[ExI] [tt] The NSA's new data center

Eugen Leitl eugen at leitl.org
Tue Mar 27 18:23:49 UTC 2012

----- Forwarded message from "J. Andrew Rogers" <jar.mailbox at gmail.com> -----

From: "J. Andrew Rogers" <jar.mailbox at gmail.com>
Date: Tue, 27 Mar 2012 11:01:01 -0700
To: tt at postbiota.org
Subject: Re: [tt] [ExI] The NSA's new data center

On Tue, Mar 27, 2012 at 9:47 AM, Eugen Leitl <eugen at leitl.org> wrote:
> ----- Forwarded message from Anders Sandberg <anders at aleph.se> -----
> From: Anders Sandberg <anders at aleph.se>
> Date: Tue, 27 Mar 2012 17:18:41 +0100
> To: ExI chat list <extropy-chat at lists.extropy.org>
> Subject: Re: [ExI] The NSA's new data center
> User-Agent: Mozilla/5.0 (Windows NT 5.1;
>        rv:10.0.2) Gecko/20120216 Thunderbird/10.0.2
> Reply-To: ExI chat list <extropy-chat at lists.extropy.org>
> On 27/03/2012 10:57, Eugen Leitl wrote:
>> The centralist high-expert-power surveillance approach will always beat
>> decentralist many-shallow-eyeballs sousveillance approach.
> I think this is an important angle.
> What properties of technologies tend to give states (or other big power
> concentrations) advantages in power compared to the citizens? A few
> possibilities:
> * High cost of entry (keeps small actors out)
> * Economies of scale (big systems do the job better)
> * Produces information or products of more use to states than individuals
> * Requires rare expertise that cannot be distributed

The important factor that is often overlooked when discussing
centralized surveillance versus decentralized sous-veillance is that
distributed/parallel algorithm optimality or efficiency is often
predicated on fairly narrow topological assumptions. In a centralized
facility, the topology can be designed for the problems being solved.
In an ad hoc decentralized network it is virtually guaranteed that
parts of the topology will be pathological, making many algorithms
effectively intractable in such environments. The reverse is not true
of course, algorithms that are not sensitive to topology will still
work just fine in a centralized data center.

It turns out that the algorithm families related to deep relationship
analysis -- the kind of algorithms you use to invade privacy -- are
very much in the former category. They only parallelize well under
topological constraints so narrow that even having a giant data center
is insufficient; you would need to deploy the data models on the
cluster in carefully designed logical topologies in addition to a
carefully designed physical topology.

In this case, "winning" is biased toward whoever has both the biggest
data centers and the best theoretical computer scientists.

J. Andrew Rogers
tt mailing list
tt at postbiota.org

----- End forwarded message -----
Eugen* Leitl <a href="http://leitl.org">leitl</a> http://leitl.org
ICBM: 48.07100, 11.36820 http://www.ativel.com http://postbiota.org
8B29F6BE: 099D 78BA 2FD3 B014 B08A  7779 75B0 2443 8B29 F6BE

More information about the extropy-chat mailing list