You may know Watson as IBM’s Jeopardy-winning, cookbook-writing, dress-designing, weather-predicting supercomputer-of-all trades. Now it’s embarking on its biggest challenge yet: Preventing cybercrime in finance, healthcare, and other fields.
Starting today, 40 organizations will rely upon the clever computers cognitive power to help spot cybercrime. The Watson for Cybersecurity beta program helps IBM too, because Watson’s real-world experience will help it hone its skills and work within specific industries. After all, the threats that keep security experts at Sun Life Financial up at night differ from those that spook the cybersleuths at University of New Brunswick.
Watson isn’t starting from scratch here. IBM researchers started training Watson in the fundamentals of cybersecurity last spring so the computer could begin to analysize and prevent threats. Now it graduates to real-world situations to further hone its skills. Think of it as the world’s smartest intern.
The computer’s done a lot of cramming since May. Cyberspace is vast, and the more of it Watson recognizes and understands, the more effectively it can discern the difference between benign threats and real problems. Watson’s great skill isn’t the ability to comb through huge amounts of information (though it does that too), it’s the ability to contextualize that information by combining structured data such as specific security events with unstructured data like white papers, research reports, and blog posts.
“Cognitive computing is 30-40 percent faster than traditional rule-based systems,” says Forrester Research analyst Andras Cser, who adds that cognitive systems like Watson also result in fewer false positives. Because it learns as it goes, it doesn’t repeat the same mistakes.
Before it could do that, though, Watson needed to study. Researchers fed Watson up to 15,000 documents each month through the fall, linking it to libraries and news feeds in real-time to keep its knowledge base current. Volume, too, is only half the battle; Watson must understand what words mean before it can know how they relate.
Take “ransomware.” You might know it as an increasingly prevalent type of hack that holds computers and systems hostage until the victim pays up. For awhile there, Watson, thought it was a place. “Our guess is that ‘Ransom’ is actually the name of several cities, and ‘ransomware’ is not something that generally appears in most dictionaries,” says Caleb Barlow, vice president of IBM Security. “We think it was trying to figure out what this thing was that was showing all the time, and asserting that it was a location.”
When IBM researchers annotated “ransomware” documents with a definition, Watson finally understood that his teachers aren’t obsessed with some far-flung metropolis.
It’s a fun anecdote that illustrates the challenges and opportunities of bringing Watson up to speed. What it doesn’t know, it can use context to guess. If it guesses wrong, it can learn. And once it learns, it never forgets. Which is where the beta program comes in.
Now, Watson’s beta isn’t like the kind you associate with consumer software.
“In a regular development project, you assign a testing matrix an go through testing to see how it deals with the tests,” says Barlow. “In this case, it’s much more like human learning. There are things it can do in grade school, things it can do in high school, things it can do when it goes off to college, and things it can do when it’s had a lot of experience in the industry. Watson will follow a very similar journey.”
In other words, Watson currently understands the fundamentals of security. Now it must learn the specific vernacular that peppers various use cases. “The language of security in the healthcare industry is probably completely different than the language of security in the energy sector,” says Barlow.
During the beta, Watson will embed with a few dozen companies and provide their security analysts with reports and recommendations. Specifically, Watson can identify whether a security event is associated with known malware and provide relevant background, as well as identify suspicious user behavior. (Think of password entry; are repeated failed attempts an absentminded user, or an attempted break-in?)
Watson doesn’t replace humans, but rather helps make them faster and more comprehensive in their response. IBM research shows that security teams sift through an average of 200,000 potentially significant events per day; a computer that prioritizes those events and reveals how they fit into the broader security climate saves worlds of time.
“I’d equate the traditional cybersecurity analysis model to standing at the side of a freeway trying to identify potential lawbreakers. As traffic whizzes by, it’s impossible to identify who is speeding or who might be in a stolen vehicle,” says Sean Valcamp, head of security at Avnet, a business-to-business electronics distributor that signed on with Watson in September. “Using Watson, on the other hand, is like flying over the same freeway in a helicopter.”
Watson won’t get it right every time, which is the point of a beta. It will learn from its mistakes—and, in the process, catch a few incidents that its human teachers wouldn’t. “You’re going to have your list of highly actionable events that come up every day, that are derived by analytics and humans. You’re going to compare and contrast that every day with what Watson came up with,” says Barlow. “Our objective here is to get Watson to find new things.”
And in the process, save companies—and their customers—from potentially crippling threats. Not a bad learning curve for a computer that until recently was looking for ransomware on a map.