Improving the quality of communication between healthcare providers and patients has the potential to improve outcomes. Simulated role-playing can be employed to teach and practice social communication skills, however existing simulation technologies fail to deliver the nuance, variety, and richness of actual human interaction. This talk presents a novel approach to dialogue and social interaction, which simulates interaction from a large database of recorded human interactions. Combining artificial intelligence, natural language processing, and crowdsourcing, can deliver an authentic experience, facilitating transfer of knowledge from the virtual world to the real world. Unlike traditional approaches to serious games, which rely on scripts and multiple choice dialogue options, this talk will demonstrate how a data-driven approach supports open-ended dialogue, where the user learns by expressing him or herself in his/her own words.