Archive for February, 2017

What to ask IBM Watson?

February 26, 2017 Leave a comment

On an uncharacteristically cloudy California afternoon looking out at gathering rain clouds I wondered what do IBM Watson – the artificial intelligence platform – and my favorite characters Sherlock Holmes and Dr. Watson have in common?  Unlike Sherlock Holmes who was introduced to the world over 125 years ago by Sr. Arthur Conan Doyle, the world came to know about IBM Watson primarily after the TV show Jeopardy.  Today you read about IBM Watson being used in every possible vertical market – from banking to finance.  Excuse me for using the tiresome cliche “when you have a hammer everything looks like a nail” but to avoid exactly that let’s look at what problems are better suited for IBM Watson over others. Watson works by collecting large amounts of data (articles, blogs, tweets, research data), generating dozens of hypothesis around this data, ranking various candidates for answers and answering with the first candidate if the confidence level is high enough.

Book Illustration Depicting Sherlock Holmes and Dr. Watson in a Train Cabin

Attribution: By Sidney Paget (1860-1908) (Strand Magazine) [Public domain], via Wikimedia Commons

Watson excels in Natural Language Processing (NLP).  An example of a question relevant to our times and posed in natural language would be: Is White House Press Secretary Sean Spicer accurate when he states that Donald Trump drew “the largest audience ever to witness an inauguration, period, both in person and around the globe”

Any question that requires active learning, that uses context based search, or that uses inference chaining would be eminently suited for IBM Watson.  You could also potentially use Watson for predictive analytics, to answer a question relevant to merchandisers “Are stay at home dads more likely to buy beer when they step out to buy diapers for the infants in their care?”   A data scientist who thinks about these matters might explain that Watson has some predictive analytics capability because it uses CHi-squared Automatic Interaction Detection (CHAID) algorithms.  For now we’ll leave the why and focus on the what.

On the other hand, topics like inductive reasoning may be better suited for tools other than Watson.  What is inductive reasoning you ask? The statement “All performers want to perform before large crowds like those at Presidential inaugurations, Elton John is a performer so Elton John must want to perform at Trump’s inauguration” would be an example of deductive reasoning even if Sir Elton John would disagree with the final conclusion.

Conversely the statement: Garth Brooks is a performer, Garth Brooks “prays” for Trump, so all performers must be praying for Trump. In this statement there is no logical movement from the initial premise to the final conclusion.  This would be an example of inductive reasoning.

Just as Dr. Watson wonders how Sherlock Holmes arrives at his seemingly fantastic conclusions you might wonder what internal reasoning is used by IBM Watson to arrive at its conclusions?  The IBM WatsonPaths solution graph might give you some clues.  More on Watson on another rainy afternoon.