Strategic Features for General Games

1998

This paper describes an architecture that begins with enough general knowledge to play any board game as a novice, and then shifts its decision-making emphasis to learned, game-specific, spatiallyoriented heuristics. From its playing experience, it acquires game-specific knowledge about both patterns and spatial concepts. The latter are proceduralized as learned, spatially-oriented heuristics. These heuristics represent a new level of feature aggregation that effectively focuses the program's attention. While training against an external expert, the program integrates these heuristics robustly.

2022

In many board games and other abstract games, patterns have been used as features that can guide automated game-playing agents. Such patterns or features often represent particular configurations of pieces, empty positions, etc., which may be relevant for a game's strategies. Their use has been particularly prevalent in the game of Go, but also many other games used as benchmarks for AI research. Simple, linear policies of such features are unlikely to produce state-of-the-art playing strength like the deep neural networks that have been more commonly used in recent years do. However, they typically require significantly fewer resources to train, which is paramount for large-scale studies of hundreds to thousands of distinct games. In this paper, we formulate a design and efficient implementation of spatial state-action features for general games. These are patterns that can be trained to incentivise or disincentivise actions based on whether or not they match variables of the s...

… JOINT CONFERENCE ON …, 1997

Game Theory is mainly seen as a mathematical theory which tries to replace pure chance and intuitive behavior in a competitive situation by calculations. This theory has been widely used to define computer programs. The aim of the research described here is to design an ...

2007 IEEE Symposium on Computational Intelligence and Games, 2007

Accuracy of evaluation functions is one of the critical factors in computer game players. Evaluation functions are usually constructed manually as a weighted linear combination of evaluation features that characterize game positions. Selecting evaluation features and tuning their weights require deep knowledge of the game and largely alleviates such efforts.

IEEE Transactions on Computational Intelligence and AI in Games, 2000

General Game Playing competitions provide a framework for building multigame playing agents. In this paper, we describe an attempt at the implementation of such an agent. It relies heavily on our knowledge-free method of automatic construction of approximate state evaluation function based on game rules only. This function is then employed by one of the two game tree search methods: MTD(f) or Guided UCT, the latter being our proposal of an algorithm combining UCT with the usage of an evaluation function. The performance of our agent is very satisfactory when compared to a baseline UCT implementation.

2008

General Game Playing (GGP) aims at developing game playing agents that are able to play a variety of games and, in the absence of pre-programmed game specific knowledge, become proficient players. Most GGP players have used standard tree-search techniques enhanced by automatic heuristic learning. The UCT algorithm, a simulation-based tree search, is a new approach and has been used successfully in GGP. However, it relies heavily on random simulations to assign values to unvisited nodes and selecting nodes for descending down a tree. This can lead to slower convergence times in UCT. In this paper, we discuss the generation and evolution of domain-independent knowledge using both state and move patterns. This is then used to guide the simulations in UCT. In order to test the improvements, we create matches between a player using standard the UCT algorithm and one using UCT enhanced with knowledge.

2021 IEEE Conference on Games (CoG), 2021

Many games often share common ideas or aspects between them, such as their rules, controls, or playing area. However, in the context of General Game Playing (GGP) for board games, this area remains under-explored. We propose to formalise the notion of "game concept", inspired by terms generally used by game players and designers. Through the Ludii General Game System, we describe concepts for several levels of abstraction, such as the game itself, the moves played, or the states reached. This new GGP feature associated with the ludeme representation of games opens many new lines of research. The creation of a hyper-agent selector, the transfer of AI learning between games, or explaining AI techniques using game terms, can all be facilitated by the use of game concepts. Other applications which can benefit from game concepts are also discussed, such as the generation of plausible reconstructed rules for incomplete ancient games, or the implementation of a board game recommender system.

2012

General Game Playing (GGP) is a specially designed environment for creating and testing competitive agents which can play variety of games. The fundamental motivation is to advance the development of various artificial intelligence methods operating together in a previously unknown environment. This approach extrapolates better on real world problems and follows artificial intelligence paradigms better than dedicated single-game optimized solutions. This paper presents a universal method of constructing the heuristic evaluation function for any game playable within the GGP framework. The algorithm embraces distinctive discovery of candidate features to be included in the evaluation function and learning their correlations with actions performed by the players and the game score. Our method integrates well with the UCT algorithm which is currently the state-of-the-art approach in GGP.

Aaai, 1999

Computer programs now play many board games as well or better than the most expert humans. Human players, however, learn, plan, allocate resources, and integrate multiple streams of knowledge. This paper highlights recent achievements in game playing, describes some cognitively-oriented work, and poses three related challenge problems for the AI community. Game Playing as a Domain Work on games has had several traditional justifications. Given unambiguous rules, playing a game to win is a well-defined problem. A game's rules create artificial world states whose granularity is explicit. There is an initial state, a state space with clear transitions, and a set of readily describable goal states. Without intervening instrumentation, games are also noise-free. For these reasons, as well as for their ability to amuse, games have often been referred to as "toy domains." To play the most difficult games well, however, a program must contend with fundamental issues in AI: knowledge representation, search, learning, and planning. There are two principal reasons to continue to do research on games, despite Deep Blue's triumph (Hamilton and Hedberg 1997). First, human fascination with game playing is long-standing and pervasive. Anthropologists have catalogued popular games in almost every culture. Indeed, the same game, under various names, often appears on many continents (Bell 1969; Zaslavsky 1982). Games intrigue us because they address important cognitive functions. In particular, the games humans like to play are probably the ones we are good at, the ones that capitalize on our intellectual strengths and forgive our weaknesses. A program that plays many games well must simulate important cognitive skills. The second reason to continue game-playing research is that some difficult games remain to be won, games that people play very well but computers do not. These games clarify what our current approach lacks. They set challenges for us to meet, and they promise ample rewards. This paper summarizes the role of search and knowledge in game playing, the state of the art, and recent relevant data on expert human game players. It then shows how cognitive skills can enhance a game-playing program, and poses three new challenge problems for the AI community. Although rooted in game playing, these challenges could enhance performance in many domains.

The Twenty-Third AAAI Conference on Artificial …, 2008

The aim of General Game Playing (GGP) is to create intelligent agents that automatically learn how to play many different games at an expert level without any human intervention. The most successful GGP agents in the past have used traditional game-tree search combined with an automatically learned heuristic function for evaluating game states. In this paper we describe a GGP agent that instead uses a Monte Carlo/UCT simulation technique for action selection, an approach recently popularized in computer Go. Our GGP agent has proven its effectiveness by winning last year's AAAI GGP Competition. Furthermore, we introduce and empirically evaluate a new scheme for automatically learning search-control knowledge for guiding the simulation playouts, showing that it offers significant benefits for a variety of games.