Source-linked AI summary
Big Brains and Changing Environments: Cause or Consequence?
Sian Heesom-Green, Jonathan Shock, Geoff Nitschke
TL;DR
The paper examines whether large brains evolve because environments are changing or whether they arise earlier under stable, energy-permissive conditions. Using neuroevolution with explicit ANN energy costs, it compares agents evolved only in changing environments with agents transitioned from static ones. Prior static evolution mainly produces larger ANNs, which perform better under unpredictable changes and comparably under predictable changes.
Problem
The paper asks whether associations between large brains and changing environments imply that changing conditions caused larger brains to evolve, as the Cognitive Buffer Hypothesis proposes.
Method
The study uses neuroevolution in an artificial seasonal foraging task with explicit ANN energy costs, comparing agents evolved exclusively in changing environments with agents transitioned from static environments.
Results
Prior static evolution produced significantly larger final ANNs and superior performance under unpredictable changes, while performance differences were not significant under predictable changes.
Takeaways & Limitations
The findings support a colonization-based account in which larger brains arise under stable conditions and later confer advantages in variable environments, especially when changes are unpredictable.
Abstract
from arXiv · showhide
Large brains are metabolically costly, and associations with changing environments do not imply they evolved there, as the Cognitive Buffer Hypothesis (CBH) would suggest. They may instead evolve in stable conditions and later facilitate colonization of changing environments. Using neuro-evolution in an artificial seasonal foraging task, we compared agents evolving exclusively in changing environments to agents first evolved in static environments before transitioning. Results show that larger neural networks in dynamic environments arise mainly from prior static evolution, achieving superior performance under unpredictable changes. Our results challenge strict CBH predictions, provide agent-based (computational) support for a colonization-based account and highlight the role of evolutionary history in brain size evolution.
1 Introduction
The paper asks whether large-brain associations with changing environments reflect adaptation within those environments or prior evolution under stable conditions. It contrasts the Cognitive Buffer Hypothesis with energetic constraints and a colonization-based alternative tested through neuroevolution.
- 1 Introduction: Associations between large brains and changing environments do not establish that changing conditions caused larger brains to evolve.The Cognitive Buffer Hypothesis proposes enhanced flexibility and learning as a buffer against seasonality.
- 1 Introduction: Energy constraints may limit larger brains in variable environments, where smaller, more efficient brains can be favored.The Expensive Brain Hypothesis treats brain size as sustainable only when energy acquisition or allocation permits it.
- 1 Introduction: A colonization-based account proposes that large brains evolve in stable conditions and later facilitate expansion into variable environments.Bird evidence reports that brain-size increases typically preceded expansion into more variable habitats.
- 1 Introduction: Figure 1 depicts a 10×10 grid-world in which cell colors identify food types and the legend illustrates seasonal color changes.
- 1 Introduction: Neuroevolution provides a controlled test by comparing agents evolved only in changing environments with agents transitioned from static environments.The artificial foraging task includes predictable and unpredictable seasonal variation and explicit costs for ANN size.
2 Methods and Experiments
The study evolves neural agents in a seasonal foraging grid-world with explicit size-dependent energy costs, online learning, and transitions between static and changing environments. It evaluates evolutionary trajectories across predictable and unpredictable conditions.
- 2.1 Task Environment: Agents forage in a 10×10 grid-world, choosing movement or eating actions while observing nearby RGB colors, location, and previous reward.Each 100-step episode contains randomly placed edible and poisonous items that are replaced after consumption.
- 2.1 Task Environment: Dynamic environments change food-color mappings across seasons, with unpredictable conditions regenerating mappings at every seasonal transition.Predictable mappings remain fixed within each season, and seasons follow a fixed sequence with equal duration.
- 2.1 Task Environment: Energy increases by +1 for edible food and decreases by -1 for poisonous food, while per-step expenditure scales with ANN size.The size-dependent expenditure represents the metabolic cost of larger brains.
- 2.2 Neuroevolution: ANNs evolve with NEAT across populations of 150 genomes for 1100 generations, using speciation, selection, reproduction, mutation, crossover, replacement, and elitism.
- 2.3 Lifetime Learning via Hebbian Learning: Fitness averages reward across 10 seeded trials, each using 100 episodes with exploration in episodes 1–50 and evaluation in episodes 51–100.Weights are updated online through reward-modulated Hebbian learning, while updates do not carry across trials or generations.
- 2.4 Experiments: Experiments compare exclusive evolution in changing environments with abrupt or gradual transitions from static environments across 20 independent evolutionary runs.ANN size and task performance are recorded across transition scenarios.
3 Results & Discussion
Prior static evolution produced larger neural networks that remained larger after transition and performed better under unpredictable, but not predictable, environmental change. These results support a colonization-based account over the view that changing environments directly select for larger brains.
- Results & Discussion: Table 3 reports Bonferroni-corrected Dunn post hoc pairwise comparisons of ANN size and task performance across transition scenarios.
- Results & Discussion: Prior static evolution produced significantly larger final ANNs after transition to changing environments (KW p<0.05), despite declines in size after transition.This indicates a lasting influence of early evolutionary history.
- Results & Discussion: Under predictable changes, task performance did not differ significantly between transition scenarios (KW p≥0.05).
- Results & Discussion: Under unpredictable changes, agents with prior static evolution performed significantly better than agents evolving exclusively in changing environments (KW p<0.05).
- Results & Discussion: Associations between large ANNs and changing environments primarily reflected prior static evolution, which later conferred advantages especially when environmental changes were unpredictable.
4 Conclusions
The study tests whether large brains evolve within changing environments or arise earlier under stable conditions and later facilitate colonization. Neuro-evolution with explicit energy constraints supports the latter account, while the simplified artificial setup limits generalization to biological systems.
- Conclusions: Changing environments did not reliably select for larger neural architectures; larger ANNs primarily arose through prior evolution in static, energy-permissive environments.
- Conclusions: Agents with larger ANNs achieved comparable performance under predictable changes and superior performance under unpredictable changes relative to agents evolving exclusively in changing environments.
- Conclusions: The findings show that brain-size associations with changing environments do not necessarily imply causality and should be interpreted with evolutionary history in mind.
- Conclusions: The simplified artificial setup limits generalization to biological systems.