CONCEPT
The Generation-Selection Ratio
Holland's insight that the quality of a complex adaptive system's output degrades when variation outpaces evaluation—the condition that describes exactly what happens when a human accepts AI-generated output faster than they can judge it.
In every complex adaptive system, two processes operate in perpetual tension: generation, which produces variation, and selection, which evaluates it. The quality of the system's output depends on both, and specifically on the ratio between them. When selection keeps pace with generation, the system amplifies what works and extinguishes what does not. When generation overwhelms selection, the population fills with untested variation—candidates that have not been evaluated, building blocks that have not been tested, outputs that are plausible but not verified.
Holland documented this dynamic in genetic algorithms, where running the mutation rate too high relative to the fitness evaluation rate filled the population with noise, degrading the quality of convergence even as the quantity of candidates increased. The insight applies with structural precision to human-AI collaboration: the machine generates building-block recombinations at a speed that vastly exceeds any human's capacity to evaluate them, and the skew between generation and selection is the mechanism by which AI-assisted work produces smooth, plausible,