Home > News > Human vs. Computer Quiz Bowl event offers first fair fight for humans

Human vs. Computer Quiz Bowl event offers first fair fight for humans

Jordan Boyd-Graber

 

By Grace Hynes and Tess Wood

(College Park, MD. – May 30, 2018)  Design a robot that plays Jeopardy? Check! Design a robot that plays Quiz Bowl- the elite scholastic knowledge competition where teams of players “buzz in” as soon as they know the answer? That’s trickier. Quiz Bowl poses a special natural language processing challenge for artificial intelligence when compared to other trivia competitions such as Jeopardy since contestants can interrupt the question as soon as they think they know the answer. When hearing a question word-by-word, at what point has enough information been received to make an appropriate guess? An incorrect answer based on too little information costs points, but delaying too long allows the other team to answer first. Last year, Japan’s Studio OUSIA quiz bowl system beat a team of elite human Quiz Bowl champions assembled at the Conference on Neural Information Processing Systems, while the University of Maryland and the University of Colorado’s QANTA machine has beaten 74-time Jeopardy champion and Quiz Bowl Expert Ken Jennings.

These scientists had to create a system that can make good predictions when information is incomplete and identify the best point to answer a question. The research behind these tasks has valuable applications in other areas - including simultaneous machine translation, which poses closely-related problems. When translating between languages with very different word orders, at some point in a sentence a human or machine is forced to either stop and wait for a crucial word or to make a guess and forge ahead, just like in Quiz Bowl.

While QANTA and Studio OUSIA’s quiz bowl system have won many matches, Jordan Boyd-Graber and others at the University of Maryland believe that current Quiz Bowl questions are not testing the technology to its limits. Quiz Bowl uses questions designed to stump humans, not computers. Their questions are in a pyramid format. The hardest clues are at the beginning and easiest clues at the end of the question. With the entirety of Wikipedia at their disposal and near infinite memory, question-answer machines are usually able to answer using just the “hardest clues.” This gives the computers an edge over humans; however, they are not necessarily “thinking” better.

As Jordan Boyd-Graber, associate professor in the University of Maryland Computer Science Department, Institute of Advanced Computer Studies, iSchool, and Language Science Center, points out: 

“This is not just a problem for Quiz Bowl. Nearly every other question-answer data set that is being used in the machine learning and artificial intelligence community has this problem. The questions are being written by humans who don’t necessarily try to challenge a computer.”

Researchers at the University of Maryland have developed an online question-making program to address this problem and invite the public to use the program to try and stump the computer. As a user drafts a question, the computer program highlights the words that the quiz bowl system would use to guess the answer and alerts the user when the system would “buzz in.” The best questions the public can come up with will be put to the test when humans face off against the superior Studio OUSIA quiz bowl system this Saturday at the Human-Computer Expo Match in the Regency Ballroom A/B, Hyatt Regency Reston, 8 PM.