AI IN GAMES

AI in Games



AI in Games

1. Introduction

The goal of any situation awareness system is to keep track of and use available information in a proper and timely manner to support planning and decision-making . Due to huge amounts of data the establishment of situation awareness must be seen as an information fusion process starting with sensory data that is propagated upwards being fused into comprehensible information (Belz, 2003, 68-99). However, merely presenting a comprehensible description of the situation, the situation picture, does not give a complete understanding of the development of a situation. Hence, the last step in the fusion process is the prediction step where one tries to predict opponent plans and suggest future courses of actions (Belz, 2003, 68-99).

The challenges and difficulties when it comes to prediction are fundamentally different depending on available time and resources. In this paper we focus on long-term operative decision-making in command centers. These situations arise in command and control (C2) where the decision situation is characterized by its multi-agent planning perspective where one wants to make predictions using historic events as well as make look-ahead predictions using assumptions regarding future events. The typical civilian example scenario is a disaster relief scenario and the typical military example is a C2 scenario where commanders make long-term decisions on the operative level. In a data fusion context these situations belong to “threat prediction” according to the so-called JDL model and , which is used to structure information fusion processes (Belz, 2003, 68-99). This has been recognized in for example Ref. where the authors suggest the use of game-theoretic algorithms for the estimation process in higher level data fusion. Indeed, game-theoretic reasoning should be used for multi-agent decision-making in command centers that need to make decisions regarding long-term goals. After all, game theory is not just another tool in a toolbox — it is the toolbox itself (Berners, 1999, 12).

There is a current discussion regarding the relevance of decision theory for human decisions, based on the perception that human decision-makers in experiments do not seem to follow the theory. While the discussion is highly relevant and potentially troublesome, we observe that there are no serious alternatives and we rely on Weibull's defense saying that the experiments where violations are observed are often designed so that it is questionable whether the perceived subject utility of outcomes represents what the designer had in mind .

One goal of artificial intelligence (AI) has been to create expert systems, i.e., systems that can, provided the appropriate domain knowledge, match the performance of human experts (Broad, 2004, 135-151). Such systems do not yet exist, other than in highly specific domains, but AI research has implied that researchers from widely differing fields have come together in order to solve questions regarding knowledge representation, decision-making, autonomous planning, etc. The results provide a good ground for the construction of C2 decision support systems. During the last decade, the intelligent agent perspective has lead to a view of AI as a system of agents embedded ...