The recent Nobel Prize for groundbreaking advancements in protein discovery underscores the transformative potential of foundation models (FMs) in exploring vast combinatorial spaces. These models are poised to revolutionize numerous scientific disciplines, yet the field of Artificial Life (ALife) has been slow to adopt them. This gap presents a unique opportunity to overcome the traditional reliance on manual design and trial-and-error methods for uncovering lifelike simulation configurations.
In a new paper Automating the Search for Artificial Life with Foundation Models, a research team from MIT, Sakana AI, OpenAI, The Swiss AI Lab IDSIA and Independent introduces Automated Search for Artificial Life (ASAL). This novel framework leverages vision-language FMs to automate and enhance the discovery process in ALife research.
ASAL demonstrates its potential across various ALife substrates, including Boids, Particle Life, Game of Life, Lenia, and Neural Cellular Automata. By employing ASAL, the researchers uncovered previously unknown lifeforms and extended the frontier of emergent structures in ALife simulations. Beyond discovery, ASAL’s framework facilitates quantitative analysis of traditionally qualitative phenomena, mirroring human-like methods to measure complexity. Crucially, ASAL’s FM-agnostic design ensures compatibility with future foundation models and ALife substrates.
ASAL employs vision-language FMs to evaluate simulation outputs, formulating the process as three distinct search problems:
- Supervised Target Search: Aligns simulation trajectories with specified text prompts, enabling targeted discoveries.
- Open-Ended Exploration: Identifies simulations exhibiting high historical novelty at each timestep, fostering innovation.
- Illumination: Seeks diverse simulations by maximizing the distance between neighboring configurations
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.
Empirical results demonstrate ASAL’s effectiveness. The framework uncovered previously unseen lifeforms in Lenia and Boids simulations and discovered cellular automata exhibiting open-ended behaviors akin to Conway’s Game of Life. Furthermore, by integrating FMs, ASAL quantifies phenomena that were once purely qualitative, aligning these measurements with human perceptions of complexity.
ASAL’s FM-based paradigm represents a significant leap forward for ALife research. By automating the discovery process, it enables researchers to explore the vast and intricate space of artificial life forms more effectively than ever before. This approach marks a departure from traditional methods and provides a scalable, innovative framework for future studies.
To the best of the researchers’ knowledge, this is the first instance of leveraging foundation models to drive ALife simulation discovery. ASAL sets the stage for a new era of exploration, promising to accelerate advancements beyond the limits of human ingenuity alone.
The code is available on project’s GitHub. The paper Automating the Search for Artificial Life with Foundation Models is on arXiv.
Author: Hecate He | Editor: Chain Zhang
The post Automating Artificial Life Discovery: The Power of Foundation Models first appeared on Synced.
Leave a Reply