IBM Watson AI gets draw on Jeopardy!
IBM's artifical intelligence draws with human player in first round of quiz show competition
IBM's Watson artificial intelligence managed a draw in the first round of its million-dollar stint on US game show Jeopardy!
The DeepQA technology machine, developed by IBM specifically for the quiz show, was equal with one of the human contestants with a total of $5,000 each in prize money by the end of the show, with the second human contestant on $3,000.
The Watson machine, named after IBM founder Tom Watson, is competing against the two best all time contestants on the quiz show over three consecutive nights, for a million dollar prize.
In terms of performance, Watson proved to be very good at answering the more straight forward questions, but did not always get some of the more subtle questions, and because it does not have an audio input, taking questions from the board as text, it repeated an incorrect answer to a question that one of the competitors had already gotten wrong.
Watson took four years to develop, and is intended to take artificial intelligence to new levels of complexity. The IBM team chose the Jeopardy! game show as its field of competition, because the show's question format includes clues in the question, which means the QA engine has to analyze subtle meaning, irony, slang, riddles, and other complexities to arrive at an answer, a highly complicated task for a machine. Jeopardy! also punishes contestants for wrong answers, so Watson has to rate confidence in the accuracy of the answers before buzzing to answer.
Watson is housed on 90 IBM Power 7 servers in ten racks, with 16TB of memory and 2,880 processor cores, The system runs on Linux, and uses a number of proprietary technologies for the specialized demands of processing a huge number of concurrent tasks and data while analyzing information in real time.