Each domain is really a dialect of English (for now, Watson is fluent only in English) that Watson must absorb, and each company's implementation of those domains is a version of that dialect. Training Watson can take weeks or months, depending on the complexity. In an interview with Forbes, IBM said the likes of catalogs, training manuals, product disclosures, emails, customer forums, and call center logs, as well as publicly available feeds and reviews from places like Amazon.com, Yelp, Trip Advisor, and technical support communities, will all be grist for Watson's mill.
When Watson is up and running in the cloud, customers such as ANZ, Celcom, IHS, Nielsen, and Royal Bank of Canada will use it to create a customer-facing feature called Ask Watson. Consumers will be able to access it via Web chats, email, smartphone apps, and even SMS.
One of Watson's most striking features -- and one that will develop much more as cognitive systems roll out -- is its ability to understand unstructured data, which comprises the vast majority of data in the world. Even doctors' notes, which are famously hard to read, can be digested and are becoming a significant part of Watson's medical knowledge, Kelly says.
Quantum computing: Coming sooner than you think
In their book, Kelly and Hamm say, "People who demand the most from computers are already running into the limits of today's circuitry."
They quote Michel McCoy, director of the Simulation and Computing program at the Lawrence Livermore National Laboratory, noting that he is among those calling for a national initiative involving U.S. national laboratories and businesses to come up with radical new approaches to microprocessor and computer system design and software programming. "In a sense, everything we've done up until this point has been easy," McCoy says. "Now we have reached a physics-dominated threshold in the design of microprocessors and computing systems that, unless we do something about it, is essentially going to stagnate progress."
Kelly says part of the answer to that problem is what he calls data-centric computers. "The designers of computing systems have long treated logic and memory as separate elements. Now, they will meld the components together, first on circuit boards and later on single microchips. Also, they'll move the processing to the data, rather than vice versa."
Assuming that Kelly is correct, that shift obviously won't happen all at once. We'll see variations on silicon-based or silicon-like transistors over the next five years, he says. Indeed, Watson is really a hybrid of next-generation software and current-current generation hardware. Kelly makes the analogy to the first programmable computer systems that appeared decades ago: They had vacuum tubes; Watson systems are built on conventional silicon.
As I noted, Kelly is an optimist. "If you asked me five years ago how long it will be until we see quantum computers, I would have said 25 years," he said. "Now," he said as he held up two hands, "you could count the years on a bundle of fingers." The point is, it's really coming.
This article, "Bigger, better, stronger, faster: How IBM's Watson upends Moore's Law," was originally published by InfoWorld.com. Read more of Bill Snyder's Tech's Bottom Line blog and follow the latest technology business developments at InfoWorld.com. For the latest business technology news, follow InfoWorld.com on Twitter.