The Red Queen Effect in a Predator–Prey Co-Evolutionary System

🧬 Introduction

The Red Queen Effect, inspired by Lewis Carroll's Through the Looking Glass, describes an evolutionary dynamic where species must continually adapt just to maintain their relative fitness. In co-evolutionary systems, this manifests as reciprocal adaptation — a change in one species drives evolutionary pressure on the other.

In this experiment, we test for the presence of the Red Queen effect using a Predator–Prey–Grass environment with multi-agent reinforcement learning (RLlib, PPO). We restrict the system to:

• type_1_predator: a single predator archetype
• type_1_prey: a single prey archetype
  to isolate pure co-evolutionary interactions without intra-species competition.

We assess whether adaptation is required for either population to maintain fitness by applying a freeze–unfreeze test. This involves fixing (freezing) one agent population’s policy while allowing the other to continue evolving.

---

🔬 Experimental Setup

Four evaluation scenarios were created by combining PPO checkpoints from two generations:

• checkpoint_iter_500 (early generation)
• checkpoint_iter_1000 (later generation)

We then ran:

Experiment	Predator Source	Prey Source	Purpose
Frozen Prey	checkpoint_iter_1000	checkpoint_iter_500	Test if predator evolved advantage
Frozen Predator	checkpoint_iter_500	checkpoint_iter_1000	Test if prey evolved advantage
Static Baseline	checkpoint_iter_500	checkpoint_iter_500	Baseline (no evolution)
Fully Co-Evolved	checkpoint_iter_1000	checkpoint_iter_1000	Full mutual adaptation

Each evaluation ran for 1000 steps in the same grid-based environment.

---

📊 Results

Experiment	Avg Prey Offspring	Avg Prey Lifespan	Total Prey Offspring	Total Reward
Frozen Prey	0.96	4.76	491	6350
Frozen Predator	0.96	6.01	523	6660
Static Baseline	0.96	4.70	500	6500
Fully Co-Evolved	0.96	8.19	529	6680

Observations:

• All setups produced similar offspring counts and average rewards — but lifespan diverged significantly.
• The co-evolved prey population lived longer than in any other setup, despite no increase in reproduction rate.
• Freezing predators allowed prey to increase their lifespan from 4.7 to 6.01 steps.
• Freezing prey resulted in shorter lifespans for prey, suggesting that predator adaptation alone is not enough to secure ongoing advantage.

Display 1: Life span comparison

---

✅ Conclusion

This experiment provides strong empirical evidence of the Red Queen effect in our co-evolutionary predator–prey simulation.

Key conclusions:

• Mutual co-adaptation yields the best prey performance (↑ lifespan)
• Freezing one side reduces the other's contextual fitness — indicating that fitness is relative and dependent on opponents
• Predator improvement depends on the ability to respond to evolving prey — not just isolated learning

In essence:

"It takes all the running you can do, to stay in the same place."
— The Red Queen, Lewis Carroll

This test confirms that evolutionary pressure is bidirectional and ongoing adaptation is necessary to maintain performance — a defining hallmark of Red Queen dynamics.

---

🔁 Future Work

Next steps may include:

• Running across multiple random seeds for statistical validation
• Introducing type_2 predator/prey variants to observe strategy replacement
• Logging mutation rates and lineage turnover for evolutionary depth analysis
• Visualizing arms race dynamics across generations