IBM Impact: IBM closes doors on new Watson pilot projects

Selling the AI technology is not the goal at present

LAS VEGAS: Any firms interested in trying out IBM's Watson technology should prepare for disappointment, as IBM is not accepting new test projects for its artificial intelligence machine.

Robert Madey, vice president for IBM Watson Solutions, said that he is inundated with requests for the technology, with 10 proposals a day passing his desk with new ideas on how to use Watson.

"If you come to me right now and ask, ‘How many millions [for Watson], I'd say sorry, no can do. I've got all the pilots I need. The big mission right now isn't selling Watson, it's proving Watson can work in a commercial environment," he explained.

"Eighty or 90 per cent of these requests don't need Watson anyway, technology already exists for what they need."

Madey was presenting at the IBM Impact conference in Las Vegas on the Jeopardy-beater's progress in the commercial space. Healthcare and financial services have been early test environments, with WellPoint and Citibank both running Watson projects.

"Ninety per cent of the world's data was created in the last two years, and businesses, governments and healthcare providers can't put it into use, make sense of it or put it into their processes," Madey said.

To train Watson for the healthcare sector, the first step was to ask nurses to sit down and write over 2,000 questions to ask the machine, with the objective being that it could tell a doctor whether a treatment being considered was authorised or the norm for a patient with those particular symptoms.

"Medical information is doubling every five years, and most of it is unstructured. We want to make this data available to specialists, that's Watson's mission for the next year in healthcare," Madey said.

IBM Impact: IBM closes doors on new Watson pilot projects

Selling the AI technology is not the goal at present

Watson is achieving this outcome by building up its diagnosis confidence levels via learning tasks.

It starts by asking a patient about particular symptoms, for example abdominal or back pain, and then moves on to analyse family history, reaching a higher confidence level.

The next stage would be finding out about patient history, such as medication being taken and any recent test results. At that point, Watson would have a very high confidence level about the diagnosis.

The next step in healthcare is a partnership with genetics specialist MD Anderson, aimed at improving the treatment of cancer.

At Citibank, the goal is to reduce customer attrition. Watson is being trained to sift through the vast amounts of customer data, much of it unstructured, to supply call centre workers with information that would help them retain or upsell card services. This all needs to be done in real time, while the agent has the customer on the phone.

Madey also gave insight into the challenges his team faced on transitioning Watson from game contestant to business provider.

"We had to take it from answering one question at one time to handling thousands of documents and answering thousands of questions that go back and forth. And the code you get from researchers and scientists isn't exactly what you want to put into production," he noted.

"It took seven weeks to retrain Watson for each Jeopardy section. If you go to a doctor and think you might have cancer, do you want to wait seven weeks for an answer? Probably not."

Madey was also keen to pre-empt any comparisons between Watson and Apple's Siri.

"People ask ‘Can't Siri do this?' You've got to be kidding me?" he quipped.

"There's a big difference between voice recognition and natural learning technology."