Murder by numbers

In London on July 10th, 2008, four people were murdered in separate incidents.
BBC correspondent Andy Tighe said "To have four fatal stabbings in one day could be a statistical freak."
Professor Spiegelhalter declares this number was predictable.

http://www.physorg.com/news156540743.html

Prof. Spiegelhalter and his team of statisticians "counted how many murders occurred on each day over a three-year period. Four murders on the same day in London would be expected to occur about once every three years, and it has done. Seven days without a murder should occur about six times a year, and it does."

The trouble here is that the overall patterns are predictable, whereas specific events are not.

I'd like to introduce you to Eliza, a computer program by Joseph Weizenbaum, designed to imitate a non-directional psychotherapist. Cutting edge stuff back in 1966, but it's now easily dominated by AIM bots.
Eliza's importance was the approach it used in developing AI, which undoubtedly inspired Robin Burgener, in 1988 to invent something you may already be familiar with, 20q.

Based on the less addicting parlor game, 20q is an AI website that can guess what you are thinking of in approximately 20 questions or less.
The thing is uncannily good in the same way Amazon.com has become eerily good at recommending Cure cds I already own after I tell it how much I enjoyed the movie Dark City.

Now this is something of a morbid thought, but could one possibly develop this AI to assist in carving away the specifics from the patterns of murder?