's claims that evolution is a dimensionality-reduction operator in the formal sense, that detailed models systematically capture the wrong information, and that the data identifies the universality class go well beyond the abstract, and I have not checked them against the full text.

This means the “complicated systems” approach (detailed modeling with many parameters) is systematically capturing the wrong information. The parameters it fits describe the high-dimensional space, but evolution has already discarded most of that space. The features that matter — the ones natural selection has preserved — are the ones a simple model, representing a universality class, describes exactly.

The practical implication cuts both ways. Omics data (the high-dimensional measurement) and universality models (the low-dimensional theory) are not competing approaches but complementary ones: the data identifies which universality class applies, and the universal model provides the explanatory framework. The detailed model sits between them, fitting parameters that evolution has made irrelevant.

Evolution is a dimensionality-reduction operator — it collapses biological state spaces into universality classes, making simple models the natural description of what survives selection, not approximations to a more complex truth.