Brian Potter argues that biological evolution and an economist's circuit simulation both depend on the same shortcut: mixing and reusing working modules to cut down a search space too vast to enumerate.
Sex is an information-acquisition hack, and so is the engineer's habit of stacking tested modules into bigger ones. Brian Potter, writing on Construction Physics, makes the case that biological evolution and an economist's circuit simulation both depend on this same shortcut to find useful complexity in a search space too vast to enumerate.
In a follow-up to his earlier piece on Arthur, Potter lays out the comparative argument. The economist W. Brian Arthur modeled technological evolution by letting randomly combined logic gates compete in a population, with the fittest variants surviving into the next round. The result, as Potter has previously described, is that the simulated circuits can grow into objects as elaborate as a 12-way AND gate or a 4-bit adder starting from nothing but NAND gates, random combination, and selection. They do so not by enumerating the vast space of all possible circuits, but by taming it.
The mechanism is modularity plus recombination. Once the simulation has produced a working subcomponent, it stops re-rolling the dice on that subcomponent and treats it as a building block. A few working modules combine into one larger module, the larger module gets tested, and the unpromising branches of the search tree get cut off. Each round of combination plus selection reaches deeper into the design space while keeping the combinatorial explosion in check. The system does not need to know what it is looking for. It just needs a way to keep good assemblies around and to shuffle them.
Biology, in Potter's framing, runs the same program. Sex, in the genetic sense of mixing two parental genomes into an offspring, lets a population test new combinations of alleles without losing the working ones parents already carried. Horizontal gene transfer does a related job for microbes, letting one lineage borrow a tool another already evolved. Gene duplication followed by divergence lets a cell keep an original function intact while experimenting on a copy. In each case, the operation is the same: keep working modules, recombine them, and screen off the combinations that do not survive.
The constructive claim that follows is sharp enough to test. A system's rate of information acquisition, its ability to discover new useful structure, is bounded by how well it can mix and reuse modules it has already built. Asexual populations can still evolve, but they pay a combinatorial cost: every useful mutation has to arise on its own and propagate through a clonal lineage. Sexual populations pay an up-front cost in finding mates and breaking up co-adapted combinations, but they get to test many variants of a working genome at once. The framing predicts faster adaptation in changing environments for populations with stronger recombination machinery, and the empirical record on recombination rates broadly supports that.
The framing has limits, and the analogy is structural rather than literal. Evolution has no selector making design choices, and Arthur's randomness is a modeling stand-in, not a description of how nature works. Calling sex an information-acquisition hack is a way of naming the trick, not of putting intention into the genome. The two systems face a search problem too large to solve by enumeration, and both arrive at the same answer by modular recombination. That is the claim, and the rest is detail.
The open question is whether the framing travels. If modular recombination really is the bottleneck on information acquisition, then the speed of adaptation in any system, biological, technological, or organizational, should track the quality of its recombination machinery. That is a testable prediction, and one that biologists, software maintainers, and industrial historians can each run on their own data.