IBM’s Watson goes from quiz champ to medical marvel
In 2011, humanity seemed to lose yet another inch of its footing as the dominant species on the planet. The reason? IBM Watson, a computer system that answers natural language queries, defeated the world’s best human players on the popular quiz show “Jeopardy!” It was yet another step in the computer’s inexorable march toward human-like intelligence – and followed the momentous victory by IBM supercomputer Deep Blue over chess grandmaster Garry Kasparov in 1997.
IBM has taken on such “grand challenges” every 10 to 15 years as a way to accelerate innovation, Koichi Takeda of IBM Research Tokyo explained to attendees at the 2016 Rakuten Technology Conference in Tokyo, where he took home the Gold Prize for his work on natural language processing and text analytics. Takeda, a leading member of the Watson project – thrilled the conference audience with an overview of the project’s design goals, technologies and potential applications.
“We wanted to have a ‘wow’ factor so that everybody – non-technical people included – could understand the importance of the technology,” Takeda said about the Watson development team. “We wanted to have a scientific impact as well as a business impact.”
Watson is an example of cognitive computing, which refers to computer systems that in some way emulate how the human mind works and respond to stimuli. The Watson project consisted of 10 teams around the world including 30 researchers at the IBM Thomas J. Watson Research Center in Yorktown Heights, New York, where the “Jeopardy!” showdown was held. The system was “trained” using 25,000 past questions from the program, and programmed to recognize keywords, associate them with concepts, formulate possible answers and choose the one with the highest probability of being correct. Running on 90 IBM Power 7501 servers, the entire process takes only a few seconds – Watson ran at about 80 teraflops, or 80 trillion floating point operations per second, a measure of a computer’s speed.
The computer system famously took home the top prize of $1 million in the quiz show tournament, but hasn’t rested on its virtual laurels over the past five years. One of the most compelling applications of Watson is assisting with medical diagnoses and treatment. While published medical information is doubling every five years, and one in every five diagnoses is estimated to be incomplete or wrong, Watson can mine enormous volumes of data and quickly analyze a set of symptoms and a patient’s history. Just as the system was trained on previous “Jeopardy!” answers, Watson was trained on cancer cases at Memorial Sloan Kettering Cancer Center in New York to be able to recommend customized courses of treatment based on supporting evidence from numerous sources.
Watson’s ability to infer from medical cases was highlighted in dramatic form earlier this year at the University of Tokyo, when it identified a 60-year-old woman’s rare form of leukemia following an incorrect diagnosis months earlier. It cross-checked her genetic data against a database of millions of oncology papers, taking only 10 minutes to identify which of her many genetic mutations were significant – something that would have taken human scientists about two weeks. After the patient’s therapy plan was changed, the patient improved and was eventually discharged from the hospital.
“In cognitive computing, you can combine hardware and software technology to build an intelligent system with a lot of deep learning, pattern recognition and inferencing capabilities,” said Takeda. “Now is a very interesting time to come up with innovative systems inspired by human behavior.”
Read more posts from the Rakuten Technology Conference 2016 here.
Photographs of IBM Watson courtesy of International Business Machines Corporation. Unauthorized use not permitted.
