“Two heads are better than one.” New knowledge can be generated by the collective knowledge of groups through collaboration. However, past studies have shown that simply adding manpower to a group may not facilitate its performance. It is important to consider the types of interactions that may facilitate the activity within the group. Multiple studies in cognitive science using field studies, experimentation, and computational simulations have shown what types of interactions facilitate performance during collaborations. In our lab, we investigate collaborative problem solving in groups with members that have different perspectives and see when and how these different perspectives can become a powerful source to facilitate the group’s problem-solving performance. We also use conversational agents as experimental tools to control the conversational behaviors of the collaborative members and to specify the effective interactions in collaborative problem solving.