t byfield on Sat, 7 Mar 1998 10:49:21 +0100 (MET)

[Date Prev] [Date Next] [Thread Prev] [Thread Next] [Date Index] [Thread Index]

<nettime> an entirely new subject

I spent a little time this afternoon composing a response to Mr.
Sanborn's eruption, but it proved to be too much of a chore; he
doesn't really seem to have a point, other than branding his in-
terlocutors as neoconservatives, neo-manicheans, Machiavellians,
and social darwinists. (Fascists can't be very far off now, can
it?) In any event, I thought to myself, Self, how about writing
something Constructive for a change? So here's something that I
have been thinking about a bit: databases [turgid spoiler here;
if you hate turgid blocks of text, don't say I didn't warn you]
<And, whether they like it or not, extra-special thanks to Pit,
Diana, and D. S. Bennahum for getting me thinking again, FWIW.>
(And if you got this as a BCC, then it's dedicated to you... :)


Until quite recently, most programming languages worth consider-
ing would fail unless every element they encountered was explic-
itly classified in advance. Put simply, if a program would need
to process a datum, the programmer would need to announce in ad-
vance how the program should parse it. While this is still true
for the most part, the accumulation of human effort that's gone
into software development has pushed most interactions (even of
"developers") to "higher" levels--that is, forms of interaction
that don't require such intimate interaction with data. I'm not
talking about pointing-and-clicking in a GUI, or optical charac-
ter recognition programs; however sophisticated, processes like
this are ways of facilitating input and output--in short, human
interaction interfaces. Rather, what interests me are databases
that don't need to know the "type" of the data they store prior
to incorporating it. That means: a database that can store--and
process, within limits--streams of data (for example, raw ether-
net traffic over N period), memory core (the momentary state of
a system *in the process of operating*), media states (an image
of a hard drive), simple matrices (a subdatabase) or complex ma-
trices (arrays of "time-series" data)...and, of course, formats
we all know, such as ASCII, numbers, hex or binary data, images
(TIFs, PICTs, BMPs, faxes), word-processing documents, video or
sound streams or analog streams (such as encoded radio signals).
It's hard to grok just how important this is until one realizes
that a database is, by definition, more than a way to put files
somewhere, anywhere, wherever, and tagging them with serial num-
bers; there's no doubt that actually making a database that can
store these various formats without mangling them was a supreme
achievement for its programmers, but it'd be useless for anyone
else. Instead, certain kinds of databases can, more than merely
storing data of undeclared "types," *analyze* data according to
the peculiar structure of that data. Since almost all available
data is structured according to standardized methods, it's fair-
ly simple to declare the range of likely structures a given dat-
um may correspond to and discover which one it is, even when it
wasn't "declared" when it was incorporated. But declaring every
single type, and then developing numerous ways to analyze every
possible format, and then developing metaformal structures that
will allow users to construct their own tools for analyzing any
given datum or array of data... In principle, it's much more ef-
ficient to develop formal methods for analyzing data in general,
because the "native" structure of any given element will reveal
its own peculiar logic and greater or lesser resemblance to any
other datum. What you're reading right now, for instance, ASCII
text, has certain peculiar traits--the logical patterns in meth-
ods of encoding characters, the patterns of justified text, the
patterns imposed by the English language (the incidence of this
or that letter, word length, grammatical and syntactic patterns,
and so on)--all of which are encoded in overlapping frequencies
in the binary data this essay consists of. Had I encrypted this
message, someone who didn't have the key but wanted to crack it
would search the data for ways to find patterns that correspond
to known linguistic characteristics; knowing that I write in En-
glish would help, but wouldn't be necessary; and knowing that I
write in a Latin alphabet would help, but wouldn't be necessary
--in both cases because the data structures in which these meta-
languages (English, ASCII) are encoded reveal their own particu-
lar patterns. Once the basic encodings are found the high-level
encodings are relatively trivial to find. And so it is with new
databases: it's often more efficient to examine a datum or data
stream for "behavioral" patterns than it is to pay attention to
what it (or someone) says it is. I've explained one reason this
is so, but there's another equally if not more important reason,
a fundamental principle of database construction: the potential
value of data isn't limited to the interrelationships we antici-
pate. By limiting the elements one incorporates into a database
to what you (*think* you) know--for example by declaring "types"
--you limit what you might discover. In other words, by telling
a database "this element is ASCII" and "that element is a video
stream," you'll lose more (by categorizing) than you might gain
(by examining). Now, I should say, this is a gross oversimplifi-
cation of the specific techniques that modern databases can use
to function efficiently in everyday practice: there'd be little
point in searching through an hour-long video file just because
someone wanted to find this or that string of text in a memo he
or she sent last week. But the fact remains that some databases
will allow you to find every use of, for example, a picture--in
video files, in image files, in images embedded in word-process-
ing documents, and in elements whose type was never declared to
the database. But I wouldn't write this much just to extoll the
wonders of modern database technologies, that's not my point at
all. Rather, what's remarkable about this, I think, is where it
will lead in terms of the way we communicate. To declare a data
type is, at root, to declare the GENRE of an utterance: this is
fiction, this is nonfiction, this is a book, this is a love let-
ter, this is an art film, this is a laundry list... And what we
are now seeing is the rise of a technique for folding every pos-
sible genre we might think up into a metastructure that regards
genres--categories that have served to structure the communicat-
ive basis of our knowledge--as *obstructions* to possible forms
of, not "knowledge" as we now understand it, but meaningful cor-
relation. The problem is, of course, that the "agent" that will
perceive this correlation and thereby construct the meanings is
no longer an individual or even a group of people; and so we'll
have no way, at least according to our current ways of thinking,
to verify the validity, worth, or use of these correlations. We
won't believe it because we won't--can't--know *how* to believe
it; and so we shall argue in various ways that the correlations
are *meaningless*. According to the constitutions of knowledges
we know, we will be right; but according to the ways of constit-
uting knowledge we do not and cannot know, we'll be quite wrong.
The question I would like to leave you with is this: Are you an
Ancient, or are you a Modern? Do you firmly believe that the ca-
pacities of an individual mind (or group of individuals) is the
basis of all valid knowledge, or are you willing to accept what
a machine says is so? One can, of course, dismiss this question
by pointing out that machines ask only those questions they are
told to ask, so the choice that I've presented is a false dicho-
tomy. To be sure--for a few years. And then what? Much as expli-
cit data-type declarations have, through cumulative labor, been
obviated by modern databases, I am very sure that explicit quer-
ies will be obviated soon enough: their structures, values, and
terminologies will be built--by human beings--into basic system
functions. So the question I posed can be put off, but it won't
go away--not forever, and in fact, not even for the foreseeable
future. Not that I could answer this question myself. No one, I
think, would relish the prospect of a fundamental shift in epis-
temology--which, if you haven't noticed, is in no way a "theore-
tical" problem.  On the contrary, it's a palpable, painful prob-
lem that's ever-present in everyday life: the lingering culture
wars over the validity of psychoanalysis (wherein data does not
correspond to its declared type), whether individual experience 
should take priority over sociological statistics, the relative
merits of legal theories of individual privacy and group securi-
ty--these all prefigure and contribute to the ways in which the
shift toward disembodied or supraindividual "knowledge" will af-
fect us all. But more important, for me, than the affect it has
on individuals is the prospect of a meta-genre: a database, may-
be many, that can produce from what it stores whichever genre a
user requests. Or, more important, what it might produce when a
user no longer needs a genre to construct any form of knowledge.

---t. byfield <tbyfield@panix.com / hieronym@desk.nl / etc.>

#  distributed via nettime-l : no commercial use without permission
#  <nettime> is a closed moderated mailinglist for net criticism,
#  collaborative text filtering and cultural politics of the nets
#  more info: majordomo@icf.de and "info nettime" in the msg body
#  URL: http://www.desk.nl/~nettime/  contact: nettime-owner@icf.de