In the well-known two sigma problem introduced in 1984, Bloom found that students tutored by a one-on-one mastery-learning tutor achieved a learning outcome two standard deviations higher than those tutored by traditional learning methods. Since one-on-one tutoring is too costly to scale to billions of students in society, technology has been perceived as a promising tool to simulate a one-on-one tutoring experience. However, current electronic learning systems still primarily consist of learning activities with limited interactions such as multiple-choice questions, review-and-flip flash cards, and listen-and-repeat practices. With recent advances in artificial intelligence (AI), we now have the potential to create conversation-based tutoring systems with the capability of providing personalized feedback to make learning much more engaging and effective, and eventually help bridge the gap between one-on-one human tutoring and computer tutoring.

In this talk, I present the design, development, and testing of four AI-based conversational tutoring systems that personalize learning for both adults and children. For adult learning, I present two systems, QuizBot for helping college students learn factual knowledge and EnglishBot for tutoring language learners in speaking English. For child learning, I present two systems integrating narratives into conversational tutoring systems, BookBuddy for helping children develop reading comprehension abilities and SmartPrimer for supplementing elementary school students’ math learning. I conduct human evaluations with these tutoring systems to understand how humans interact with AI in the educational setting. The test with over 500 students show that conversation-based tutoring systems that leverage new natural language processing (NLP) and reinforcement learning (RL) techniques can provide adaptive feedback, engage students more, motivate them to spend more time using the tutor, and improve student learning outcomes, when compared to current learning systems.