IBM plans to introduce technologies from its Watson computer, which beat humans on the game show "Jeopardy," for information-discovery use in enterprises.
The company hopes that Watson's underlying technologies will provide precise answers to complex queries using natural language interfaces, said Guru Rao, an IBM fellow last week. When applied in enterprises, underlying Watson technologies will be able to analyze warehouses of structured and unstructured data to determine answers with high levels of confidence.
[ Keep up on the day's tech news headlines with InfoWorld's Today's Headlines: Wrap Up newsletter. ]
There are many industries where Watson's technologies could be applicable, Rao said. Health-care providers could get precise answers to diagnose diseases or banks could detect credit-card fraud in real time. The technologies could also provide accurate answers through evidence-based searches related to cases in the legal industry.
"In a health-care world, if you know enough information about the diseases then it will be much more precise information. Using a natural-language query to get that degree of confidence could be much higher," Rao said.
In an ultimate game of man versus machine, IBM's Watson computer in February beat former champions Brad Rutter and Ken Jennings in the game of Jeopardy. The game showed how computers can answer queries with high levels of confidence, much like humans. Watson answered full questions posed by the host in a synthesized voice, with the computer's confidence levels to an answer being shown on a display.
Watson is a highly parallel computer running on Power7 processors with approximately 2,680 CPU cores, Rao said. Of the 15TB of information stored in the memory, about 1TB was being used on a consistent basis to answer questions, Rao said. Specific algorithms enabled processing of natural language queries, while other software and algorithms associated queries with keywords, known hits and other information stored on the system.
The system was able to process information in parallel by keeping the data in memory, which cut down any possible latency and I/O issues involved if data was stored on a disk, Rao said. The combination of in-memory processing, analytics and advanced hardware allowed Watson to answer queries in two to three seconds, Rao said.
Watson used the GPFS (general purpose file system) and software such as Hadoop, which provides a platform for analyzing structured and unstructured data in parallel across many servers.
IBM will likely work with partners to build server architectures around Watson for organizations to use, Rao said. Watson's software technologies could be tweaked for appliances dedicated to specific functions like security and compliance.
Watson demonstrated advances in computer science, much like its previous Deep Blue computer, which in 1996 lost a chess match series to reigning world chess champion Garry Kasparov. In a 1997 rematch, Kasparov lost in a series to an upgraded and more powerful Deep Blue.
"I don't see IBM providing a Watson-based game-playing service," Rao said. "Our goal is not to produce computer games, but [to] apply this ground-breaking computer science to business and enterprise-level issues."
IBM also wants to show new ways of human-computer interaction and the ability to weed out relevant information from warehouses of data.
The company has been adding FPGAs (field programmable gate array) to perform specific data processing functions on servers, while adding more cores to speed up chips. IBM is also redesigning servers to reduce latency between the CPU and memory, which helps speed up application performance. The company's EX5 server design, which was introduced last year, steps away from traditional x86 server designs by decoupling memory and processors into separate boxes. The separation provides servers access to a larger memory pool, and a special chip designed by IBM sitting near the CPU reduces latency between the memory and processor.
Down the road, the EX5 technology could include flash storage as a memory option, Rao said.
(Joab Jackson in New York contributed to the story.)