Automating the Search for Artificial Life with Foundation Models
What this work does
Uses foundation-model embeddings to search simulations for semantic targets, historical novelty and diverse collections of behaviors.
Connection to discovery
An explicit example of representation-dependent novelty: changing the observer embedding changes what counts as a new behavior.
Evidence
Inspected the three search formulations and simulation/model choices such as CLIP, DINOv2 and Lenia.
Scope and limitations
Semantic or embedding novelty is not biological life, explanatory knowledge or unbounded complexity. Simulator parameterization remains a fixed discovery boundary.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.