Data-Sifting Intelligent Agents Nourish Virtual Environments

Neural network software acts as early warning system to prevent equipment failures, predict business trends.

Sophisticated, pattern-recognizing artificial intelligence agents are solving quandaries faced by organizations that are being inundated by massive amounts of information. The design of these technodrones is based on the characteristics of structures that allow the human body to function. It enables systems administrators, both military and commercial, to monitor and pre-empt network catastrophes and allows corporate leaders to tap available data and take advantage of opportunities.

The increasingly fluid and fast-paced nature of business and military environments has chief executive officers and commanders scrambling for ways to contend with a deluge of information quickly and effectively. Faced with doing more with less, a situation currently aggravated by a shortage of qualified technical personnel, many organizations are turning to technology to address the challenges technology has created.

Experts agree that the linchpin of future success in the virtual environment will no longer be data management but rather knowledge and predictive management. Systems administrators and corporate leaders will rely on the synaptic speed of artificial intelligence to avoid network breakdowns and make business-generating decisions.

One software program developed by Computer Associates International Incorporated, Islandia, New York, employs neural network technology to provide military commanders and business development managers the ability to learn from the past and plan for the future. Unlike conventional artificial intelligence agents that require a subject matter expert to develop solutions to foreseeable problems and then encode the steps to address them, these neural agents, or Neugents, understand situations without manual intervention. By automatically learning how certain situations emerge, they effectively program themselves and can even suggest policy changes. They are typically deployed at the server level.

The company has incorporated this technology into two different product suites. Unicenter The Next Generation (TNG) is an enterprise management package that includes tools for network, desktop, server, database, operations, Internet, security, application and help-desk administration. Within this context, Neugents help systems administrators identify and repair problems within information technology systems. Jasmine The Next Dimension (TND) provides an information infrastructure that enables organizations to use technology as well as existing data and logic to build, deploy and manage intelligent electronic commerce solutions. Still in its initial stages, the program works within this tool to examine data and alert decision makers about previously unnoticed relationships between sections of information.

The common denominator in both of these environments is the existence of patterns—the precise material from which Neugents are designed to learn and work. By employing adaptive pattern recognition, the technology helps manage a chaotic environment.

Because organizations are taking advantage of an ever-increasing number of applications, systems administrators are becoming overwhelmed by the chore of keeping networks up and running. These tasks are reduced by employing Neugents not only to examine activity, but also to alert personnel when systems are reaching critical operational levels.

Computer Associates’ experience in the information technology field allows the company to provide clients with a basis for monitoring equipment. Unicenter TNG Neugents include Computer Associates-supplied information about more than 1,200 performance parameters; however, it is still necessary for them to learn about a specific customer’s systems. A client defines a certain scope or problem domain that Neugents observe for a set amount of time to identify common modes of behavior, Brandon J. Musler, director of product strategy at Computer Associates, explains. Once this data has been collected, the neural agents evaluate how hundreds of operating variables fluctuate over time and learn to associate patterns of change with changes in system performance. “Neugents know what pathways are outside of the state you have defined. They then can alert the systems administrator and say, ‘Every time you’ve gotten to this status in the past, 30 percent of the time this was the outcome, and 70 percent of the time that was the outcome,’” Musler says.

In one example, Unicenter TNG Neugents predicted a 95 percent probability that a specific server would run out of virtual memory in approximately 45 minutes. The systems administrator received the statistics that supported this prediction. These included data indicating that message and World Wide Web input was higher than expected, virtual memory read from disk was lower than expected, and virtual memory swapping, read operations and write operations as well as available random access memory were within acceptable limits. With this information, technicians were able to further examine systems and take the appropriate action to avoid a problem before it occurred.

The system is not totally foolproof. If, for example, the Neugents’ learning period was conducted during the summer months, it may send out alarms at the end of the year when a company is closing out its business for the year. “It will say, ‘I’ve never seen this activity before.’ Then you can override the alarm because you know this is an exception to the time period you monitored. Ideally, you can identify all states, but you can’t always do that,” Musler explains.

Although specific conditions may produce false alarms, the technology is an improvement over standard operating procedures for monitoring systems. Many technicians currently set specific thresholds. When activity exceeds these, erroneous alarms are sent to the systems administrator. “It’s like the boy who cried wolf. They can see false alarms so many times that they begin to ignore them. So when a real problem situation comes along, they don’t react. It’s like the old artificial intelligence. The old artificial intelligence used if-then logic. If this gets too complicated, then the programming gets too complicated, and the rules get too complicated. With Neugents, this is not a problem,” Musler states.

According to Patrick Dryden, senior industry analyst, Giga Information Group, Dallas, Texas, monitoring information technology systems is critical for more than pre-empting technical problems. In one incident, the technology detected that a company employee was downloading an extraordinary amount of data on a Saturday when activity was typically slow. In this case, the employee, who was scheduled to leave the firm, was taking proprietary information. By identifying this activity, Neugents helped company officials detect a data theft situation, Dryden relates.

The capability to study networks constantly and consistently is especially important today when there is a critical shortage of technical personnel. “You could set up someone to monitor and to identify busyness [on the network], but you couldn’t have someone baby-sit these servers. So by putting Neugents on the machine, it’s like having a person monitoring all of the time,” Dryden says. Using a computer to conduct this scrutiny also provides equipment examination by an entity that does not get bored or tired, he adds.

Future dependence on the digitized battlefield requires a means to offer an early warning system for equipment failures, Musler says. “In the electronic battlefield, everything is wired, even if it’s wireless. You have radar, sensors … it’s very important to be able to predict failure of critical systems,” he explains.

In addition to the technology’s role as a systems monitor, an emerging application for this learning program is as an intelligence-gathering tool. Faced with information pouring in from a number of sources, business and military personnel are searching for ways to digest data and put it into a useful context. “There’s a glut of information out there, and Neugents love a glut of information,” Dryden says.

Computer Associates has integrated the neural agent technology into Jasmine TND, a suite of tools that enables the rapid deployment of Internet, intranet and client-server applications. The product supports several deployment devices including palmtops and personal digital assistants, allowing corporate managers to deliver information directly to employees.

Unlike in the Unicenter TNG application for Neugents, the company cannot supply a set of metrics because the data being scrutinized for patterns can be sales, supplies, income or any number of other business-related variables. The agents examine this client-supplied information in search of patterns. The results can then be incorporated into future business plans or policies.

On the battlefield, commanders could use the neural agents to sift through the flood of information collected from any number of sources, including radar, unmanned aerial vehicles or reconnaissance patrols. The agents would look for past patterns and determine what will happen in the future. Commanders would be able to examine data collected from satellites over a long period of time and be alerted to a potentially dangerous upsurge in troops or equipment without designating a military unit to monitor and collect the information. “You don’t have to keep watching an area because Neugents will do this for you and alert you when it hits the level you’ve predetermined. There are no subjective errors,” Musler says.

“Neugents are good wherever there is a pattern. They can then go on and say that the last time this kind of buildup occurred, this and that happened,” he adds.

Because the software specializes in pattern recognition, organizations such as the National Security Agency can use Neugents to conduct cryptographic analysis and code breaking.

The technology also has been employed by private industry. To help increase profits during slow periods at a racetrack, one Asian company used Neugents to examine the various factors affecting attendance during a designated time period. The agents reviewed data supplied by the organization, including the types of races, jockeys, times of day, weather, days of the week, concessions and concurrently competing sports events.

“Just looking at the data and using common sense, they assumed that A-class races had the best horses and jockeys and that’s why they drew so many people. But when they looked at the results produced by Neugents, they saw that, although there were more people at the A-class races, at the B- and C-class races people were betting more money because the odds weren’t as bad,” Musler recounts. The racetrack could then make business decisions based on this information. For example, they might decide to focus on concessions or move A-class racing to slower days. “A lot can be said for perceptions, but sometimes these perceptions are wrong, and Neugents identify patterns that may never have been thought of before,” Musler explains.

Dryden calls this the bottom-line effect. “The business ramifications could be as mundane as how many cups or hot dogs or buns or employees do we need for this event on this day,” he says. Although the agents are not necessarily contributing to increases in business, they help identify ways to increase profits.

This capability also can be applied to electronic commerce in the virtual world. Musler offers the example of online booksellers, but adds that the concept applies to any business that provides a shopping cart as a tool for its customers. “It’s been determined that two-thirds of the people who put items in their [electronic] shopping carts leave the site before actually buying the products. This could be performance related, that is, people are frustrated that it takes so long to process orders, and they just leave. So you want to look at the performance of the networks and make sure that those people who buy are getting optimum service and faster performance. To monitor this, you don’t pay attention to all the traffic until someone puts something in a cart. Then, you can see what they are buying to identify usual patterns—for example, someone always buys romance novels. First, you correlate to the Internet protocol address and so on. Then we can move these people from being a suspect to being a prospect and raise the level of service we give them. Then, they are more likely to buy. So you can go from maybe copper service level to nickel to gold to platinum service to the point that, once they are serious about buying, you can now add things like suggestive selling. For example, a prompt that would say, ‘You like this author. Did you also want to order his new book?’ We’re not at the level of technology where we can offer, for example, faster performance because there are too many technology variables, but this will change. When it does, you’ll already have this in place to take advantage of it,” Musler says.

While neural network technology has been around for some time, Dryden explains that Computer Associates’ offerings are different in that they are providing technology that, rather than working the same way each time, can be applied in different situations.

Although Neugents are valuable in circumstances that involve data and patterns, they are still unable to predict random events. “They couldn’t predict the weather, but for example, here in Texas, we’ve had a lot of planes landing at the airport that experience problems during storms. Neugents could look at occurrences of other factors like wind speed and direction and could then say it may be unsafe to land a plane, for example, because in the past there was a problem under these conditions,” he says.

Musler agrees that there are limitations. “The temptation is to think that they [Neugents] work by themselves. We’re not selling a brain in a box. These are tools to look at intelligence in a context that’s useful,” he says.

Although not the total solution that some have been predicting, the technology is a step in that direction, Dryden says. “For years and years, there’s been a dream of self-healing networks so that they would have enough intelligence that they could detect problems and solve them and then report what’s been done to the person in charge. We have a whole boatload of information so no human can keep on top of it. Operators see all errors, and it’s too much so they empty it all and see what comes back and then address these problems. Systems administrators don’t need to be tied down looking for problems. The information systems group has to pay attention to business needs,” he adds.