Baldwin Effect in Dynamic Environments
Summary of Current Results
Our computational experiments in the context of a non-zero sum game (IPD) environment shows:
- Evidences of the Baldwin effect,
- Emergence of extremely effective adaptive strategies.
Overview
The Baldwin effect is known as interactions between learning and evolution,
which suggests that individual lifetime learning can influence the course
of evolution without the Lamarckian mechanism.
Our concern is to consider
the Baldwin effect in dynamic environments, especially when there is
no explicit optimal solution through generations and it depends only
on interactions among agents. We adopted the iterated Prisoner's Dilemma
as a dynamic environment, introduced phenotypic plasticity into strategies,
and conducted the computational experiments, in which phenotypic
plasticity is allowed to evolve.
The Baldwin effect was observed
in the experiments as follows: First, strategies with enough plasticity
spread, which caused a shift from defect-oriented population to cooperative
population. Second, these strategies were replaced by a strategy
with a modest amount of plasticity
generated by interactions between learning and evolution.
By making three kinds of analysis, we have shown that
this strategy provides the outstanding performance. Further experiments
towards open-ended evolution have also been conducted so as to
generalize our results.
Papers
- Reiji Suzuki and Takaya Arita, "Meta-Pavlov: Strategies that Self-Adjust Evolution and Learning Dynamically in the Prisoner's Dilemma", IPSJ (Information Processing Society of Japan) SIG Notes, 99-GI-1, pp. 15-22, 1999 (in Japanese).
- Reiji Suzuki and Takaya Arita, "The Baldwin Effect in the Iterated Prisoner's Dilemma", Proceedings of the 13th Annual Conference of JSAI (Japanese Society for Artificial Intelligence), pp. 277-278, 1999 (in Japanese).
- Reiji Suzuki and Takaya Arita, "Simulation Analysis for the Effect of Learning on Evolution", Ninth JAMB (Japanese Association for Mathematical Biology) Symposium (JAMB Newsletter, No. 30, p. 31), 1999 (in Japanese).
- Reiji Suzuki and Takaya Arita, "How Learning Can Affect the Course of Evolution in Dynamic Environments", Fifth International Symposium on ARTIFICIAL LIFE AND ROBOTICS, Vol. 1, pp. 260-263, 2000.
- Reiji Suzuki and Takaya Arita, "Simulations and Analyses for an Interaction between Learning and Evolution: The Baldwin Effect in the Iterated Prisoner's Dilemma", Journal of Japanese Society for Artificial Intelligence, Vol. 15, No. 3, 2000 (in Japanese) (to appear).
- Takaya Arita and Reiji Suzuki, "Interactions between Learning and Evolution: The Outstanding Strategy Generated by the Baldwin Effect", Proc. of Artificial Life VII, pp. 196-205, 2000 (PDF format).
- Reiji Suzuki and Takaya Arita, "Interactions between Evolution and Learning in a Population of Globally or Locally Interacting Agents", Special Session: The interaction between evolution and learning at ICONIP-2000 (7th International Conference on Neural Information Processing), 2000.
- Takaya Arita, "A Constructive Approach toward Life based on the Images of Dynamic Fitness-Landscapes", Philosophy of Science, Philosophy of Science Society Japan (in Japanese), Vol. 33, No. 2, 2000.
- Reiji Suzuki and Takaya Arita, "The Baldwin Effect Revisited: Three Steps Characterized by the Quantitative Evolution of Phenotypic Plasticity", ECAL 2003 (in press).