Evolution, neural networks, games, and intelligence

Abstract

Intelligence pertains to the ability to make appropriate decisions in light of specific goals and to adapt behavior to meet those goals in a range of environments. Mathematical games provide a framework for studying intelligent behavior in models of realworld settings or restricted domains. The behavior of alternative strategies in these games is defined by each individual's stimulusresponse mapping. Limiting these behaviors to linear functions of the environmental conditions renders the results to be little more than a façade: effective decision making in any complex environment almost always requires nonlinear stimulus-response mappings. The obstacle then comes in choosing the appropriate representation and learning algorithm. Neural networks and evolutionary algorithms provide useful means for addressing these issues. This paper describes efforts to hybridize neural and evolutionary computation to learn appropriate strategies in zero-and nonzero-sum games, including the iterated prisoner's dilemma, tictac-toe, and checkers. With respect to checkers, the evolutionary algorithm was able to discover a neural network that can be used to play at a near-expert level without injecting expert knowledge about how to play the game. The implications of evolutionary learning with respect to machine intelligence are also discussed. It is argued that evolution provides the framework for explaining naturally occurring intelligent entities and can be used to design machines that are also capable of intelligent behavior.

Employing Game Theory and Computational Intelligence to Find the Optimal Strategy of an Autonomous Underwater Vehicle against a Submarine

“…Among many benefits of neural networks is an application for nonlinearly separable problems and generalization of acquired information [22]. In this study, Multi-layer Perceptrons (MLP) were used as a popular model often applied in a broad class of problems [5].…”

Relationship Between Generalization and Diversity in Coevolutionary Learning

“…A game has a specific set of rules that constrains strategies to certain behaviors (legal moves), with goals for strategies to meet (to win the game), and rewards for those that better achieve the goals under constraints of finite resources (payoff for a move). A game has enough subtleties to allow representation of a wide range of complex behaviors (a diverse set of strategies) [11]. Games can capture intrinsic properties of complex, real-world problems where EC methodologies are developed to obtain solutions, which can range from solutions (strategies) for board games to economic games [11], [12], but are sufficiently simple to enable extensive and in-depth analysis of EC methodologies.…”

“…A game has enough subtleties to allow representation of a wide range of complex behaviors (a diverse set of strategies) [11]. Games can capture intrinsic properties of complex, real-world problems where EC methodologies are developed to obtain solutions, which can range from solutions (strategies) for board games to economic games [11], [12], but are sufficiently simple to enable extensive and in-depth analysis of EC methodologies.…”

Scripting the game of Lemmings with a genetic algorithm

“…Later Schaeffer et al [11], [12] developed an automated Checkers player called Chinook that took the world title from Marion Tinsley who had been champion for over forty years. Chellapilla and Fogel [13], [14] and Fogel [15] co-evolved a checkers program in the late 1990's that was able to defeat expert players. In contrast to Deep Blue and Chinook, that both used large opening and endgame databases, Chellapilla and Fogel's program started with no knowledge of the game and learned strategies through co-evolution.…”

“…Chellapilla and Fogel [13], [14] and Fogel [15] successfully co-evolved a program to play a game of checkers. Members of the population played against each other to compete for survival into the next generation.…”