As I mentioned in the introduction, during my last two years at the Office of Technology Assessment, I was part of a three-person team that studied the use of data mining technologies to look for evidence of money laundering in large databases of financial transactions. Criminal money laundering is the attempt to hide the origin or destination of funds gained through illegal activities.

An agency of the Treasury — the Financial Crimes Enforcement Network (FinCEN) — collects and analyzes data on currency transactions of more than $10,000 dollars. In the mid-1990s, FinCEN received about 10 million new currency transaction reports each year. FinCEN analyses these records, and other records, in an effort to detect money laundering. It forwards its findings to federal, state, and local law enforcement agencies.

The Senate Permanent Subcommittee on Investigations, who requested the OTA study, was interested in whether Congress should require the collection of an order of magnitude more data — upwards of 100 million wire transfers sent each year. Their concern was that such transfers could be used in money laundering, and that FinCEN might miss many cases of money laundering unless they could access this source of data.

Our report focused on three primary questions. First, the study examined whether the analysis of wire transfers would substantially improve the ability of FinCEN to detect money laundering. Second, we examined whether the analysis of wire transfers could be automatic and sequential. Some consultants lobbying the Subcommittee argued that each transfer could be examined individually by an automated system and that such a system could detect whether the transfer was involved in money laundering. Third, we examined the impact of the proposed systems on individual and corporate privacy.

I was the "technical muscle" on the project. My dissertation, completed in 1992, focused on what we now call data mining systems. As a result, my talk will deal primarily with the first and second questions. That said, questions of the privacy impact of data mining systems are extremely important, and I will touch on these questions as well.