Virtual humans are computer-generated characters that can take the part of humans in a variety of limited contexts. These can include acting as role-players in simulations and training systems, where they play a variety of parts, such as acting as friendly or hostile forces, or locals in the environment. Other uses for virtual humans include acting as museum guides, marketing assistants or characters in entertainment systems. Because virtual humans are intended to mimic a broad range of human behaviors, they must integrate a diverse set of AI technologies, including speech recognition, natural language understanding and generation, dialogue management, non-verbal communication, perception, automated reasoning, and emotion modeling. In this talk I will give an overview of virtual human research, focusing on the opportunities this area provides and the lessons learned and their larger implications for the field of artificial intelligence. I will conclude with some thoughts on future directions for virtual human research.