This position paper argues the case for the application of constructivist theories to Artificial Intelligence, with particular emphasis on Piaget's theory. The idea of building an artificial baby is an old one in Artificial Intelligence, yet it is difficult to execute because little is known about the information processing mechanisms which babies use to learn. That which is known comes from non computing disciplines and has not been exploited very much in Artificial Intelligence. Part of the difficulty is that many AI researchers do not know enough about these other disciplines; another difficulty is that where AI researchers do know some of the theories from other disciplines, many do not see their value. This paper tries to make a case for the value of Piaget's theory in particular.