Machine Learning for Intelligent Systems

Abstract

Recent research in machine learning has focused on supervised induction for simple classification and reinforcement learning for simple reactive behaviors. In the process, the field has become disconnected from AI's original goal of creating complete intelligent agents. In this paper, I review recent work on machine learning for planning, language, vision, and other topics that runs counter to this trend and thus holds interest for the broader AI research community. I also suggest some steps to encourage further research along these lines. Introduction A central goal of artificial intelligence has long been to construct a complete intelligent agent that can perceive its environment, generate plans, execute those plans, and communicate with other agents. The pursuit of this dream naturally led many researchers to focus on the component tasks of perception, planning, control, and natural language, or on generic issues that cut across these tasks, such as representation and search. Over ...