Detecting intelligent agent behavior with environment abstraction in complex air combat systems

Abstract

Intelligence can be defined as an emergent property in some types of complex systems and may arise as a result of an agent's interactions with the environment or with other agents either directly or indirectly through changes in the environment. Within this perspective, intelligence takes the form of an `observer' phenomenon; externally observed at a level higher than that of agents situated in their environment. Such emergent behavior sometimes may be reduced to the fundamental components within the system and its interacting agents and sometimes it is a completely novel behavior involving a new nomenclature. When emergent behavior is reducible to its parts it is considered to be a `weak' form of emergence and when emergent behavior cannot be reduced to its constituent parts, it is considered to be a `strong' form of emergence. A differentiating factor between these two forms of emergent phenomena is the usage of emergent outcomes by the agents. In weak emergence there is no causality, while in strong emergence there is causation as a result of actions based on the affordances emergent phenomena support. Modeling a complex air combat system involves modeling agent behavior in a dynamic environment and because humans tend to display strong emergence, the observation of emergent phenomena has to exist within the knowledge boundaries of the domain of interest so as not to warrant any new nomenclature for the computational model at the semantic level. The emergent observed phenomenon has to be semantically tagged as `intelligent' and such knowledge resides within the bounds of the semantic domain. Therefore, observation and recognition of emergent intelligent behavior has been undertaken by the development and use of an Environment Abstraction (EA) layer that semantically ensures that strong emergence can be modeled within an agent-platform-system, such as Live, Virtual and Constructive (LVC) training in a Distributed Mission Operations (DMO) testbed. In the present study, various modeling architectures capable of modeling/mimicking human type behavior or eliciting an expected response from a human pilot in a training environment are brought to bear at the semantic interoperability level using the EA layer. This article presents a high level description of the agent-platform-system and how formal modeling and simulation approaches such as Discrete Event Systems (DEVS) formalism can be used for modeling complex dynamical systems capturing emergent behavior at various levels of interoperability. The ideas presented in this paper successfully achieve integration at the syntactic level using the Distributed Interactive Simulation (DIS) protocol data units and semantic interoperability with the EA layer.