IBM was so sure that it had something special that it invited the makers of Jeopardy! to set up a stage at its research center and invite their two top scorers to tape a series of episodes of the popular game show. The third player was Watson, IBM’s latest innovation.

Watson is the result of the work of 25 IBM research scientists, who spent four years developing a system that can evaluate the equivalent of about 200 million pages of content, written in natural human language, to find answers to questions. It is powered by 10 racks of IBM POWER 750 servers running Linux and uses 15 terabytes of RAM and 2,880 processor cores. And it is capable of operating at 80 teraflops.
The choice to showcase Watson on Jeopardy! was not merely a carnival trick. The show’s trivia-style format is an enormous task for a computer as most computers do not understand natural language. It also requires a great deal of speed and, according to IBM literature, the ability to decipher clues with subtle meanings, irony, riddles, and other language complexities.

The research obviously paid off because Watson walked away with the game. Now IBM faces the challenge of introducing Watson to “the real world,” a world where businesses are drowning in an overwhelming amount of data and rely on business analytics to try to make sense of it all. The oil and gas industry is a prime target.

Information overload

The oil and gas industry has proven to be quite adept at finding ways to take more and more measurements in exploration, drilling, and production. It is having a harder time figuring out what to do with all of that information. According to John Brantley, IBM worldwide manager, Chemicals and Petroleum, Watson can help.

“Watson will allow you to have the information within the machine,” Brantley said. “We then build out different algorithms over time that allow you to go in and get different information.” This information could include lessons learned and best practices related to drilling analogs, extraction procedures from certain types of reservoirs, etc. “You can rapidly go after the information within a matter of seconds and also put a percent of confidence level on the information that’s being provided,” he said.

Already IBM is in talks with a number of oilfield clients, and Brantley said that one area of interest is seismic interpretation. “To the extent that you can teach interpretation to Watson and have access to that information, it also helps with the aging of the workforce and the requirement that you have to continue to do more exploration,” he said. “Watson does have the ability to learn over time.”
He added, “Our issue is not finding the demand.”

Jeopardy’s top winners vie against Watson, an English-speaking computer system.

A computer that talks back

The biggest difference between Watson and other computers is its ability to understand human language. Brantley said this ability allows companies to put volumes of written data into the system, so that analytics can be performed on the written word as well as numbers.

It is, in fact, so adept at language that it can pick up context clues to understand the true meaning of a word that can be used in multiple ways. For instance, if Watson is told that a person is running for president, it can understand that the word “running” means for office, not in a race.

Brantley said when IBM came out with its first PC in the early 1980s, computers knew with 95% accuracy about 5,000 words. “The major difference with Watson is that it understands 100 billion words,” he said. “That’s the big breakthrough – opening up the human language and being able to interact with a computer the way we interact with other humans.”

While Watson is currently queried via a keyboard, the next step is voice recognition. IBM also is examining multi-language skills. “I know English is supposed to be the global language, but the ability for a question to get answered in one’s native tongue would allow people to collaborate better,” he said.

Watson also can gauge the probability that the answer it comes up with is correct. “It allows you to go off into your database in multiple areas and try to string together a thought and then come back with the probability of success,” Brantley said. “If that probability is high, you want to check it. I don’t think anyone is ready to turn the world over to a system, but again it allows you to make better use of your data.”

Brantley sees a great future for Watson in the area of “streaming computing,” where data is coming in from the field in real time – 24/7. Watson could be trained to monitor all of the data and look for potential problems, not like a simple alarm system but a system that can recognize patterns and offer suggestions for action or improvement.

It is these high-value areas that should be Watson’s first challenge, Brantley believes. “What we’re talking about with customers is in the exploration and production space,” he said. “The gains are huge, and the costs can be high if it doesn’t work out.”

Dave Ferrucci, IBM scientist and Watson project director, shows the brains behind the system.

Watson’s future

Watson has tremendous potential in industries like oil and gas and medicine, Brantley said. In a sense, these will be easier challenges than a quiz show. “The algorithms in Watson were specifically written around Jeopardy!, which is very broad,” he said. “You’re not too sure what the question is, or the context. But once you go into an industry where the terminology and nomenclature is smaller, then you’re working with more of a known universe.”
Eventually Watson won’t require a large room equipped with multiple racks. John Kelly, head of IBM Research, claims that within 10 years the power of Watson will be available in a handheld device.

But the time to start experimenting is now. “This is a toolbox that builds on itself over time,” Brantley said, adding that Watson’s Jeopardy! “fame” is creating some competition among oil companies wanting to be the first to try out the new technology. “We are pushing customers not only on Watson but also around streaming computing – instead of storing data that gets produced from wells, actually doing something with it as it’s coming in. We’re very excited about the analytics-based work, and the next part of that is to tie it into a Watson-based system.”

Overall, the future is very exciting indeed. “I believe this is going to change the way very complex systems like E&P take place,” Brantley said. “We’re just on the cusp. It’s very exciting stuff.”