This article in AI Magazine, by David Ferrucci and a large IBM Research team, is the primary technical account of DeepQA, the architecture behind Watson. The paper frames the project as an explicit challenge: build a computer system that could compete at the human champion level, in real time, on the American TV quiz show Jeopardy. The authors stress that this meant fielding an actual real-time contestant on the show, not just running a laboratory experiment, and they report that after about three years of work by a core team of roughly twenty researchers, Watson was performing at human-expert levels in precision, confidence, and speed.
DeepQA’s design is the paper’s lasting contribution. Rather than a single algorithm, it is a massively parallel pipeline that takes a question, generates many candidate answers from many different sources and techniques, and then scores and combines evidence for each candidate, producing not just an answer but a confidence estimate. That confidence value was essential for Jeopardy, where a contestant must decide whether to buzz in and how much to wager. The approach deliberately blended many weak methods rather than betting on one strong one.
The work sat at the intersection of natural language processing, information retrieval, knowledge representation, and machine learning, and it became a landmark demonstration that those fields could combine into a system that handled the puns, wordplay, and broad trivia of open-domain questions.
For a business reader, DeepQA is the engineering story behind one of AI’s most famous public moments, and the template IBM tried, with mixed success, to carry from a game show into industries like health care.