:format(webp)/cdn.vox-cdn.com/uploads/chorus_image/image/63703866/ibm_watson.0.1467738535.0.jpg)
IBM plans to jumpstart Watson with a $100 million venture fund to build a business around the supercomputer that famously beat two human champions at the TV game show “Jeopardy!” in 2011.
The fund to boost startups building apps for Watson is part of a $1 billion investment in Watson that IBM CEO Ginni Rometty unveiled today in New York. The aim is to transform the machine designed to answer natural language questions into a real business. IBM will devote 2,000 people to the effort in the new business unit.
Since Big Blue put the machine on national television in an elaborate publicity stunt in 2011, IBM has created a series of new jobs for Watson: Analyzing and improving customer service; cancer researcher; and a physician’s assistant.
Yet there are signs of trouble in the world of Watson. Earlier this week, The Wall Street Journal, citing leaked transcripts of internal conference calls, reported that sales of the machine are failing to meet expectations. On the call, a marquee Watson customer, University of Texas MD Anderson Cancer Center, was described by one IBM exec as being “in a ditch.” With only $100 million worth of Watson sales recorded so far, an internal goal of hitting $1 billion by 2018 seems pretty far-fetched.
IBM announced two new Watson-based products today. One is the Watson Discovery Advisor, described as being designed to cut through data quickly and find unexpected connections in raw data. It’s being aimed at the pharmaceuticals industry.
The second new product is the IBM Watson Analytics Advisor, which is supposed to take natural language questions and run them against complex sets of business data. It’s potentially useful in part because data questions are more often than not the specialized domain of data scientists trained in the interpretation of business data that can then be turned into decisions.
One big challenge to turning Watson into a business is that mainstream businesses simply don’t have the time to train Watson to become useful. It takes a lot of time to index all the metadata it will later have to sort through in order to solve problems. In being prepped for the cancer research product, Watson was fed more than half a million medical evidence documents in what IBM called at the time a “training period.”
The point is that when presented with questions, Watson’s job is to provide weighted answers — that is, answers that are graded according to the amount of confidence the machine has in the options. That’s probably more complex than most companies need, and the time and expense to get up and running is probably turning a lot of companies off.
I reached out to Steve Baker, a former colleague and the author of “Final Jeopardy,” a book on Watson and its game show project. He told me that Watson is actually a combination of four technologies: Information retrieval, natural language, question-answering and analytics. IBM already sells systems that use these technologies on a standalone basis. It’s more complicated to work with all four rolled into one.
“So if IBM has a goal to reach $10 billion in sales for Watson, it can just brand those revenue streams as Watson, since so many of them feature at least pieces of the package,” he said.
Baker says the fundamental change that Watson represents — computers that reliably can answer questions that people want to ask — is coming. Whether or not IBM is the one that delivers the platform that we all end up using is another matter.
“It has to do with building machines that can handle language, knowledge and nuance. They’ll change our civilization, they’re already busy doing it,” he told me. “The fact that IBM might have trouble turning this into revenue this year or next doesn’t make much difference.”
This article originally appeared on Recode.net.
