"It doesn't just learn from what it knows today," Rhodin adds. "You can add new data to it. It reads new books every day. It connects the dot from what it just read to what it has already read. Sometimes what it just read contradicts what it's already read. It has to sort that out. As we start to move forward, Watson's getting smarter. We're adding new capabilities to it. It's learning to reason, to think through things."
Cognitive computing really comes down to three abilities, according to IBM:
- The ability to perform deep natural language processing and analysis both for information ingestion and research, as well as to provide human-style communication.
- The ability to statistically generate and evaluate series of evidence-based hypotheses to be able to answer questions in a relevant and meaningful manner.
- The ability to adapt and learn from training, interaction with humans and outcomes related to hypotheses it generates.
"These abilities make Watson ideal for applications where large amounts of information need to be ingested and understood, complex decisions are made and feedback is available to train the application to improve its decision making over time," says IDC's David Schubmehl, research director for the research firm's Search, Content Analytics and Discovery research.
"Today, consumers continue to struggle to find and use actionable information," Schubmehl adds. "Traditional search systems deliver web pages, documents, video and audio to users when what they are really looking for is answers and advice."
"The technology behind IBM's Watson takes all of these information sources and distills them down to the important facts, events and relationships. Its natural language capabilities, hypothesis generation, cognitive analytics and machine-learning components then utilize these facts, events and relationships to answer questions in the same manner that a human would. In addition, if it is wrong or incorrect, Watson has feedback facilities built into it so that it can learn and get "smarter" over time," Schubmehl says.
IBM Struggling to Gain a Big Data Foothold With Watson
And yet, for all that, IBM has been struggling to gain traction for Watson in the commercial sphere since the high-profile Jeopardy! win, but response has been tepid. According to The Wall Street Journal, Watson has brought IBM only $100 million in revenue in the three years since its debut.
Part of that may be attributed to what ran under Watson's hood. The Watson that won Jeopardy! was built on 90 IBM Power 750 Express servers powered by 8-core processors--four in each machine for a total of 32 processors per machine. At the time, Power 750 servers were running $34,500 a piece, adding up to about $3 million.
Sign up for Computerworld eNewsletters.