Duncan Frissell's email (Sep 13 17:52:55), in reply to John Young, gives some problems regarding an AI screening program for detecting illicit electronic money transfers:
Not the least of which [problems] is that money launderers can use "AI Software" to generate a stream of real and dummy money transfers that emulates "normal" money transfers. Not to mention the fact that monopoly money transfer networks that can be surveilled by the Feds (FEDWIRE and SWIFT) are not long for this world. They will be replaced by encrypted, open, net-based systems.
There are also more basic problems. An essential initial stage in the development of such AI programs, a kind of expert system, is the analysis of the behaviour of the human "experts". In this case the experts being the willy money launderers. The program can then, aledgedly, spot all those trademark traits of the money launderer and nab him/her. Trouble is, the program therefore represents a kind of blue-print of what the aspiring money launder should *not* to do when making illicit electronic money transfers. Any "expert system" that operates by recognizing, and acting upon, human behaviour (regardless of whether or not that be on the net) can be foiled by altering, or disguising, that behaviour (the "dummy money transfers" mentioned by DCF would be one way of duping the system). Of course, the officials at law will claim rousing successes and cite numerous convictions -- but as with the drug trade, that will only be the tip of the ice-berg with no real clues as to the size of the base. The tip of the ice-berg will simply be what the AI program has succeeded in formally "capturing". Pete ================================================================= Dr Peter Madden, Email: madden@mpi-sb.mpg.de Max-Planck-Institut fuer Informatik, Phone: (49) (681) 302-5434 Im Stadtwald, W-66123 Saarbruecken, Germany. Fax: (49) (681) 302-5401