Automating the Search for Artificial Life with Foundation Models

TL;DR

This work presents Automated Search for Artificial Life (ASAL), a framework that leverages vision-language foundation models to automatically discover and analyze life-like simulations across diverse substrates. By formalizing three search modes—Supervised Target, Open-Endedness, and Illumination—ASAL can find target phenomena, cultivate temporally open-ended novelty, and illuminate a broad landscape of interesting simulations. The approach yields previously unseen lifeforms in Lenia and Boids and enables quantitative, human-aligned assessments of emergence and diversity. Its substrate- and model-agnostic design promises to accelerate ALife research by expanding the scope of exploration beyond manual design and trial-and-error. The use of FM embeddings as a common representation enables both discovery and post hoc analysis, offering a practical pathway to map, quantify, and understand life-like complexity in computational universes.

Abstract

With the recent Nobel Prize awarded for radical advances in protein discovery, foundation models (FMs) for exploring large combinatorial spaces promise to revolutionize many scientific fields. Artificial Life (ALife) has not yet integrated FMs, thus presenting a major opportunity for the field to alleviate the historical burden of relying chiefly on manual design and trial-and-error to discover the configurations of lifelike simulations. This paper presents, for the first time, a successful realization of this opportunity using vision-language FMs. The proposed approach, called Automated Search for Artificial Life (ASAL), (1) finds simulations that produce target phenomena, (2) discovers simulations that generate temporally open-ended novelty, and (3) illuminates an entire space of interestingly diverse simulations. Because of the generality of FMs, ASAL works effectively across a diverse range of ALife substrates including Boids, Particle Life, Game of Life, Lenia, and Neural Cellular Automata. A major result highlighting the potential of this technique is the discovery of previously unseen Lenia and Boids lifeforms, as well as cellular automata that are open-ended like Conway's Game of Life. Additionally, the use of FMs allows for the quantification of previously qualitative phenomena in a human-aligned way. This new paradigm promises to accelerate ALife research beyond what is possible through human ingenuity alone.

Figures (19)

• Figure 1: Overview: Our method, ASAL, searches for interesting ALife simulations by using a vision-language foundation model to evaluate the simulation's produced videos. ALife lifeforms are discovered across different substrates with three different mechanisms: (1) found via a text prompt, (2) found via searching for open-ended simulations, and (3) illuminating a set of diverse simulations.
• Figure 2: ASAL: Our proposed framework, ASAL, uses vision-language foundation models to discover ALife simulations by formulating the processes as three search problems. Supervised Target: To find target simulations, ASAL searches for a simulation which produces a trajectory in the foundation model space that aligns with a given sequence of prompts. Open-Endedness: To find open-ended simulations, ASAL searches for a simulation which produces a trajectory that has high historical novelty during each timestep. Illumination: To illuminate the set of simulations, ASAL searches for a set of diverse simulations which are far from their nearest neighbor.
• Figure 3: Discovered target simulations: Using Equation \ref{['eq:supervised']}, ASAL discovered simulations that result in a final state which matches the specified prompt. Results are shown for three different substrates. More prompts can be seen in appendix Figures \ref{['fig:supervised_extra_lenia']}, \ref{['fig:supervised_extra_boids']}, and \ref{['fig:supervised_extra_plife']}.
• Figure 4: Discovered temporal target simulations: Using Equation \ref{['eq:supervised']}, ASAL discovered simulations that produce a sequence of events which match a list of prompts. The second row shows how the first simulation generalizes to a different initial state. The results are shown for the NCA substrate.
• Figure 5: Discovered open-ended simulations: Using Equation \ref{['eq:oe']}, ASAL discovered open-ended simulations in the Life-Like CAs substrate. Simulations are labeled in Golly notation eppstein2010growth to denote the number of living neighbors required for birth and survival. (a) The discovered CAs rendered over a simulation rollout. (b) The temporal trajectories of three simulations in CLIP space. The pixel-space simulation (red) exhibits a convergent trajectory, whereas the FM-space simulation (green) demonstrates a more divergent trajectory, even exceeding that of Conway's Game of Life (blue). (c) All Life-like CAs plotted based on the UMAP mcinnes2018umap projection of the CLIP embedding of their final state, colored by open-endedness score. The resulting structure reveals distinct islands of similar simulations, with the most open-ended CAs grouped together near the bottom.