So what's next for IBM's Watson?
If you are waiting for a financial services version of the powerful artificial intelligence system that won a game of Jeopardy against two of the highest winning champions of all time -- Brad Rutter and Ken Jennings -- don't hold your breath ... yet.
Unfortunately for the Wall Street techno-geeks and quants looking for another tool to add to their algorithmic arsenal, IBM isn't working on a financial services version of Watson at this time, according to Dr. David Ferrucci, Lead Researcher and Principal Investigator on IBM's Watson/Jeopardy! Project.
"We have started to adapt Watson for the medical space," said Ferrucci in an exclusive interview with Wall Street & Technology at the SIFMA Financial Services Technology Leaders Forum in New York. "But we are definitely looking at other areas and financial services is one of them." IBM has partnered with Nuance Communications to build an industry-specific version of Watson for the medical field.
In financial services, Ferrucci envisions Watson being able to monitor and analyze a wide variety of content and information quickly to detect opportunities or market risks. "The learning framework can take examples of natural language content," Ferrucci said. "Will there be a merger, or a default event? Watson can look at the surrounding information contained in analyst reports, financial reports, blogs articles and many other sources and could tell when an event might occur.
"Watson can take the evidence and predict that something might happen," Ferrucci added. "It automates the collection and analysis of the data and can present it to humans to make a final decision".
Watson is designed to be relatively easily adapted to new applications and markets, Ferrucci says. "When we designed Watson for Jeopardy!, we wanted to build general capabilities that could be applied elsewhere," Ferrucci explained. "We did not want to ‘over fit' Watson just to answer Jeopardy! questions," which would have made Watson harder to adapt for other applications." For instance, if Watson's algorithm was just designed to answer plain language Jeopardy!-style questions, it would have been almost useless for professionals in healthcare, financial services or other industries, Ferrucci said. "When we were testing Watson, we were running blind data -- not Jeopardy! questions -- to make sure it worked," he adds.
Watson is also designed, Ferricci says, to run without a team of IBM's computer scientists and technicians tweaking the system. "The idea is we would build the capability, and it would grow with the company that is running it," Ferrucci said. "[Watson] does not need the depth of skills we have here at IBM to run in a company.
For instance when IBM build a version of Watson for the healthcare space, IBM set up the application and helped get it running, but it is being run by healthcare professionals who have a lot of knowledge about their industry. "But the person that grows the technology doesn't have to be a computer scientist," Ferrucci said.
Watson, built on IBM Power 750 servers running Linux, does "deep analysis of data," Ferrucci explained. It can process hundreds of information sources, thousands of possible answers and pieces of evidence and test hundreds of thousands of hypothesis almost instantly, says Ferrucci. For a frame of reference, to answer a typical Jeopardy! question on a single CPU would take the Watson algorithm two hours to complete. "That would make for a pretty boring game of Jeopardy!" joked Ferrucci.
Greg MacSweeney is editorial director of InformationWeek Financial Services, whose brands include Wall Street & Technology, Bank Systems & Technology, Advanced Trading, and Insurance & Technology. View Full Bio