The present study examines the use of multiple agents as experimental tools for inves- tigating human collaboration behaviors. The study suggests two types of methods to ensure the effective use of agents for human experiments: (1) instructing participants to believe that they are collaborating with actual human partners, and (2) using embod- ied agents equipped with social cues that use certain human characteristics. These two methods were investigated by conducting a controlled experiment in a collaborative problem-solving task where members had to insightfully change their perspective to discover the solution. The following two situations were investigated to determine the degree of perspective change based on previous studies on human group dynamics: (a) engaging with members with different perspectives (majority), and (b) engaging with a member with a different perspective (minority). Results showed that perspective-taking behaviors were detected in the two suggested methods. Specifically, the effects of the dynamics of group members (agents) with different perspectives had a stronger influence on perspective change. The study contributes to collaboration studies demonstrating new methodologies, such as using agents as experimental tools. Further, it presents interesting views on human interaction; interactions succeed not only when humans think they are interacting with other humans but also with artifacts that have simple designs like humans.