Behavioral diversity, coverage and versatility metrics

Washburn (1908) discussed the starfish. Washburn reported a writer who held that its versatility tells against its perfecting any one movement through experience¹. Child (1924) took up the same animal, citing Jennings. In Child's account the starfish's versatility results from the physiological equivalence of the rays, any one of two or three of which may become dominant. He added that all radiate animals show some of this versatility and that the ameba is perhaps the most versatile². For higher animals, Child placed versatility in the combinations of fixed, specialized mechanisms, not in any one flexible part³.

Lotka (1925) treated versatility of aim, together with accuracy, as a main determinant of success among animals competing for food and safety. He drew from this the trend of evolution⁴. Calvo and Baluška (2015) used the behavioral repertoire of eukaryotes as evidence on minimal intelligence. They argued that the repertoire, and the organization that sustains it, resists interpretation in reactive, non-cognitive terms⁵. Blackiston and colleagues (2023) described xenobots as autonomous and as showing a diverse array of behavioral repertoires. They then asked whether such constructs are intelligent⁶.

Etcheverry, Moulin-Frier, Oudeyer and Levin (2025) separated two terms. Diversity is the ability of the exploration agent (IMGEP) to reveal a wide range of behaviors in a gene regulatory network. Versatility is the capability of the network itself to reach diverse goal states⁷. Versatility cannot be observed directly. The authors used the exploration agent's diversity, measured with the threshold coverage metric, as a proxy⁷. Random networks with four or fewer attractors stayed below a versatility of 0.026⁷. They then selected the most versatile networks for further study⁸.

Etcheverry and colleagues noted that their selection covered limited equation classes. They suggested that broader and more complex classes might reach higher behavioral diversity⁸. Gumuskaya and colleagues (2025) described a related gap for Anthrobots. The rules governing their motion are unknown. The authors proposed experiments in enriched environments with living and nonliving features to probe the behavioral repertoire and extend understanding of basal cognition⁹.