DOE designs system for secret documents
- By Elana Varon
- Jan 04, 1998
The Energy Department plans to develop a cutting-edge computer system to speed the declassification of some 280 million pages of information dating from World War II to the present.
The system, described in a solicitation issued last month, would combine imaging, work flow and artificial intelligence technology to automate the review of still secret documents. Ultimately, DOE's Office of Declassification wants to replace humans with software to identify sensitive materials. The software would use linguistic databases and programs capable of "learning" how to make classification decisions.
According to the request for proposals, DOE needs the system to cope with the volume of materials in its vaults. A study commissioned by DOE from DynMeridian and Integrated Resources Group Inc. and published earlier this year said the department's declassification staff is already "working at full capacity." Under a 1995 presidential order, all federal agencies are required to declassify millions of Cold War documents, and the department has already released about 11 million pages.
In the past few years, agencies have deployed a variety of technologies to aid their declassification work, including scanners to digitize paper documents, search engines to locate words in these documents relating to classified subjects, "redaction" tools that allow reviewers to block out sections of documents that must remain secret and workflow software to route documents through the entire process. "Those work for speeding the process," said David Lipstein, director of market development with Eastman Software Inc.'s federal division.
But Lipstein, whose company has supplied software for declassification and related Freedom of Information Act review systems in the Defense Department and the Immigration and Naturalization Service, said that to date, imaging and workflow technologies have not been successfully integrated with artificial intelligence software on a broad scale. "That part is cutting edge," he said.
DOE officials could not be reached for comment at press time, but according to the solicitation, the department plans to purchase the system incrementally, "fielding components as they are proven." According to the market research firm Federal Sources Inc., the five-year contract would be worth about $6 million.
How advanced a system DOE is able to develop may depend largely on its budget, observers said, especially because Congress did not budget money for the project this year.
"To achieve what they are talking about is a very, very expensive system," said Brad Willcockson, vice president of business development with Severn Companies Inc., Lanham, Md. Severn provided the State Department with a system for reviewing and redacting documents several years ago.
The DynMeridian study said the Office of Declassification could achieve "near term" improvements in its productivity by enhancing the capabilities of its Electronic Document Declassification System to include automated workflow and better online reference materials. The report said DOE also could revamp existing decision-making software and deploy new tools that currently exist as prototypes, but applying artificial intelligence to the review process, would require more research and development.
According to information posted on the World Wide Web site of the Declassification Productivity Research Center, a federally funded think tank at George Washington University that is helping agencies assess technologies for declassification, "full manual review of each paper document by two highly trained and certified reviewers is still the best and only reliable means of determining" whether a document still needs to be kept secret.
Steven Aftergood, who heads a project on government secrecy for the Federation of American Scientists, said agencies are devoting too many resources to automating their review processes and not enough getting documents into the open. "In principle, automation might help, but at this stage it is detracting from the actual release of documents," he said. "That, of course, is the test, not whether the software is elegant but whether I can get my hands on documents."