The Boundedness Illusion: Projections from early evolution can underestimate evolutionary potential

Open-ended evolution researchers seek to create systems that continually produce “new” evolutionary outcomes, attempting to mimic the power and diversity of evolution in nature. The specific metrics used (novelty, complexity, diversity, etc) vary by researcher, but the holy grail would be a system where any of these can accumulate indefinitely. Of course, one challenge that we face in reaching this goal is even recognizing if we have succeeded. To determine the evolutionary potential of a system, we must conduct finite experiments; based on their results we can predict how we would expect evolution to progress were the run to have continued. Here we begin to explore how such predictions might be made and how accurate they might be. In this initial study, we focus on predicting fitness; this metric can be easy to calculate, and often correlated with increases in traits like novelty and complexity. We find the best fit to measured values of fitness in a simple digital evolution experiment, and demonstrate that projecting this fit out usually predicts that fitness will be constrained by an asymptote. Extending the experiments, however, we see that fitness often shoots past this asymptote, belying the boundedness that it implies. Extending past a premature end point allows us to see through this “boundedness illusion”.