Following are several general organizing principles around which the specific goals, principles, and preferences, which comprise an agent's way of appraising the world, can be selected. In practice, an agent might well be comprised of a mix of these general types. Whether or not a purely supportive model is better at helping us to achieve the pedagogical goals of the system than more controversial models is not at issue here. This might well prove to be true. On the other hand it is worth discovering, in a formal, repeatable way, which set of teaching personalities best assists the student- especially in the long term where less complex agents can become tiresome.
Mentor. The most general model, and the one usually thought of in the tutoring context, is the mentor, which is focused on the development of the student as measured by progress on the tutoring tasks. The AR would tend to model this by giving the agent goals having to do with the student's progress, and by having principles based on evaluation of progress.
Competitive. Other teaching modes exist besides that of mentor. Some teachers engage in (friendly) competition with students, and peer-learning also often has this quality. Competition is modeled in the AR by giving agents goals that are paired with different goals of the student (e.g., AR agent: ``Based on past history, I'll try to select four questions from the medium-hard group that you cannot answer. If you get more than two right, you win, and I lose.''). The relationship is characterized by the mix of goals and principles, wherein, although the AR would be distressed if it lost, it would simultaneously admire the student for winning. If the AR won it would be happy about it, and might gloat, but would also feel pity for the student.
Caring passionately about one's own (pedagogically sound) goals. Some teachers are not sensitive to a student's own goals, and do not interact much with them. Nonetheless they are effective partly because they are extremely interested in their subject. This enthusiasm can be imparted to students. In the AR this is modeled by having the agent's goals and principles based on interactions about the domain material: the AR agent may be delighted if the student asks for an explanation about some aspect of the domain. The agent may be distressed if a student decides to terminate the tutoring session before a major point has been delivered. The agent may have strong preferences for certain aspects of the domain material, and disklikes for other parts (which would then presumably be covered by a different agent). The agent can be effective at delivering content material from a particular point of view: ``Only an idiot would argue against strong data-typing! Here is why...''
Caring about the student's own perception of how well they are doing. Some teachers tune their presentations by measuring the student's own perceived state: how is the student feeling about their own progress? Are they satisfied? Are they meeting their own goals? Are they having fun? AR agents model this compassion by having goals about the student's satisfaction level, and by having principles that relate to their ability to satisfy the student's desires. Such an agent might be remorseful about presenting the too-difficult material such that the student gets frustrated.