Out of randomness: How evolution benefits from modularity

Abstract Brute force random search, effective in exploring solution spaces, often becomes inefficient or infeasible in real-world scenarios with vast solution spaces. A more effective method, akin to natural evolution, involves recombining existing modules into new ones, a concept known as “evolution as tinkering” introduced by François Jacob. Understanding these mechanisms is crucial for comprehending evolution and designing evolution-inspired algorithms. This study employs genetic algorithms (GAs) to quantitatively explore how evolution-like processes, marked by mutation and crossover, search for complex solutions. Compared to random search, GAs significantly improve the probability of finding solutions, especially complex ones. This improvement varies, showing biases towards more intricate solutions, likely due to the crossover process in GAs that facilitates the recombination of smaller modules into larger, more complex ones. Our experiments reveal that grouping module components rather than scattering them aids in forming larger, more complex solutions. This mirrors a pattern observed in real biological systems, where the sequences encoding individual genes are clustered together in all prokaryotic organisms. These findings highlight the importance of spatial correlations in the development of larger, more intricate modules and solutions, underscoring how modularity and modular recombination enhance solution space exploration.