Reader's advisory: Wired News has been unable to confirm some sources for a number of stories written by this author. If you have any information about sources cited in this article, please send an e-mail to sourceinfo[AT]wired.com.
As financial transactions continue to be processed electronically, fraud-busting software tools need to move beyond the cyber-version stage of a high school hall monitor randomly checking bathroom passes.
Enter the next generation of intelligent systems, sometimes called artificial intelligence: technology that can think, learn, and react as humans do, but won't get bored by repetitive work or distracted by the cute person in the next cubicle.
Engineers are presently developing hybrid intelligent systems that incorporate all of the tools developed during the first wave of AI technology.
The London Stock Exchange uses MonITARS (Monitoring Insider Trading and Regulatory Surveillance), developed by Searchspace. The application melds genetic algorithms, fuzzy logic, and neural network technology to detect suspicious trades within the vast amount of data that electronically passes through the exchange each day.
"The system can even detect dealing rings where several individuals or one individual with several accounts trade across all their accounts for the purpose of insider dealing or market manipulation," Searchspace marketing director Jan Tellick said.
Searchspace calls its hybrid technology iTM, or intelligent transaction monitoring. ITM differs from the standard monitoring tools, Tellick said, because it can learn. It can understand business practices based on its analysis of transaction activity, using more flexible reasoning logic than the traditional on-off, yes-no type of processing.
Instead, iTM uses information collected from an internal profiling database. The technology dynamically updates and educates itself each time it handles a new transaction.
The newest high-profile Searchspace client is the Bank of New York, which was investigated last year for allegedly laundering $7 billion.
The bank was not charged with a crime but agreed to improve its auditing, due diligence, and risk management. The bank will supplement its existing monitoring systems with iTM-powered applications to comply with the mandate.
Fraud-detecting technology has been widely deployed throughout the financial industry, but until recently the available tools have focused on one specific application of artificial intelligence.
Genetic algorithms are intelligent systems inspired by biological evolution; these systems use a "survival of the fittest" strategy to eliminate all but the best solution to a problem. The algorithms are used in bankruptcy prediction, financial forecasting, and scheduling for big events like the Olympics.
Neural networks can recognize patterns, make associations, compare and contrast activity within the system to past events, and formulate ideas based on these concepts.
Neural nets are being used for fraud detection at Visa International, Fidelity Investment, Citibank, and American Express. The networks learn to spot fraudulent activity by comparing data on legitimate card usage against known cases of fraud.
HNC Software is the largest supplier of NN-based credit card fraud-detection software. Its Falcon system is used by 30 card issuers worldwide, monitoring over 90 million accounts. HNC says Falcon users, on average, have cut their losses due to fraud by 25 percent.
HNC has created a data warehouse of the transactions of over 200 million cardholders to build the predictive models. These models recognize the transaction behaviors of people using fraudulent or stolen credit cards. The same fraud detection software is applied in other markets, including insurance.
"We're very satisfied with the results of the initial use of HNC Insurance Solutions' VeriComp software," said Bob Short, senior vice president of a Utah state fund providing both employers and employees with workers' compensation insurance.
"We've saved money, and have also obtained a consistent fraud identification method, a more focused review process, better resource management, and earlier detection of potentially fraudulent claims."
Nestor is another software supplier that produces NN-based fraud detection systems. It supplies a system called FDS, which it claims has reduced fraud by 20 to 40 percent in live tests. Mellon Bank, which installed FDS in 1992, reports that the system has increased fraud detection rate by a factor of 20 while reducing the number of false positives by a third.
Searchspace's Tellick says more sophisticated criminals constantly change and evolve their methods so any automated approach to fraud detection will be outdated almost as soon as it becomes effective. The criminals simply shift their attention to new weak spots in the system.
But since iTM can actually learn as opposed to simply remember, it can easily adapt to new circumstances and adjust its response accordingly. And if a system can learn from past experience, it's unnecessary for the designer to specify all the operating conditions under which the system is to perform.
"A system that is free to make its own associations between the events that it experiences should be capable of finding relationships that were unknown to its programmers," Tellick said. "In terms of fraud detection, this potential is extremely valuable, because it means that the system itself can contribute creatively to the detection process."
