[ExI] The NSA's new data center

[Anders Sandberg]

It is worth noticing that changing the budget assumptions by a factor of 
ten in my model only shifts the arrival dates by a few years. Eventually 
the Railroad Retirement Board might decide to put apart some of its 
budget for the task...


On 24/03/2012 18:34, Adrian Tymes wrote:
> Suggest this to any actual prison warden, and they'll point out 
> there's no way they'd have the budget for that. Some of this can be 
> argued as priorities - money goes to things that are important. But if 
> it really was that cheap to do complete surveillance, this wouldn't be 
> enough to rule it out. Conclusion: it is not, in fact, that cheap to 
> do complete surveillance, likely due either to costs not accounted for 
> or the assumption of more capital than is in fact available. 

It is not cheap or efficient to do total surveillance *yet*. When data 
processing, storage and analysis is expensive you cannot afford to do 
more than to focus on persons of interest, and you will have to throw 
away superfluous information because you cannot store it. As they get 
cheaper you still want to ignore noise and there might be legal 
constraints on who you are allowed to intercept, so you add filters. But 
eventually the cost of data gathering and storage will become small 
enough that it is cheaper to first get all data, and *then* filter out 
what you want from it.

I suspect that the expensive part will remain analysis. Sensors are 
getting cheaper as per Hendy's law by a factor of about 100 per decade; 
the Internet of Things will be an excellent surveillance tool paid for 
by consumers, and so on. Analysts are paid salaries and can only process 
a working day of information per day. The key factor influencing how 
efficient they can become is software support: automatic pattern 
matching, decision support, all sorts of AI apps - as they become 
better, then the same staff can do much more. So what I *really* would 
love to know (and my friends at IARPA will *never* tell me :-) ) is 
whether that software is showing exponential-ish improvement. I suspect, 
given current trends in language crunching, that this might be happening.
http://www.nytimes.com/2011/03/05/science/05legal.html?pagewanted=all
This might not lead to better eventual decisions, but it might very well 
enable managing total surveillance on a limited staff budget.


-- 
Anders Sandberg,
Future of Humanity Institute
Philosophy Faculty of Oxford University