Cell-level competency and evolutionary efficiency

Evolutionary algorithms served as search tools in these works. Lobo and Levin (2015) evolved a population of candidate regulatory networks in parallel until one matched the experimental results with zero error.¹ Kriegman, Cheney and Bongard (2018) compared a control called Evo, which kept a fixed morphology and control policy over a robot's lifetime, with a variant that included development.² They reported an incremental path that selection could climb, as legs grew and controllers were refined together.³ Kriegman and colleagues (2020) fed patterns from deployed organisms back into the algorithm as constraints.⁴

In the work of Shreesha and Levin (2023), each competent embryo was assigned a competency level. This integer set how many bubble-sort swaps the embryo could perform within its developmental cycle.⁵ They then placed competent embryos among hardwired ones and varied the starting proportion. At 2.5% of the population, competent embryos failed to dominate even at the highest level tested. At 10%, a level of 75 sufficed. At 20% the required level fell to 40, and at 30% a level of 10 was enough.⁶ Where competent embryos dominated, they did so in two or at most three generations.⁶

Hartl, Risi and Levin (2024) used neural cellular automata. They scaled competency continuously, from a direct encoding of the target pattern with no competency to a multi-scale competency architecture.⁷ Their result depended on noise. With little or no noise, the search was most efficient at low competency, and direct encoding looked preferable.⁸ With more noise, higher competency gave much greater efficiency. At noise levels of 0.375 and 0.5, the algorithm found no solution for direct encoding within 2000 generations.⁸

Pio-Lopez, Hartl and Levin (2025) developed an evolutionary framework with multi-scale competency to study aging. They stated that, as far as they knew, aging literature so far lacked computational models that considered cellular information processing and multi-scale competency.⁹ Their aim was an evolutionary model that included multi-scale competency in tissue.