MSPD Metric Directs Open-Ended Evolution with Physics-Based Formula
A new renormalization-group-inspired metric outperforms black-box complexity measures in artificial life.
Deep Dive
MSPD (Multi-Scale Path Divergence) is introduced as a gradient-free fitness function for open-ended evolution in artificial life. Unlike black-box neural-network metrics, MSPD is an explicit formula inspired by renormalization group theory, quantifying temporal multiscale organization. It directs evolution more effectively than random parameters and bridges artificial life to physical theories of complexity, working across Flow-Lenia, Life-like CA, and Particle Life++.
Key Points
- MSPD is a gradient-free fitness function computed from population trajectory using a renormalization-group-inspired estimator.
- High MSPD correlates with scale-dependent frustration, linking directly to Vanchurin et al.'s physical theory of biological complexity.
- The metric generalizes across multiple artificial life substrates: Flow-Lenia, Life-like cellular automata, and Particle Life++.
Why It Matters
Provides a principled, interpretable metric for directing artificial life evolution, connecting simulations to fundamental physics of complexity.