Deep Swarm and Evolution for Generative Artificial Intelligence

Generative AI or LLM (Large Language Model) is currently flourishing, but its core technology is mainly based upon correlation inference from big data, making it difficult to derive deep knowledge. This book provides theoretical and practical knowledge about swarm and evolutionary approach, i.e., deep swarm and deep evolution, to generative AI. While the central theme of the book is generative AI, it also develops a discussion of AI in a broader sense. The development of such tools contributes for better optimizing methodologies with the integration of several machine learning and deep learning techniques. In particular, we will discuss how the âemergenceâ concept can contribute to the improvement of AI.

Another goal of this book is to model human cognitive function in terms of âemergenceâ and to explain the feasibility of AI. In other words, to understand how intelligence emerges, to map it to the real world, and to provide causal explanations by means of evolutional and psychological mechanisms. To this end, this book focuses on human perceptions of âutility.â We describe the emergence of various cognitive errors, and irrational behaviours in the above-mentioned multi-objective situations. We also discuss the cognitive differences and similarities between humans and LLMs by asking the same psychological questions to LLMs as to humans and observing how LLMs answer the questions. Such studies are important when applying LLMs to real-world tasks that involve human cognition, e.g., financial engineering and market issues.

Furthermore, this book illustratively describes the intelligent behaviour of living organisms. For instance, we explain how ants choose a preferred nest while solving a kind of optimal problem (i.e., Buffon's needle problem) and how ants sacrifice themselves to build a bridge. This is to clarify how to achieve AI in the direction of artificial life. We also describe sexual selection, which is a well-known natural phenomenon that troubled Darwin, i.e., why evolutionarily useless items evolved such as peacock feathers and moose antlers etc. Sexual selection is extended as ânovelty searchâ for the application of generative AI.

Yet another emphasis is its real-world applicability. We provide empirical examples from real-world data to show that the concept of deep swarm and evolution is successfully applied when addressing tasks from such recent fields as robotics, e-commerce WebShop and image generation etc. In addition, a GA (Genetic Algorithms) simulator is available in the appendix to test and experiment with evolutionary multi-objective optimization for the readerâs studies.