Re: Are there enough FBI agents to handle Digital Telephony?????

[Tatu Ylonen <ylo@cs.hut.fi>]

> For instance, if 1% of America is on the phone during the peak
> hours of the day, then that puts 2.5 million Americans on the
> phone or 1.25 million conversations. If the FBI wants access to
> 1% of that, that is still 12,500 simulataneous conversations.
> That would seem to imply 12,500 people to listen to the tapes,
> right? Would that take agents off the streets?

On the other hand, if you only want to collect rough background
information about people who might become significant later, it is
enough to store the conversations in a computer (storage is orders of
magnitude cheaper than the man-time to listen to the tapes), and only
listen the tapes if the person becomes interesting.

Besides, computer technology is approaching the point where you can
eliminate the human from the link entirely, except for final
verification.

1. Speech recognition already works quite well.  There was an article
about a 20.000 word speaker-independent system a few years ago,
operating 1/7th of real time on an alpha workstation.

2. Automatic speaker recognition from voice works quite well if my
understanding is correct.  (Useful for picking up interesting
conversations for futher analysis when you get them from sources you
don't normally monitor).

3. Computers have been able to pick up potentially interesting
conversations by keywords for decades.  Also useful for picking up
interesting conversations for further analysis from sources you don't
monitor very actively.  (Of course, you can additionally use phone
numbers, mobile phone *phone* identification codes, etc.)

4. A lot of work is being done in classifying transcript based on
their content, on message understanding.  In other words, lot of the
analysis work can be automated now or in near future.  The computer
can then answer questions from the data and for example select
individuals for futher analysis based on complex criteria.

5. A lot of work is being done on data mining (i.e., finding new data
from small pieces of individual data in a database, such as purchase
logs, etc.).  This is one of the hot topics in database conferences
right now.

6. Research is being done in massive databases.  There was an
Intelligence Community research initiative a couple of years ago on
massive databases; if my memory serves me right, they were talking
about 2-3 *petabytes* (10^12) as the size of the final database (no,
it was not gigabytes and it was not terabytes).  I believe I still
have the announcement saved somewhere if somebody wants it.


All of these technologies are feasible now or in near future.

Then add a little spices: the hundreds of thousands of surveillance TV
cameras around (did you know that there is a computer system that can
recognize and look up 25 faces per second from a database of a million
faces - used to control football huligans and shoplifters for
instance, but has other uses as well), car movement records from
highway payment systems, purchase records obtained from
credit card companies, banks and retail chains, link up to medical
records, tax databases, employment records, etc.  Add full knowledge
of flight and other travel reservations, some fax, e-mail and telegram
intercepts.

Now, what have you got (besides effective tools for finding criminals)?

    Tatu