How Should Prey Animals Respond to Uncertain Threats?

Joel Zylberberg, Michael Robert DeWeeseFrontiers Comput. Neurosci 2011 · OpenReview

TL;DR

Prey animals face a constant trade-off between fleeing a potential predator and conserving energy/mating opportunities. This paper shows that uncertainty about the predator's presence fundamentally changes the optimal escape strategy, making economic factors crucial. Computer simulations confirm that animals naturally evolve to adopt these uncertainty-driven strategies.

What Problem Does This Paper Address?

When a prey animal encounters a potential threat, it must decide whether to flee or not. Fleeing too late means injury or death (Type I error), while fleeing unnecessarily wastes energy and misses mating or foraging opportunities (Type II error). While previous models have explored these trade-offs, they often assumed prey animals had perfect information about their environment. This paper tackles the crucial question of how uncertainty itself influences the optimal escape strategy, arguing that it's not just the likelihood of danger but the degree of certainty that dictates behavior.

Methods

The authors employ a probabilistic automaton model to represent the life cycle of a prey animal. This model captures discrete time steps and various states (start, flee, mate, struggle, death). The core of their approach is a decision-theoretic framework where animals aim to maximize their rate of reproduction (a proxy for evolutionary success). They use Lagrange multipliers to find the optimal escape strategy by maximizing this objective function subject to constraints imposed by the animal's sensory capabilities (represented by Receiver Operating Characteristic - ROC - curves). To validate their analytical findings, they conduct computer simulations of a population of prey animals, allowing their escape strategies to evolve naturally.

Datasets

Not discussed in the paper. The study uses simulated data generated by their probabilistic automaton model.

Key Insights and Ideas

  • Uncertainty is Key: The presence of uncertainty fundamentally shifts the optimal escape strategy. When certainty is high, animals simply flee if danger is likely. However, as uncertainty increases, economic factors (like lost mating opportunities) become critical in determining the threshold for fleeing.
  • Economic Factors Emerge with Uncertainty: The paper demonstrates that economic considerations only become relevant in determining the escape threshold when the animal is uncertain about the threat. This is a significant departure from previous models that assumed perfect information.
  • Evolution Favors Uncertainty-Driven Strategies: Computer simulations show that populations of prey animals, when subjected to predation and allowed to evolve, naturally adopt the escape strategies predicted by the model that incorporates uncertainty and economic factors. This validates the authors' choice of objective function.
  • "Score" as a Decision Metric: The model uses a "score" (potentially from a neural network or sensory input) to represent the animal's assessment of danger. The decision threshold on this score is what evolves.

Key Results

  • The optimal escape strategy shifts from a simple "flee if likely dangerous" to a more nuanced approach influenced by economic costs when uncertainty is high.
  • For Gaussian-distributed threat scores, the optimal threshold deviates from a maximum likelihood estimator as uncertainty increases, with economic factors playing a larger role.
  • Computer simulations confirm that a population of prey animals, through evolutionary processes, converges to the predicted optimal escape strategies derived from the model. For example, in one simulation, the average escape threshold asymptoted to -1.605, closely matching the theoretical prediction of -1.645.

Limitations and Future Work

  • Simplified Model: The probabilistic automaton model omits factors like maturation periods, learning over a lifetime, and sexual reproduction (though they note simulations showed sexual reproduction didn't change results).
  • Innate vs. Learned Behavior: The study focuses on innate, instinctual responses rather than learned behaviors, which are also important in real-world scenarios. Combining these is a direction for future work.
  • Generalizability of Distributions: While the conclusion about economic factors becoming more important with uncertainty holds for Gaussian and other unimodal exponential distributions, it's not proven for all possible score distributions.
  • Quantitative Prediction Difficulty: While the paper addresses the correct objective function and the role of uncertainty, accurately modeling real prey animals' life cycles with probabilistic automata remains a challenge for making precise quantitative behavioral predictions.
  • Future Directions: The authors suggest generalizing the model to continuous distances and exploring the integration of learned behaviors with innate responses. They also propose that this decision-theoretic framework can be applied to other biological systems where decisions are made under uncertainty with unequal error costs.

Summary updated

Abstract

A prey animal surveying its environment must decide whether there is a dangerous predator present or not. If there is, it may flee. Flight has an associated cost, so the animal should not flee if there is no danger. However, the prey animal cannot know the state of its environment with certainty, and is thus bound to make some errors. We formulate a probabilistic automaton model of a prey animal's life and use it to compute the optimal escape decision strategy, subject to the animal's uncertainty. The uncertainty is a major factor in determining the decision strategy: only in the presence of uncertainty do economic factors (like mating opportunities lost due to flight) influence the decision. We performed computer simulations and found that in silico populations of animals subject to predation evolve to display the strategies predicted by our model, confirming our choice of objective function for our analytic calculations. To the best of our knowledge, this is the first theoretical study of escape decisions to incorporate the effects of uncertainty, and to demonstrate the correctness of the objective function used in the model.
BibTeX
@inproceedings{zylberberg2011how,
  author = {Joel Zylberberg and Michael Robert DeWeese},
  title = {{How Should Prey Animals Respond to Uncertain Threats?}},
  booktitle = {Frontiers Comput. Neurosci},
  year = {2011},
  url = {https://openreview.net/forum?id=seza-R5ul0k},
}
Figure