Watson Jeopardy! refers to the 2011 competition in which IBM's Watson computer system, developed under the DeepQA project, played against two of Jeopardy!'s most successful champions, Brad Rutter and Ken Jennings. Watson won the two-game match, taking the first-place prize of US$1 million, and demonstrated significant advances in artificial intelligence and natural language processing. The event was broadcast in February 2011 and marked a milestone in computing's ability to understand and answer natural language questions.
Watson was named after IBM's founder and first CEO, Thomas J. Watson, and was built by a research team led by principal investigator David Ferrucci. The system was designed to answer open-domain questions by combining multiple AI techniques, including information retrieval, knowledge representation, automated reasoning, and machine learning. Unlike later deep learning systems, Watson's architecture relied on hundreds of parallel algorithms rather than deep neural networks.
System Architecture
Watson's software was built on IBM's DeepQA framework and the Apache UIMA (Unstructured Information Management Architecture). It was written in Java, C++, and Prolog, and ran on SUSE Linux Enterprise Server 11 with Apache Hadoop for distributed computing. The system included strategy modules: one calculated Final Jeopardy wagers based on confidence scores and opponent scores; another used Bayes' rule to estimate the probability of a Daily Double using historical data from the J! Archive. If a Daily Double was found, the wager was computed by a two-layered neural network similar to TD-Gammon, developed by Gerald Tesauro in the 1990s. Parameters were tuned by benchmarking against a statistical model of human contestants.
Hardware comprised a cluster of ninety IBM Power 750 servers, each with a 3.5 GHz POWER7 eight-core processor with four threads per core, totaling 2,880 threads and 16 terabytes of RAM. The system processed 500 gigabytes per second (equivalent to a million books) and had a Linpack performance of 80 TeraFLOPs, about half the cutoff for the Top 500 Supercomputers list. All content was stored in RAM during the game because hard drive access would be too slow.
Data and Knowledge Sources
Watson's knowledge base included encyclopedias, dictionaries, thesauri, newswire articles, and literary works, as well as structured databases and ontologies such as DBpedia, WordNet, and YAGO. IBM provided millions of documents for Watson to build its knowledge. The system parsed questions into keywords and sentence fragments, then executed hundreds of language analysis algorithms simultaneously. The more algorithms that independently converged on the same answer, the higher the confidence. Watson then checked potential solutions against its database to verify plausibility.
Comparison with Human Players
Watson had both advantages and disadvantages compared to human contestants. It lacked contextual understanding, and humans often generated responses faster, especially for short clues. Watson was programmed not to buzz until it was confident, but it had faster reaction times once a response was ready and was immune to psychological tactics like category jumping. In 20 mock games, humans used the six to seven seconds Watson needed to process a clue to decide whether to signal. Watson's electronic circuitry received the ready signal and evaluated confidence before buzzing, giving it a speed advantage over human reaction times except when humans anticipated the signal. Watson's voice was synthesized from recordings by actor Jeff Woodman for an IBM text-to-speech program.
Aftermath and Commercial Applications
In February 2013, IBM announced Watson's first commercial application: utilization management decisions for lung cancer treatment at Memorial Sloan Kettering Cancer Center, in conjunction with WellPoint (now Elevance Health). However, the Watson Health division struggled commercially. In 2022, IBM divested Watson Health, spinning it off into Merative, which was sold to Francisco Partners for $1 billion, after IBM had invested $4 billion in development. By 2023, Watson had cost IBM 10% of its stock value, lost four times more than it brought in, and led to mass layoffs. Despite these setbacks, the Jeopardy! victory remains a landmark in AI history, influencing subsequent research in machine learning and large language models.