Abstract
Distinguishing genuine Darwinian evolution from random genetic drift is a fundamental challenge in artificial life (ALife) research. Traditional metrics, such as Bedau-Packard statistics, rely on genetic persistence as a proxy for adaptation. However, these metrics are susceptible to false positives in systems with non-heritable high-persistence traits. This paper introduces Fitness Transmission, a robust statistical metric that measures the correlation between the reproductive success of parents and their offspring. By using the number of grandchildren (NOGC) as a universal fitness proxy, this method reliably identifies true adaptive evolution and filters out stochastic noise.
Introduction
In complex artificial life simulations, identifying the emergence of adaptive evolution is often non-trivial. While populations may exhibit dynamic changes, these can often be attributed to random genetic drift or chaotic system dynamics rather than natural selection. For decades, the Bedau-Packard (BP) statistics have served as the standard for measuring evolutionary activity. BP metrics track the "persistence" of components (e.g., genes or strategies) over time.
However, a critical flaw exists: BP metrics essentially measure the variance in persistence. If a simulation randomly assigns long life or high reproductive potential to certain individuals without a heritable basis, BP statistics will incorrectly signal adaptive evolution. This paper addresses this limitation by proposing a metric rooted in the fundamental requirement of Darwinian evolution: the heritability of fitness.
The Problem with Persistence
The core issue with existing metrics is that high genetic persistence does not necessarily imply selection. In systems where fitness is assigned randomly at birth, certain lineages may still dominate by chance, creating "telic waves" of activity that mimic true evolution.
As shown in Figure 1, a system with purely random fitness assignments can generate high levels of cumulative evolutionary activity. This false positive indicates that persistence-based metrics are insufficient for validating the presence of natural selection in complex ALife environments.
Methods: Fitness Transmission
To overcome the limitations of persistence-based metrics, the author proposes Fitness Transmission. This metric calculates the statistical correlation between the fitness of parents and their children.
Number of Grandchildren (NOGC)
The method utilizes the number of grandchildren (NOGC) as a universal proxy for reproductive success. Unlike the number of children, NOGC ensures that the offspring were themselves viable enough to reproduce, providing a more accurate measure of evolutionary fitness.
Statistical Normalization
To ensure the correlation is not spurious, two adjustments are made:
- Decoupling: A child's success naturally contributes to a parent's NOGC. To isolate the child's independent fitness, the child's own offspring are subtracted from the parent's NOGC before comparison.
- Finite Population Correction: In finite populations, the success of one lineage limits others. The parent's modified NOGC is normalized by the total grandchildren of all other parents in the same time slice.
The final metric is the Pearson correlation between these normalized parent and child NOGC values across generations.
Results
The proposed metric was tested using Genetic Algorithms (GAs) solving standard optimization benchmarks. These were compared against "neutral" versions where fitness was assigned randomly.
As illustrated in Figure 2, Fitness Transmission successfully distinguishes between the two scenarios. In the neutral system (green), the correlation remains near zero, correctly indicating the absence of heritable selection. In contrast, the standard GA (black) shows a consistently positive correlation, reflecting the transmission of adaptive traits across generations. This signal persists even when the population reaches an optimum, as selection continues to cull deleterious mutations.
Conclusion
Fitness Transmission provides a rigorous, genealogically-based signature of natural selection. By focusing on the heritability of reproductive success rather than the persistence of specific components, it avoids the false-positive pitfalls of previous metrics. While it may struggle in cases of extreme stabilizing selection or low population diversity, it serves as a powerful validation tool for researchers aiming to prove the existence of open-ended evolution in artificial systems.
Summary updated