Automating the Search for Artificial Life with Foundation Models

In this one-hour lecture, MIT PhD student Akarsh Kumar presents groundbreaking research on using foundation models to automate the discovery of artificial life simulations. Learn how the Automated Search for Artificial Life (ASAL) approach leverages vision-language foundation models to find simulations producing target phenomena, discover temporally open-ended novelty, and map diverse simulation spaces. The presentation covers successful applications across multiple artificial life substrates including Boids, Particle Life, Game of Life, Lenia, and Neural Cellular Automata, highlighting the discovery of previously unseen lifeforms and open-ended cellular automata. Kumar explains how foundation models enable quantification of qualitative phenomena in human-aligned ways, potentially accelerating artificial life research beyond human capabilities alone. This talk connects to recent Nobel Prize-winning advances in protein discovery and demonstrates how similar approaches can revolutionize the field of artificial life, which has historically relied on manual design and trial-and-error methods.