Reproduce “Guiding Evolution of Artificial Life Using Vision-Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Nikhil Baid, Hannah Erlebach, Paul Hellegouarch, Frederico Wieser
ASAL++ introduces a framework that uses Vision-Language Models to autonomously guide the open-ended evolution of artificial life, building sequences of increasingly complex Lenia forms. The system generates new evolutionary targets iteratively, demonstrating the potential for AI-driven scientific discovery in complex systems without continuous human input.
Key insights
- Foundation Models can effectively serve as "AI scientists" by autonomously generating hypotheses (new evolutionary targets) and driving exploratory processes based on observational data.
- Different target generation strategies (EST vs. ETT) present a trade-off between promoting visual novelty across generations and maintaining coherent, interpretable evolutionary sequences.
- The rich semantic understanding of VLMs enables translating abstract natural language concepts into concrete emergent behaviors in computational substrates like Lenia, advancing open-ended search.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.