Introduction Should computer generated forces (CGF) systems include automated learning capabilities? The CGF research literature contains many statements by CGF experts that the ability to learn will be generally valuable, even necessary, in future CGF systems. A variety of significant benefits for CGF systems and military simulation in general are claimed to follow from automated learning. However, upon closer examination, it seems to be not so obvious that learning by CGF systems would

more »

... rily be beneficial for many uses of CGF systems. This paper takes a respectfully skeptical position regarding CGF learning and provides arguments that CGF learning could compromise and confound the utility of CGF systems for the most common CGF applications. This paper begins by defining CGF systems and grouping CGF simulation applications into three broad types. Calls in the CGF research literature for automated learning by CGF systems are surveyed. Categories of learning-modified behavior for CGF systems are defined based on what behaviors have been learned. Arguments are given, organized by application and behavior category, explaining how learning could increase and/or reduce the utility of the CGF system for the application. Real and notional examples are provided. Finally, specific applications where learning by CGF systems might be useful are identified.