IBM’s new Watson Group, launched to develop and commercialize cloud-based cognitive computing systems, plans to invest $1 billion, including $100 million in venture money to cultivate an ecosystem of applications developed by third parties.
The announcements on January 9 at an event in New York (where the Watson Group will be based) put the system made notable by its 2011 victory on “Jeopardy!” at the center of IBM’s future commercial plans for a world where unstructured data is growing unbounded, and professionals and consumers, all overwhelmed by this information, need help finding answers to questions.
“I think what’s most important now for all of us, why we took on Watson—it’s built for a world of big data,” said CEO Virginia Rometty. Watson, she added, “is not a super search engine. It can find a needle in the haystack, but it understands the haystack.”
IBM research scientists have been working on the project known as Watson since 2006. The system uses natural language processing and machine learning capabilities to sift through large volumes of text and data, with the ability to pick up on the nuances of vocabulary and meaning and context, and then rank potential responses for a user to select.
In that way, a system like Watson has the potential to be a valuable assistant to someone who already knows a lot about a subject but who can’t keep up with accumulating data, or sort through too many information options.
Jose Baselga, physician-in-chief at Memorial Sloan Kettering Cancer Center, said oncologists working with cancer patients are a good example. Doctors can’t keep pace with the torrent of research, the growth in the number of therapies for specific cancers, the rise of genomic techniques and the onslaught of data from patient records—there’s too much for one highly-trained scientist to manage. “The traditional process is not working any longer, and that is where Watson comes into play. It becomes our partner, our colleague” and a resource for integrating a range of data sources, Baselga said.
The Watson Group’s launch comes as the system, still considered very young in its commercial life cycle, is the subject of media reports citing IBM’s slow progress for generating substantial revenue from its sales. Here are five things to know about Watson at this important point in its history.
1. Watson is different from Apple’s Siri or Google Now.
As IBM works to bring Watson into the mainstream, it’s useful to recall the buzz surrounding Siri’s debut, as the recorded voice responded to iPhone users’ questions about schedules, weather and other relevant facts. The version of Watson that won “Jeopardy!” was about sifting through data to sort out trivia questions and answers, and in that way it shares a theme with applications like Siri or the Google Now app.
But the comparison quickly breaks down, IBM executives say, pointing to the system’s subsequent progress in terms of the data it is scanning for information and the serial questions it is designed to answer. Watson identifies options for cancer treatment in hospital settings, as Baselga discussed. Other applications under development, if successful, will help arrange travel plans and act as a personalized shopping assistant based on criteria presented by end-users.
This points to another difference: scalability. The “Jeopardy!” version of Watson was designed for one user on one question at a time, while subsequent instances serve many, and IBM executives envision systems that serve multitudes more.
2. Watson learns from interactions with subject matter experts—and the process takes time.
IBM believes that Watson ushers in a third era of cognitive computing, after earlier eras of tabulating (that began in the mid-19th century) and programming (rules-based systems begun with mainframes and still dominant). In the Watson environment, researchers consciously decided to avoid rules (if X, then Y) to mimic the way people formulate and express thoughts, said Rob High, IBM Watson Solutions CTO at November’s Information on Demand event in Las Vegas. Watson uses natural language processing and machine learning technologies to make inferences and correlations about the content it ingests.
Then the training begins. Stephen Gold, IBM Watson Solutions vice president, said IBM provides a set of tools, including tools to ingest data into Watson; training tools to help customers “advance Watson’s understanding of the content” loaded into it; integration tools, to help fit the technology into existing business processes, and to develop user interfaces for it.
The training can involve a set of question and answer trials. What is the range of conditions or outcomes that match the set of circumstances described? Watson provides a list of ranked responses—and experts score them, creating a feedback cycle designed to improve subsequent responses.
Researchers at University of Texas MD Anderson Cancer Center worked with IBM for a year to develop a prototype of its Oncology Expert Advisor powered by Watson. A prototype was made public in November, and a release date for the system has not been announced.
3. Watson serves up probabilities and evidence, rather than definitive responses.
IBM describes cognitive systems as probabilistic, and Watson generates responses to questions scored according to levels of confidence. (Those confidence levels are subject to change when subject matter experts grade the responses, as noted above.) IBM officials said Watson also has the capability to show the evidence for its responses—what data there is to back up the answer and the confidence score.
That is an important component, said IDC analyst Henry Morris at a November IBM panel discussion on Watson. Morris noted that “people don’t do a good job of dealing with probabilities” and making decisions based on probability and risk, as shown by researchers such as the economist Daniel Kahneman.
One of the things that is useful about Watson is “it shows a transparent way of looking at a solution you might not have thought of, and then understanding what the probability of success was so that you can feel more comfortable in dealing with that,” Morris said. “The ability to be able to look at all the evidence, surface some hypotheses that maybe the human agent, the decision agent, may not have thought of, giving the probabilities of success of a particular hypothesis of working. And then giving that to the human agent. What this does is, it changes the way human decision makers can work by having this type of assist.”
That approach is different from the rules-based systems determining many actions in business, such as ERP systems, Morris said, where managers must confront exceptions that are not covered in the existing polices.
4. For now, Watson works with text in English.
Watson ingests and processes text documents in English. The size of the documents has increased from about three sentences for the “Jeopardy!” system to 20-page medical reports, Manoj Saxena, IBM Watson general manager, noted in November. But the technologies that enable the system to understand a question about “2 and 2” points to four as the answer, but also may mean the seating orientation in a typical sedan or the orientation of a two-parent, two-child nuclear family, are English-only now.
Researchers have begun to work on other languages, Gold said in an interview, but IBM isn’t saying which ones.
What about still images, audio, video? At the January 9 event, Guruduth Banavar, vice president of cognitive computing at IBM Research, said scientists are working to bring video, audio and animations into Watson.
5. The cloud-based ecosystem of application developers will provide signals of Watson’s strength.
Since the TV game show victory the Watson team has been working with external experts on use cases, first with cancer researchers in health care, then moving to financial services and an application tailored to the customer service function. Each instance has involved working in tandem with end-user organizations to install Watson and teach it domain expertise. (On January 9, IBM announced a new Watson customer, DBS Bank of Singapore, for a wealth management system.)
In November, IBM unveiled a program to invite third-party developers to partner with Watson on developing new applications and use cases. The new ecosystem includes a “Watson developer cloud” that provides APIs, tools and methodologies for developers to work with a Watson system; a content store that provides data—both free and fee-based content—to feed into new applications; and a “talent hub” of about 500 subject matter experts, from IBM and third parties, to provide support on machine learning and natural language processing technologies, user experience design and other elements of building an application.
Developers need to apply for the program and have to pay for compute charges and agree to share revenue from applications created with Watson.
“We need entrepreneurs with bright ideas [who] can imagine the future that we can’t see,” said Mike Rhodin, senior vice president of the IBM Watson Group, at the New York event. Rhodin said the ecosystem has about 890 companies signed up so far.
The ecosystem partners publicized so far come from health care (MD Buyline, a procurement advisor for hospital managers and Welltok, a consumer health app), retail (Fluid Retail, which is developing a personalized shopping assistant) and travel.
Terry Jones, the founder of Travelocity.com and Kayak.com, said that travel—the biggest ecommerce category—is poised for a disruption like Watson. The average consumer visits more than 20 websites when trying to plan a leisure trip. There is an abundance of information, including text and video. Prices change quickly. Reviews for destinations and services are readily available. What’s missing, he said, is advice. (He said he uses a travel agent for leisure trips.)
In a test, Jones said Watson looked at 64 million reviews and 15 million articles to suggest Bali (“97 percent confidence” for the answer) when asked where a family with children should go for an “adventurous vacation.” A user can react to the system and provide feedback, to refine the answer to stay inland, for example. “Because it’s about a conversation, I can refine the answer,” Jones said. “If we do this, we can move travel outside the box.”
Michael Goldberg is the editor of Data Informed. Email him at Michael.Goldberg@wispubs.com.
Home page photo of headquarters of IBM Watson Group via IBM.