IBM continues its push to convert the original Watson, a standalone supercomputer that gained fame as winner of...
a Jeopardy exhibition match, into the role of a more general-purpose, cloud-based offering. As part of that effort, the company last month released Watson Analytics, an application that places Watson technology in front of more business users.
"Watson Analytics is like a data scientist in a box, but it is comfortable for a business person," said Marc Altshuller, vice president of IBM Watson Analytics. "It has cognitive discovery tools, and you can use natural language to ask a question of the system."
The software runs on IBM's SoftLayer cloud, and is available through the IBM Cloud marketplace, he noted.
This is not the full-featured, deep-learning Watson often discussed -- it is instead a collection of some Watson elements focused on analytical discovery, with other established IBM analytics tools taking over after a problem space is narrowed down or discovered.
Also underlying the analytics services are several data refinery tools. Data refinery could be described as IBM's take on the "data lake," which has been much discussed in Hadoop circles.
IBM has had a big hand in promoting the concept of cognitive computing to describe Watson's objectives. Cognitive computing is a broad amalgam of such diverse elements as natural language processing, dialog and inference engines and machine learning. The latter part often gets special attention, as it underlies recent advances in personal digital assistants and recommendation engines.
The company positions cognitive computing as a crucial capability as bigger and bigger volumes of data start to overwhelm human decision makers.
Since its Jeopardy victory, IBM has brought out versions of Watson for medical diagnostics, call centers and even recipe selection. Just last week, it announced results of a scholastic competition won by students at the University of Texas at Austin, who created a mobile app to handle natural language questions about United Way social services in the state.
Because the full-fledged Watson is still under development, appearing in snippets -- and because Watson seems like a bet-the-company proposition -- programs like the scholastic competition and point-releases like Watson Analytics are closely watched.
Because cognitive computing includes a host of technologies, including some that go back to AI in the 1980s, IBM has had its hands full just describing Watson.
"The key point is that cognitive computing is blending with machine learning, AI and ambient intelligence environments that are especially responsive to humans," said Steve Ardire, an independent advisor to software companies.
Cognitive computing also includes emotional intelligence, virtual personal software agents and more, he said.
"Watson and cognitive computing are a greenfield opportunity for IBM to reinvent itself," Ardire said. But there is risk. "It is a huge bet," he said, and others are in the hunt.
"Watson is getting the most attention because IBM has the most marketing dollars. Watson Analytics is an impressive app, but there are other ways to build impressive cognitive analytics apps," Ardire said. He mentioned startup MetaMind, baed in Palo Alto, Calif., which is pursuing software services for text sentiment analysis, object recognition and other AI tasks.
IBM may not enjoy "a market of one" for long. Microsoft's Azure Machine Learning software, IPSoft's Amelia cognitive knowledge agent and anticipated offerings from two cloud heavyweights -- Google and Amazon -- could vie with Watson.
Mike Rhodin, senior vice president of IBM Watson, was not shy in admitting some challenges to commercializing Watson. At last fall's EmTech MIT 2014 conference, he said having continuous dialogues is difficult.
IBM is working on a variety of challenges to successful Watson implementation; some of the challenges trace back to the first days of AI and expert systems. Summarization and reasoning engines have been added to the original Watson blueprint, Rhodin said, so that it can serve more varied uses.
How many cogs in cognitive?
According to a leader of the University of Texas team that won the recent Watson competition, the latest APIs and cloud-based version of Watson are not difficult to work with. The evolution of Watson has been rapid, to hear Bri Connelly speak of it.
"It is not a question of which Watson you are working with; it is a question of what Watson we built," said Connelly, team leader and undergraduate computer science student at the University of Texas at Austin.
At one point Watson was "a bunch of computers in a room" -- that was the Jeopardy Watson," she said.
"Now, it is a service. It is really just a collection of algorithms in [the IBM] cloud that we interact with," said Connelly.
She described the implementation, work on which was begun last September, as being at the prototype stage, as it still needs user testing. Connelly said training Watson, or educating the system, has been straightforward so far.
"It worked fairly well off the bat. But we are not done training it. It could do better. There may be better answers it can give, or answers it can give more succinctly," she said. "Shorter answers are what we are after."
As with IBM's Watson itself, the University of Texas implementation is still a work in progress: Both the programmers and the systems learn as they go. Meanwhile, how quickly all parties learn will be watched by people interested in computer progress generally, and IBM watchers specifically.
Learn about AI issues that could affect Watson's outlook
Find out more about advanced and predictive analtyics
Listen to a podcast about cognitive computing