Learning Agents Teachable by Typical Computer Users

One way to identify new and promising directions for Machine Learning research is to define an important challenge problem that has broad applicability, and then to identify what kind of machine learning capabilities are required to solve it. This paper introduces a great challenge problem, developing agents by typical computer users, that has far reaching consequences on our society, and raises several Machine Learning research issues from a different paradigm than that used in current Machine Learning research. The paper identifies these issues and extends them to general research directions for Machine Learning. It also presents some preliminary results on approaching the challenge problem and these issues, in order to support their feasability.