Swarm Intelligence: A Reading Note
Chapter 5: Genes and Memes
The evolution of populations and the transformation of societies share many similarities. Genetic inheritance occurs across generations through selection, crossover, and mutation, ultimately leading to offspring. The term "meme," originating from the Greek word "mimeme" meaning "imitation," can be considered the evolutionary gene of culture. When a person generates an idea, and it is transmitted to others through social interaction and learning, the idea is retained if accepted by others. However, these ideas are not static; others may integrate their own thoughts, restructure the idea with their cognition, or generate new ideas before passing them on.
While the evolutionary concepts of genes and memes are similar, there are differences. "Survival of the fittest" is a constant in nature, but an idea prevalent in society is not easily extinguished. There are several reasons: First, in genetic evolution, selection is a primary mechanism of evolution, with variation already existing in genes. Populations develop adaptive abilities through selection, but minds usually adapt by changing. Humans are more likely to change ideas than discard them. As cognitive dissonance theory suggests, we tend to resolve cognitive dissonance by changing cognition, leading to a faster rate of variation and derivation of ideas than the evolution of population generations. Secondly, the evolution of memes is regional and requires social interaction for idea transmission, with social influence depending on the relationship and importance of two people. Therefore, the evolution of memes takes time to spread and is unlikely to impact an entire culture abruptly, unlike genes that can rapidly adapt to drastic environmental changes.
However, with the widespread use of the internet, memes have transcended geographical barriers and spread rapidly online, giving rise to a distinct "meme culture." Online meme evolution resembles genetic evolution, typically having a fixed keyword or structure, such as a specific style or image accompanied by text relevant to a situation. Internet users create new memes by combining current affairs or other memes. Interesting memes endure, while uninteresting ones quickly disappear after a brief popularity.
John Holland proposed the Genetic Algorithm in the 1960s, a search algorithm for solving optimization problems. Problems are encoded as chromosomes, with each gene representing a parameter, and a chromosome representing a possible solution. Adding a fitness function evaluates the quality of the chromosomal solutions. The evolutionary process—selection, crossover, mutation, and offspring generation—ensues, ultimately converging the chromosomes to an optimal solution. Holland's initial goal with this algorithm was to study adaptive systems (Holland, 1975, 1992), though its current application in engineering mainly focuses on optimization.
In the 1990s, Burke and Smith proposed the Memetic Algorithm, similar to the Genetic Algorithm but incorporating local search techniques inspired by the regional spread of memes, addressing the issue of the Genetic Algorithm converging too quickly to suboptimal solutions (Burke & Smith, 2000).
Simultaneously, Cowan and Reynolds proposed the Cultural Algorithm. A key difference between gene and meme evolution is that the mind itself is a system with memory and beliefs, processing external stimuli and internal cognition. Like the Genetic Algorithm, the Cultural Algorithm generates a population of potential solutions that evolve over time. However, it incorporates mechanisms of memory and belief into the algorithm, divided into two layers: a population space for regular evolutionary processes and a belief space representing the collective memory of the population, signifying ways to change and adjust individuals. The belief space mutates individuals in the population space, which in turn selects individuals and updates the belief space's adjustment methods. Both spaces evolve independently, creating a reciprocal interaction (R. Reynolds, 1994; Cowan & Reynolds, 1999).
These algorithms all draw on the characteristics of population evolution for optimization. Compared to contemporary AI, Genetic Algorithms require extensive computation of fitness functions and involve uncertainty in the evolutionary process. My experiments have shown they don't always yield optimal solutions like other algorithms. Yet, Genetic Algorithms can be surprisingly effective for certain challenging problems, especially in fluid dynamics applications, such as airplane wings or train fronts. Airplanes are a modern marvel of technology, with current fluid dynamics theories still insufficient to fully explain the phenomenon of flight. Although my advisor believes it to be an optimal method for solving difficult problems, I think, given current technological trends, wearable devices and smart homes require real-time, distributed systems. The time-consuming and computationally intensive nature of Genetic Algorithms makes them less applicable. However, I see potential in their use as adaptive systems. Robots interacting with the real world may encounter complex situations, and applying the characteristics of population evolution could enable rapid adaptation to highly dynamic environments.
Holland, J. H. (1975). Adaptation in natural and artificial systems: An introductory analysis with applications to biology, control, and artificial intelligence. Oxford, England: U Michigan Press.
Holland, J. H. (1992). Genetic Algorithms. Scientific American, 267(1), 66-73.
Burke, E., & Smith, A. (2000). Hybrid evolutionary techniques for the maintenance scheduling problem. IEEE Transactions on Power Systems, 15(1), 122-128.
Reynolds, C. W. (1987). Flocks, herds and schools: A distributed behavioral model. In Proceedings of the 14th annual conference on Computer graphics and interactive techniques, 25-34.
Reynolds, R. G. (1994). An introduction to cultural algorithms. In Proceedings of the 3rd annual conference on evolutionary programming, World Scientific Publishing, 131-139.