Watson knows the definition of “brake” –
the device that stops something; or the act of stopping something. But what
about the hundreds of ways that the car-driving public describes problems with
their brakes to a mechanic? The brakes might “squeal,” or “judder.” Or maybe
they’re “soft.” All of those descriptors could indicate a different issue that
aren’t understood by a computer. Until now. Thanks to IBM Research’s Dan Gruhl,
Watson can help the mechanic and the car manufacturer pinpoint the problem,
based on common, even slang descriptions.
Watson Concept Expansion, on demonstration
at the Cognitive Computing Colloquium in San Francisco, expanded the cognitive
system’s dictionary of related words, concepts, euphemisms, and colloquialisms
to better-understand context across any industry or field. Dan’s team feeds
Watson unstructured data from blogs and news articles, to Twitter’s firehose,
among other sources.
Then, because this is a human-machine partnership, meant
to augment one another’s intelligence, the human expert begins to tell Watson
what is relevant (squeaky brakes), and not (I need a break). While Watson
gathers and organizes by relevance online documents and discussions about how
to fix various kinds of brake problems.
“It's a game of patterns. If I start with
'apple' and 'blackberry' in the tool, it needs to interact with me a few times
before it realizes I am talking about fruits and not cell phone manufacturers.”
|Dan Gruhl at 2015 Cognitive Colloquim|
That context from such a wide and diverse
source that only Watson could gather and comprehend (in seconds) gives domain
experts a new way to understand their industry, brand, or product. Try it out
on Bluemix, now.
Labels: bluemix, cognitive computing, cognitive era, IBM Watson, machine learning