In many fields including healthcare, accurate perceptions of one’s conversational partner can improve information exchange and outcomes. However, perceptions of conversational partners are difficult to validate and improve. Improving the accuracy of perceptions (e.g. avoiding stereotypes) requires learners to actively participate in improving their mental models of others and receive timely feedback. Creating an accurate perception can be achieved by transferring the mental model of the conversational partner to a tangible medium, and then validating that model. This validation can be achieved by sharing one’s current perception with other people and receiving feedback about the perception. The feedback enables one to either validate or revise one’s mental model. To achieve those goals, we propose to apply the theory of constructionist learning to have learners create, share, and receive feedback on virtual human (VH) representations of their conversational partners.