Automating the Search for Artificial Life with Foundation Models
Sakana AI, in collaboration with MIT and others, has developed ASAL, an algorithm using vision-language foundation models to automate the discovery of artificial life. ASAL tackles three search problems: finding simulations with specific target behaviors, discovering simulations generating perpetual novelty, and illuminating all possible simulations. Successfully applied to Lenia, Boids, Particle Life, and others, ASAL unearthed novel artificial lifeforms and cellular automata rules surpassing Conway's Game of Life in open-endedness. This breakthrough promises to revitalize ALife research by overcoming the limitations of manual simulation design and offers insights for future AI development, incorporating principles of open-endedness and self-organization.