This paper describes the development and evaluation of an Intelligent Team Tutoring System (ITTS) for pairs of learners working collaboratively to monitor an area. In the Surveillance Team Tutor (STT), learners performed a surveillance task in a virtual environment, communicating to track hostile moving soldiers. This collaborative problem solving task required significant communication to achieve the common goal of perfect surveillance. In a pilot evaluation, 16 twoperson teams performed the task within one of three feedback conditions (Individual, Team, or None) across four trials each. The STT used a unique approach to filtering feedback so that teams in both individual and team conditions received a similar amount of feedback. In one performance measure, Team condition participants made fewer errors in one task than those in other conditions, though at a potential cost of mental workload. Feedback condition also significantly affected participants' subjective rating of both their own performance and their teammate's. This ITTS is one of the first automated team tutoring systems that provided real-time feedback during task execution. Recommendations are offered for the design of the optimal team task for future ITTSs that offer tutoring for small teams performing collaborative problem solving.

This chapter describes five disciplinary domains of research or lenses that contribute to the design of a team tutor. We focus on four significant challenges in developing Intelligent Team Tutoring Systems (ITTSs), and explore how the five lenses can offer guidance for these challenges. The four challenges arise in the design of team member interactions, performance metrics and skill development, feedback, and tutor authoring. The five lenses or research domains that we apply to these four challenges are Tutor Engineering, Learning Sciences, Science of Teams, Data Analyst, and Human-Computer Interaction. This matrix of applications from each perspective offers a framework to guide designers in creating ITTSs.

Teams have the ability to achieve goals that are unobtainable by individuals alone. However, there is little agreement on a standard model for researching the performance of distributed teams. Initial pilot results suggest that the Multiple Errands Test (MET), when adapted to a team in a virtual environment, is a platform for evaluating the impact of feedback characteristics. To demonstrate the potential of the Team MET as a platform for future team research in the broader CSCW community, an example study is described in which team members are given feedback in one of four conditions: individual private, team private, individual public, and team public.

Challenges arise when developing a computer-based Intelligent Team Tutoring System (ITTS) that attempts to deliver feedback to teams as effectively as a human tutor. The purpose of this current work is to outline elements of feedback that should be considered when designing feedback for an ITTS. The authors present the results of a study that consisted of 32 participants grouped into 16 teams of two. Each team conducted a surveillance task where they received individual or team feedback. Feedback content was written using either the bald (direct feedback; no need for interpretation) or off-record (general feedback; interpretation needed) etiquette strategy. The results showed that feedback delivered using the bald etiquette strategy positively correlated with improved performance. The results also showed that team level feedback positively correlated with more accurate self-assessment among participants. This suggests that in an ITTS, direct feedback can lead to better performance, and that feedback provided at the team level can help to align self-interpretation of performance with actual task performance.

I would like to thank my wife for her love and support during the development of this thesis. I would also like to thank my committee for helping and guiding me through my thesis. I would also like to thank Kelsey, Sam, and Anton for jumpstarting this work by developing the initial version of the final program used for this current work. It is also important to note that this