Last week on the special episode of Jeopardy, we witnessed a transformational milestone in the history of computing. Facing off against super champions (and humans) Ken Jennings and Brad Rutter was an LCD panel in the center podium nicknamed Watson. Behind Watson was four years of work by IBM scientists resulting in a 2,800 core, 15 terabyte memory supercomputer designed to compete in a field we’ve believed is something only humans can do: natural language processing.
httpv://www.youtube.com/watch?v=lI-M7O_bRNg
As the contest began, the machine was holding its own. Then the brilliance of its algorithms added up into dollars, and at a pivotal moment – shall we say perhaps a little bit of machine luck – Watson picked the Double Jeopardy question. The final question on 19th century novelists sealed the deal. Classy Ken Jennings said later “we saw something new here.” Show host Alex Trebeck rightly pondered “where do we go from here?”
Is this a big deal? Is it really a contest that a machine the size of ten refrigerators is competing against a 3-lb. human brain? How is this better than Google returning 3.6 million results back in 0.14 seconds?
To understand, realize the fundamental difference about this challenge: Watson was not returning a response from a structured database. Through deep analytics, it was analyzing patterns, generating hypotheses, gathering evidence, and filtering, merging and scoring the results to pick the one response with the highest confidence. All within fractions of a second.
Technically described as “massively parallel probabilistic evidence-based architecture” by program lead Dr. David Ferrucci, Watson was quite simply trying to understand the meaning behind the question before answering it. It was not pulling a response from a category, but sorting out the uncertainty and ambiguity of an open-ended query. That does have its limits. Watson’s answer to a question on U.S. cities was … Toronto???? (But yes, there are half a dozen cities in U.S. named Toronto!)
So for analytics professionals, what are the lessons for sales and business intelligence?
First, filter and reduce the responses through contextual understanding. Then deliver the ones with highest confidence, simplifying the actions of the users. Provide what you determine to be the most important and actionable information.
Second, sales thinking is intuitive, unstructured and fast-reacting, much like a Jeopardy session. When we try to help a sales rep or manager, describe what the insights mean to them, what story it is telling them. The response provided should flow with the way sales people think.
Is all this futuristic? As Dr. Ferrucci says, “this doesn’t mean we have conquered natural language processing, it’s only the beginning.” Certainly we can apply these learnings today. On the report you’re about to deliver next week, think some more behind why it’s needed, who is using it, what they will do with it and how it will enrich their life. That’s just common sense and empathy we humans are already gifted with.