IBM’s Watson supercomputer reached a milestone in artificial intelligence last February when it beat two Jeopardy! champions. Millions watched, and while some experts dismissed it as a publicity stunt, IBM said Watson would soon be helping doctors diagnose illness, and hinted at talks with gadget companies about Watson helping consumers with questions.
As IBM prepares to celebrate the first anniversary of the televised contest on February 16, though, it is not yet offering the question-answering system for sale. Although limited trials using Watson technology are underway in health and financial services businesses, the AI prodigy is having its biggest impact by pulling in new customers for existing business products—as IBM persuades them to organize their data into formats that an AI like Watson can better understand. IBM has created a slogan, “Ready for Watson,” to help sell its products that way.
IBM hasn’t disclosed how much it spent developing Watson, but the lengthy research and development process is believed to have cost in the tens of millions of dollars. To play Jeopardy, the system needed to understand the meaning of the answers posed as clues, and to rapidly apply general knowledge—distilled from the Internet and other sources—to identify possible answers. That required novel software and an expensive supercomputer.
“Customers are coming to us and saying ‘I’d like a Watson,’ ” says Stephen Gold, IBM’s director of worldwide marketing for Watson. Eventually, that might be possible, but first they need to have the right data sets for Watson to operate on. Watson acquires knowledge by digesting piles of text data, and many businesses simply don’t have it on hand, or don’t have it organized in the right way. Alternatively, a company may not currently operate in a way that would make a question-answering computer very useful. Instead, IBM can offer its more established products and services, such as more basic data storage, processing, and business analytics. These tools can help uncover hidden trends in a company’s data, but lack Watson’s unprecedented ability to answer questions as a human would rather than just spitting out numerical analyses or results.
It’s a strategy that skeptics may say vindicates the view of Watson as little more than a marketing stunt. But Manoj Saxena, who, as general manager of IBM Watson Solutions, leads efforts to commercialize Watson, says ” ‘Ready for Watson’ is an on-ramp that allows a business to start with the subsystems [of Watson] and get value as they build up to a full Watson.” That helps sell business analytics software in particular, an area of great importance to IBM, which has spent $14 billion buying up analytics companies in the last five years.
Seton Healthcare, which operates hospitals and clinics serving 1.8 million people in central Texas, recently signed up for a system branded “Ready for Watson” that tries to predict if heart surgery patients will need to return to the hospital. Natural language processing software reads through medical records to look for relationships between factors like lifestyle, medications, and even details like whether a person previously missed appointments. Although Saxena says the system uses the same language processing ability as Watson, IBM offered such software before Watson came along.
At the heart of Watson is a system known as DeepQA. This is not yet for sale, although trials are underway in both health and financial services. DeepQA is the element that allows Watson to answer a question it has never seen before, drawing on what it has learned from a mass of other text that doesn’t directly answer that same question.
“As we speak, Watson is helping nurses to deal with insurance approvals for Wellpoint,” says Saxena. Wellpoint, which insures one in nine Americans, uses its version of Watson to examine patient records and narrow down possible diagnoses. The system draws on Wellpoint’s own guidelines, medical research papers, news reports, and what it has learned from past records and diagnoses. If a patient called in reporting flu-like symptoms, for example, the system might suggest that he actually suffers from allergies, based on medical literature saying that allergies produce similar symptoms, and local news reports of high pollen counts. Nurses reviewing the upcoming appointments are shown a list of five possible diagnoses, similar to the three possible answers Watson generated on Jeopardy.
Another Watson system to help doctors plan cancer treatments, which can be extremely complex, is also in the works, says Saxena. In the next few months, IBM will announce an implementation of Watson in finance, too. “If you’re a bond trader, [foreign exchange] trader, then you have to deal with a lot of information. Bond prospectuses are usually around 60 pages long.”
Hemant Bhargava, a professor of management and computer science at the University of California, Davis, researches the strategy of how new technologies are introduced. “There are many levels at which IBM can sell stuff, and it makes sense to use them all,” he says. Analytics is already big business today, but it is growing rapidly as new sectors and businesses adopt it, and Watson can help draw that new business IBM’s way, he says.
Making a success out of software that works the way Watson did on TV will be a longer process, says Bhargava. “None of those will be a slam dunk,” says Bhargava. “There may be complex liability issues around the advice a system gives, and IBM will have to show for certain that it is cheaper and better.”
One obvious area where Watson-like technology could be helpful, but that IBM hasn’t said much about, is social media, Bhargava points out. Investor interest in Facebook, which filed to go public last week, is based on the assumption that the company will find new ways to extract value from the activity of its many users, such as better targeted ads. “There’s so much text out there and a need for smarter analysis,” says Bhargava. “There may be many applications for Watson-like technology.”
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
Data analytics reveal real business value
Sophisticated analytics tools mine insights from data, optimizing operational processes across the enterprise.
The Biggest Questions: What is death?
New neuroscience is challenging our understanding of the dying process—bringing opportunities for the living.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.