I thought it might be interesting to look at some phone usage statistics. I went to the FCC's website and looked at the most recent report that I could find:
Unfortunately, the most recent stats that I can find are for 2003. Anyway, as far as I can tell from the table on page 10-4, there are over a trillion phone calls placed every year, and I do not believe that includes cell phones. This would average out to about 2.7 billions calls placed a day, which comes down to about 2 million calls placed per minute. This is probably a conservative estimate, since it appears not to take cell phones into account.
We know that these calls are placed into a database, and supposedly used to monitor potential terrorist communications. I assume the idea would be to have the system flag calls that are related to potential terrorists, or individuals who are suspected of terrorist ties. In theory, this could be used to find a correlation between different individuals who are working together, but may be trying to hide their connection to one another.
Let's assume that the system is ridiculously accurate, and that it properly classifies 99.999% of the calls accurately. This would mean that 0.0001% of the calls are classified incorrectly, i.e. they are false positives, or normal benign citizens engaging in typical phone calls. So, 0.0001% of 2 million means that they could be averaging around 200 false positives per minute. Over the course of a year, this could average out to almost 2 million false positives. Keep in mind that I am ignoring the cases where the system misses terrorists all together.
I assume that alarms generated by the system will involve further investigation from intelligence personnel. Assuming that it takes an agent an average of 8 hours to thoroughly investigate a suspect, this would mean that they would need a staff of:
So, they would need 96,000 people at any given hour ready to investigate the false alarms kicked out by the system. Clearly, the cost of this system is going to be very prohibitive. In addition, all of the estimates are VERY conservative. In all likelihood, there are more calls placed, and the system is probably nowhere near as efficient as 99.999%.
Conclusion: The odds of this system accurately predicting possible connections between terrorists is very low. At best, it may be used with other sources to establish some sort of correlation. Most likely, it will be used as retroactive evidence of a successful system. In other words, let's say 4 seemingly independent individuals engage in 4 correlated terrorist attacks. After the attack has occured, the NSA could look through their database, and establish a connection between those 4 individuals. Then they could announce a press conference and say, "See... the system works. We know the connection between these 4 terrorists." However, it won't matter, because it will be too late. Plus, millions and millions of dollars will have been spent for a system that produces unreliable alarms that tie up valuable man-hours. In addition, I'm not certain that any statistical models have been tested to ensure the viability of a system of this magnitude. In other words, they are probably using an un-tested system, in which the false alarm could lead to significant troubles for a multitude of otherwise innocent people.
I thought this was a forum for information concerning science and technology. Why is/. covering this story if it has nothing to do with any of that. I see politics everywhere I go, does that mean I now have to deal with politics from here?
Slashdot Mirror
I thought it might be interesting to look at some phone usage statistics. I went to the FCC's website and looked at the most recent report that I could find:
/ FCC-State_Link/IAD/trend605.pdf
http://www.fcc.gov/Bureaus/Common_Carrier/Reports
Unfortunately, the most recent stats that I can find are for 2003. Anyway, as far as I can tell from the table on page 10-4, there are over a trillion phone calls placed every year, and I do not believe that includes cell phones. This would average out to about 2.7 billions calls placed a day, which comes down to about 2 million calls placed per minute. This is probably a conservative estimate, since it appears not to take cell phones into account.
We know that these calls are placed into a database, and supposedly used to monitor potential terrorist communications. I assume the idea would be to have the system flag calls that are related to potential terrorists, or individuals who are suspected of terrorist ties. In theory, this could be used to find a correlation between different individuals who are working together, but may be trying to hide their connection to one another.
Let's assume that the system is ridiculously accurate, and that it properly classifies 99.999% of the calls accurately. This would mean that 0.0001% of the calls are classified incorrectly, i.e. they are false positives, or normal benign citizens engaging in typical phone calls. So, 0.0001% of 2 million means that they could be averaging around 200 false positives per minute. Over the course of a year, this could average out to almost 2 million false positives. Keep in mind that I am ignoring the cases where the system misses terrorists all together.
I assume that alarms generated by the system will involve further investigation from intelligence personnel. Assuming that it takes an agent an average of 8 hours to thoroughly investigate a suspect, this would mean that they would need a staff of:
(200*60)calls/hour * (8)investigator-hours/call * (1)investigator/investigator-hours = 96,000 investigators/hour
So, they would need 96,000 people at any given hour ready to investigate the false alarms kicked out by the system. Clearly, the cost of this system is going to be very prohibitive. In addition, all of the estimates are VERY conservative. In all likelihood, there are more calls placed, and the system is probably nowhere near as efficient as 99.999%.
Conclusion: The odds of this system accurately predicting possible connections between terrorists is very low. At best, it may be used with other sources to establish some sort of correlation. Most likely, it will be used as retroactive evidence of a successful system. In other words, let's say 4 seemingly independent individuals engage in 4 correlated terrorist attacks. After the attack has occured, the NSA could look through their database, and establish a connection between those 4 individuals. Then they could announce a press conference and say, "See... the system works. We know the connection between these 4 terrorists." However, it won't matter, because it will be too late. Plus, millions and millions of dollars will have been spent for a system that produces unreliable alarms that tie up valuable man-hours. In addition, I'm not certain that any statistical models have been tested to ensure the viability of a system of this magnitude. In other words, they are probably using an un-tested system, in which the false alarm could lead to significant troubles for a multitude of otherwise innocent people.
-D
I thought this was a forum for information concerning science and technology. Why is /. covering this story if it has nothing to do with any of that. I see politics everywhere I go, does that mean I now have to deal with politics from here?