Watson may not know it all, but it sure taught IBM a few things

IBM’s Watson natural language query/cognitive computing prodigy was a huge PR coup for Big Blue. Three years ago, Watson defeated Jeopardy champ Ken Jennings on national TV and beat other challengers like a drum on a subsequent victory tour. (Ask Gigaom’s own Stacey Higginbotham about that sometime.) IBM rode that wave for years to show that despite its woes, it can still do really hard stuff. IBM wants Watson to be a $10 billion business by 2023.

But, unfortunately for [company]IBM[/company], there is “not a lot of commercial application to playing Jeopardy,” Mike Rhodin, [company]IBM[/company] SVP for Watson, acknowledged at Emtech 2014 at MIT on Tuesday.

Mike Rhodin, SVP IBM Watson.
Mike Rhodin, SVP, IBM Watson

IBM invested untold millions in Watson, so it’s now time for Watson to, in the tortured words of another Emtech presenter, become “a market-based solution.” It needs to earn its keep. IBM has been moving down that road, making Watson more accessible to outside developers via APIs and launching a new Watson division early this year. And it’s pushing Watson into healthcare and financial services applications. Last week it started offering Watson analytics as a freemium cloud-based service.

But at the most basic level, here’s what Watson — or rather, the Watson experience — has taught IBM, based on Rhodin’s Emtech talk.

  • While the “deep Q&A problem” is real and very hard — the goal is for a computer to be able to answer any question — that is only one piece of the puzzle.
  • Watson is more likely to answer correctly if it knows about the question, not just the question itself. As in life generally, context is key and that context needs to carry over from conversation to conversation. USAA’s work with Watson helped on this, Rhodin said. For military veterans asking questions about benefits, etc., Watson really needs to know where the questioner lives and what her medical history is — which, of course, raises privacy issues.
  • Watson needs to know when not to answer the question — when it does not have enough information to answer at all.
  • Training never ends. The fact that Watson needs to be trained is one gating factor to broader adoption. This is no turn-key solution, as early “bleeding edge customers” have learned, Rhodin said.
  • A question is fine, but a conversation is often better. Watson needs to engage in a give-and-take with the questioner to get to what is really being asked. IBM bought Cognea to help with that.
  • Parsing the question: “We’re not always posed questions. Sometimes people hand us problems which have to be broken into questions,” Rhodin said. IBM’s work with the Cleveland Clinic has led to the creation of a “reasoning engine” that will help with that.

Watson is a work in progress. Rhodin says all this work around cognitive computing, including natural language processing, is reminiscent of what happened in the world of computer operating systems way back in the 1960s.

Jeopardy was a great start, so we have a Q&A engine, but we also are building a summarization system and a reasoning engine that are all part of a broader platform,” Rhodin said.

Fifty years ago, generally programmable systems became popular, but they were augmented over time with new subsystems — things like file systems and database systems, which were “added layer by layer to fill out the platform that eventually automated the world’s financial systems [and] travel industry,” Rhodin said. “We are seeing the beginnings of that system that will emerge in the next 50 years.”

I have no quibble with that. But with other companies — [company]Google[/company], [company]Apple[/company], [company]Microsoft[/company] and many we probably don’t know about yet — also attacking these big problems, it’s not a forgone conclusion that IBM will be the leader.

It knows what it has to do — build or attract applications that take advantage of Watson’s evolving power. Now it has to execute. To quote Gigaom’s Derrick Harris writing about IBM’s decision a year ago to open up Watson APIs to programmers:

If IBM actually delivers a workable cloud platform around Watson and developers actually take advantage of it to build new, smart applications, it will be a big fricking deal.