Learning to Engage in Conversations for AI Systems2018-2019

People are interacting with artificial intelligence (AI) systems more every day. AI systems play roles in call centers, mental health support, and workplace team structures. As AI systems enter these human environments, they  inevitably will need to interact with people in order to achieve their goals. Most AI systems to date, however, have focused entirely on performance and rarely, if at all, on their social interactions with people, and how to balance the AI’s goals against their human collaborators’ goals. Success requires learning quickly  how to interact with people in the real world. Stanford Computer Scientists Ranjay Krishna and Apoorva Dornadula were awarded a Magic Grant to create a conversational AI agent on Instagram, where it will learn to ask engaging questions of people about the photos they upload. Its goal will be to simultaneously learn new facts about the visual world by asking questions, and learn how to interact with people around their photos in order to expand its knowledge of those concepts.