IBM is preparing to give third parties access to its Watson supercomputer with the aim of spurring the growth of applications that take advantage of the system’s artificial intelligence capabilities.
Watson, which is derived from IBM’s DeepQA project, drew worldwide attention in 2011 after it soundly defeated human opponents on the Jeopardy! game show.
IBM has been applying Watson’s machine learning—or “cognitive computing”—technology to domains such as health care, but now the company is ready to share Watson with the broader world.
“We’ve been developing, evolving and maturing the technology,” said Rob High, an IBM fellow who serves as CTO of Watson, in an interview. “It’s stable and mature enough to support an ecosystem now. We’ve become convinced there’s something very special here and we shouldn’t be holding it back.”
Watson has “come a long way” since the Jeopardy! competition, High said. IBM decided to focus on health care initially because of the industry’s “particularly challenging” linguistic qualities. “We thought if we could master that, it would open the door for other domains,” he said.
Watson is also now a dramatically smaller piece of hardware. The Jeopardy! implementation involved a 2,900 core system with 15TB of RAM. But now, a basic Watson configuration has been between 16 and 32 cores with 256GB of RAM, according to High. IBM can chain these smaller Watson boxes together as needed for greater scale, he added.
Other improvements include support for additional document types as well as the ability to recognize more elements within those documents, such as embedded tables. Watson can fine-tune the way it answers questions as well, High said.
On Jeopardy! the problems are posed as answers and contestants must reply in the form of a question. “In the real world, people want simple noun-and-phrase answers,” he said.
The Watson cloud
IBM is initially working with a handful of partners on the Watson cloud service, and each is developing specialized applications.
Fluid is creating a Watson-powered program for retail. “The model is to have a running dialogue between the consumer and Watson,” which helps them make more informed buying decisions, High said.
The Watson cloud will include a development toolkit, access to Watson’s API (application programming interface), educational material and an application marketplace. IBM also plans to work with venture capitalists to find startups that want to build software on Watson.
Interacting with Watson is fairly straightforward for any programmers familiar with RESTful APIs, High said. Their real focus should be on understanding the fundamental difference Watson offers compared to past programming platforms.
“Cognitive systems are different in that they have the ability to simulate human behavior,” he said. “For the most part humans have had to adapt to the computer. As we get into cognitive systems we open up the aperture to the computer adapting to the human.”
Some details of the upcoming Watson cloud service, such as those involving pricing, have yet to be finalized.