Neuroevolution: Harnessing Creativity in AI Agent Design (neuroevolutionbook.com)

Sebastian Risi, Yujin Tang, David Ha, and Risto Miikkulainen have released a comprehensive new textbook and companion website, Neuroevolution: Harnessing Creativity in AI Agent Design (MIT Press, 2025). The book synthesizes decades of work in evolutionary algorithms, indirect encodings, novelty/diversity search, coevolution, neural cellular automata, evolutionary neural architecture search (NEAT, Backprop-NEAT), and evolutionary meta-learning, and pairs theory with extensive practical material: notebooks, demos (bipedal walkers, Atari, Mario, soft robots, NCAs), assignments, conference tutorials, and an instructor portal. The authors — senior researchers with roots at Google Brain/DeepMind, Sakana AI, and major universities — also link to community resources (EvoJAX, community GitHub) and public course recordings that make the methods reproducible and scalable. For the AI/ML community the book is significant because it reframes neuroevolution as a modern, complementary toolkit to gradient-based deep learning and reinforcement learning. It provides concrete pathways to scale evolutionary methods (CMA-ES, ES, MAP-Elites), use indirect encodings (CPPNs, hypernetworks, self‑attention as dynamic encodings), and combine evolution with RL and LLMs (evolutionary fine-tuning, model merging, procedural content generation). Practically, readers gain actionable techniques for robust control, architecture search, open‑ended behavior discovery, and creative generative systems — making neuroevolution a viable option for problems where diversity, modularity, or exploratory creativity matter.