Building intelligent tutoring systems presents significant challenges -- one challenge arises because tutoring is concerned with unobservable inner workings of the human brain; another results from the formidable task of knowledge representation and reasoning; still a third is due to the competing theories of teaching and learning. Over the past four decades, the intelligent tutoring systems community has made significant progress. Recently the field has witnessed some important and substantive debates, centered broadly on issues of knowledge representation, tutoring paradigms and deep infrastructures. We broaden the debate and raise new issues that can advance the state of the art in research and practice. We argue that there is little to distinguish constraint-based and cognitive tutors in terms of deep infrastructures. Further, we contend that the most productive way to pursue this debate and to more generally advance the field of intelligent tutors is to a) have the same researchers build tutors using different paradigms, b) use currently defined high standards for evaluation, c) open the knowledge bases of existing tutors and d) make the tutors themselves available for examination and testing by the intelligent tutor community.