Four Jeopardy! champions competed against CMSC 470 Natural Language Processing Students’- Question Answering Systems

Students in Associate Professor Jordan Boyd-Graber’s course used the tools of natural language processing—information retrieval representations, distributed semantics, sequence models, and reinforcement learning—to create question answering systems

On May 13, four former Jeopardy! champions took on students’ final projects to see whether trivia whizzes or artificial intelligence could better answer questions

These question answering systems played a game called quiz bowl.  Unlike Jeopardy! where you can only answer at the end of the question, you can interrupt a quiz bowl question when you know the answer. Trivia aficionados consider this a “truer” test of who knows more about a given subject, particularly because the questions are structured so that easier clues that more people know follow the initial obscure clues.  If you can answer on the obscure clues, you must be an expert.

The experts who took on the undergraduates’ systems were Roger Craig—whose one day Jeopardy! record stood for a decade until broken by James Holzhauer, Monica Thieu—who placed second in Jeopardy’s first All-Star tournament, Kristin Sausville—a five time Jeopardy! Champion who previously defeated Boyd-Graber’s question answering software, and Aaron Lichtig—another Jeopardy! Champion.

Members of the public also came out to the event to see Jeopardy! celebrities, including Sara DelVillano, a band teacher from Lanham, Maryland, who later that night became a Jeopardy! Champion herself in the Teacher’s Tournament. 

These events are part of Boyd-Graber’s project, Question Answering is Not a Trivia Activity, which uses question answering events to help explain what AI can (and cannot) do to the general public.

The Department welcomes comments, suggestions and corrections.  Send email to editor [at] cs [dot] umd [dot] edu.