Integrating AI and optimization for decision support: A survey

Optimization models have been the workhorses of computer based decision support. Their emphasis is on model structure, quantification and solution efficiency. However, there also are important meta-modeling, analysis and interpretation activities associated with practical decision making. AI methods, because of their aim of emulating human reasoning and thinking activities, have the potential of providing computer based support for these other decision making activities. There has been a growing literature on the integration of AI and optimization techniques for decision support. Much of this body of work describes techniques that are application or problem specific. Others describe more general methods addressing different specific aspects of decision making. In this paper we use a conceptual framework to survey and analyze these efforts. Our survey shows that efforts to integrate AI and Optimization have been focused mainly on model formulation and selection. Other activities such as post solution analysis or solver selection, have received considerably less attention for automated support. The dramatic difference in paradigms between AI and Optimization result in vastly different data structures and control primitives in their respective software implementations. We conjecture that this disparity will continue to thwart the development of general tools for seamless AI/Optimization integration.