Science & Technology

Could Superior Search Technology Have Helped the FBI with McVeigh?

PRNewsWire - June 8th, 2001

Like a spent boxer on the ropes, America's Federal Bureau of Investigations reels from one information management crisis to another. The bureau couldn't find evidence to convict nuclear scientist Wen Ho Lee of any serious allegations; it took 15 years to find alleged Moscow mole Robert Hanssen, despite abundant evidence of his misdeeds; and more recently, it lost 4,000 pages of critical documents in the most prominent and deadly case of terrorism in US history.

How is this "investigative agency" retrieving clues, documents, and files, and what can technology do to resolve obvious problems?

The FBI has spent over $100 million dollars in some recent years just to electronically copy and file its massive reams of paper, so it must also have an expensive electronic knowledge management bill. And to retrieve whatever it manages is more complex than in less secretive organizations. Some government documents remain filed away in the most obscure and classified places even if they refer to historical events. For example, Harpers reported in May 2001 that the US government still maintained under seal 4.6 million documents related to the Iran-Contra investigation. So an organization like the FBI obviously squirrels away lots of files in lots of places. That's fine, as long as its agents can find the files.

The FBI can turn to the latest Internet and Intranet search engines to provide solutions. But these all face known problems of information retrieval. Many are strained if you invent your own classification schema to match specialized needs; others may permit an organization to write linguistic rules or add keywords in their classification system to build a customized search engine. But this may create more problems than it solves, for example, if editors trained in cloak and dagger work but not computational linguistics place documents in inappropriate boxes. Some engines rank information with the help of a popularity contest, but this won't help you find something that is intentionally hidden or compartmentalized.

Most serious for the FBI and other organizations researching highly specialized information will be the problem of exclusion presented by any system using any keywords to seek relevant data. It might be disastrous if a document does not contain a keyword (e.g., "McVeigh") but nonetheless contains words relevant to the investigation (e.g. "John Doe 2").

One company founded by a former Princeton University professor, Mnemonic Technology, claims that it can solve many of the problems presented by traditional search technologies. For the FBI or anyone else carrying out an ongoing investigation, it offers special "intelligent agents" that are designed for recurring searches. How do these special intelligent agents work?

"Our information management system is unique in that it is designed for recurring searches," says Dr. Eric Sven Ristad, MIT graduate, computational linguist, author, founder, and CEO. "It enables the user to express interest or lack of interest with respect to any 'hits' that a search engine returns. It then statistically models expressions of positive and negative interest to rank all documents in order of greatest relevance."

In the case of the Oklahoma City bombing files, a critical, useful feature of the Mnemonic technology is that no document is excluded from returned results; the documents are merely ranked in order of likely relevance. A diligent researcher could go through any number of documents ranked by relevance, or retrieve all of them. The system also automatically notes key, recurring words and phrases in all documents that a researcher considers interesting. The search then becomes increasingly accurate as trained agents learn from the researcher about how previous search results have matched his preferences. As information is retrieved from the database, an electronic agent acts like an increasingly intelligent seeing eye dog, trained to re-prioritize the most important morsels of knowledge and deliver them like a pair of familiar slippers.