Matthew Fuller
Break the Law of Information: Notes on Search Engines
and Natural Selection//2003
BackUp
There is no permanent back-up device for the Internet. A number of institutions
have made one-off synchronous copies of the Web that claim to be complete for
the duration of the period it takes their spiders to reach the furthest corners of
the Web.' There are also systems which actively monitor every email or
newsgroup message sent on behalf of the odder fans of political action and
dissolution.2 But of course both of these are partial. As what Gregory Ulmer has
usefullycalled 'a pre-broken system', the net has no centralized back-up. Copies
of elementsof the net are backed up across thousands of storage devices off and
across itself.Were the whole lot to go down, putting back together the results of
such disgraceful degradation would be an insane task.
Inevitably,some large areas of the Web would slide back into place with the
efficiencyof a file amongst filingcabinets. Others however - made on the fly,left
without a double - would go missing. Still more would have their links tangled,
lost - a mess of stringy,gloopy,gone data.
Natural Selection3 is in a sense the result of this crash before it happens. It is
the Web, picked up in pieces by people who never made it, put back together in
a way that makes more sense. It is obvious that the BritishAirportsAuthoritysite
would 'want' to link to information about people getting killed by antiimmigration police.4 It is obvious that someone looking for information on
fertility treatments would want to find a site mixing the histories of different
acts of racial 'improvement' into a corporate brochure. And it is obvious that
anyone looking for traces of overtly racialized political activism on the net is
going to want their browser to become infested by a fast-breeding Jamaican
virus in the guise of a triple-consciousnessed poet.5
BlackBox
A search engine is a black box. That is to say, the relationship between its input
and output is masked. The tension between making a search engine usable,
predictable and 'refinable', and the commercial necessity of maintaining its
constituent algorithms and processes as proprietary information, provides one
of the key contexts for this project.
Piracy of any electronic technology is usually carried out by reverse
engineering: feeding it inputs and monitoring its output until a tabulation of its
workings can be drawn up. The black box is not actually cracked open and
Fuller//Break the Law of lnfonnation//127
scrutinized, but is fed line after line until.it can be simulated. What is true of the
search engine is also true of the user. When you use HotBot,for instance, your
search-string and all the Linksyou make from the results to your enquiry are
logged and recorded aJong with your domain and LPaddress.
This information is useful for many reasons, and it becomes increasinglyso
over time with repeated use. It allows the more accurate targeting of search
results to specific enquiry entries (according to frequency of choice) and
advertising (according to proportional interest from certain domains). Any
impression that engines are there to provide a useful service financed by
incidental advertising is of course pure 'company history', as Yahoocalls it.
Just as the engine helpfully provides a field in which you can type in your
search-string, the engine is also opening up an aperture through which the user
can be interrogated. There are always at least two black boxes in this equation.
The interface is where they meet and engineer each other.
Natural Selectionsits on top of any of the mainstream search engines, filtering
their results and, if certain strings are fed to it, slipping in a result from the
bottom of the deck.The work is composed by making an interface to an interface.
The user comes to the engine and presents a search string, a history of others and
an Internet location. The composition is driven by the lure of access to data. As
they come together, the various elements, devices and drives produce a doubleblinded bait-and-switch.
It is this context, of the specific rules of formation, of the grammars of
information put into play by various search engines - that Natural Selection
operates in, and which this story hopes to open up a little. I... )
Instructions
In search engines using automatic indexing to produce inverted files - the
number of occurrences of a word related to the number of occurrences of all
words in the file- an apparent transparency is generated by the simple procedure
of pulling the words directly from the sites that they log. What is important,
what is notable, appears simply by virtue of being logged by the crawler.
Inevitably, this flat perspective produces an outcry about quality control. For
information science,'it is entirely possiblethat vast concatenations of information
resources are just not susceptible to any really effective measure of control and
access, at least not by any automatic procedures applied on a large scale'.6 The
Web challenges not only the status of the document, the corpus, but also the
index. This challenge is the result of an attempt to refuse, or at least avoid, the
burden of classificationbuilt into the architecture of the Web.
The UniformResourceLocator,(URL)of the World Wide Web is an attempt to
'point to any document (or any other type of resource) in the universe of
128//INFORMATION'S MODALITIES
informarion' 7 without a priori specification of the resource within any syste1nor
classification. Within this schema. anything that is digltizable can be stored,
networked and located. The architecture of the WWW is an attempt to route
round classification. or at least provide a situation where many such
interpretative formations can operate without universalization. This is
embedded nor just in the device of the URLbut also in details such as the simple
geek idealism of physical and logical styles, (where. in the latter. the site
designer can specify that the individually determined settings of the user's
browser can determine the way to display certain forms of text emphasis). The
WWW attempts to produce a 'preconceprual'8 technology that - in a similar
way to a computer's memory, storing both the data it usesand the programs the
data is used with - produces a field within which concepts coexist alongside the
rules to which the field is subjected.
Elsewherethere has been discussion of how particular modes of use of HTML
have tended to lock down the potential of the network into an idealized form of
paper-based media.9 Here, though. I hope to move towards a consideration of
how the rules or formation that operate in the construction or search engines
produce many particular overt or anonymous ways of making and using the
Web. It is not suggested that these are secret rules for the confabulation of the
world beyond the magic gate internet into something that it is not. nor are they
necessarily n1erelyillusions. misinformation. bad habits of thought. but a series
of perceprual devices which are themselves reliant on and interwoven with
others - and which are as such open to reinvention.
The Eating
[ ... I The most simple method of constructing a semantic network is to use a
precodified arrangement of terms and their correlates - using a document like
an expanded thesaurus to widen and clarify search possibilities•0 by aiding query
formulation through synonym relations between words and hierarchical and
other relations between concepts. Whilst top-down, 'hard' semantic networks
come into being on the basis of automatic word-sense disambiguation - the
syntactical machine that allows and disallows conjugation with what word, with
what it means - they aggravate polysemy in the sense that all 'dictionaries
advertise semantic discrepancies'.''which allow other diachronic and synchronic
meanings,other takes to leak through. However,hard semantic networks are like
dictionaries again. in that even if effectively and constantly enriched by
lexicographers,they still only really compose logs or past signifiance.
At the same time. whilst their hierarchical organization allows for an
extremely coherent co-ordination of words from a single perspective, they are
unable to negotiate the variable status of words as they are used.Allegory, irony,
Fuller//Break
the Law of Intonnattonj
/129
metaphor. repurposing - the rich meshwork of language that mitigates against
its petrification - all go like woodcutters amongst the trees in the database,
whilst utterly incommensurable figurations of language such as sacredness
throw up errors. not just of computation. but of the essencethey pretend to.
One way of attempting to avoid disjuncture between the semantic model of
the information referenced by the search engine and that of its users, in the way
commonly imposed by hierarchically organized systems. is to assume that the
words that are most commonly found together are actually related in some way.
This is a technique that makes best use of machine rather than human
interpretation of data. In Latent Semantic lndexing,' 2 a set of documents is
represented by a matrix whose entries are measures of frequencies of
occurrences of terms in each document. The greater the frequency of cooccurrence, the greater the relevance. This matrix is then approximated by a
sum of rank-one matrices determined using a singular value or related
decomposition. This allows for great compression of the data and fast, relatively
robust and encompassing referencing or sites. However. what is found to be
most related in terms of quantity doesn't necessarily match for each user. or
each search. Latent Sen1anticIndexing finds a way of working the inverted file
- mined with homonyms. synonyms and routes to a thousand pointless choice
trees - that plays to the data processingstrengths of computers, and thus avoids
the implicitly ontological nature of directories. However, these techniques are
always caught in a self-organized conformism ro the homeostatic pressuresof
what is statistically deemed, by frequency of use and by frequency of cooccurrence. to be the most relevant.
Searching for the most relevant, the most related - simply rendered as the
most linked to - is another technique adopted by a variety of search engines in
compiling their results from the inverted field. AltaVista, Excite, lnfoseek,
lnktomi 13 rate sites higher in their list of search results if a substantial number of
sites link to it. Google, usefully. makes this processvisible and useable.allowing
backlinking from its results. Relatedto 'citation analysis' - the scheme whereby
the relative importance of scientific papers is accorded by the frequency of their
being cited by other scientific papers - is another approach adopted in different
ways both by the Web Stalker•• and by the Clever15 project at IBM: ·we have
developed a new kind of search engine that exploits one of the Web's most
valuable resources- its myriad hyperlinks. By analysing these connections. our
system autom.atically locates two types of pages: authorities and hubs. The
former are deemed to be the best sources of information on a particular topic:
the latter are collections of links to those Jocations'.16 'Authorities· are sites that
are linked to many times; 'hubs' are sites that contain many links.17
Despite their claims, Clever has no way of actually defining a topic other than
130//INFORMATION'S MODALITIES
by resorting to imposinghierarchicaltaxonomies.What it does do is -by defining
a limited set of properties (links-to and links-from, as a way of organizing an
interpretation of the topologicalform of the Web) - use some of ways the Web is
actually put together to develop a representation of the en1ergent,unpredictable
interplay of multipolargravitational tendencies within the hyperlinkedstructure
of the Web.Someof these may be sufficientlycloselyclustered to be interpretable
fromspecificperspectives as a topic. Other sites which might be highly pertinent
from other perspectives might not link at all. What Clever has usefully done,
though. is to distinguish the structure of the Web from its content and use it as
an organizational principle. What they collapse in their descriptions of the
project is the distinction between the hierarchical pretence to objectivity and
what they actually represent, which is an aggregate of subjective decisions by
the thousands of people making sites to incorporate hyperlinks.Here,'objectivity
is less about realism than it is about intersubjectiviry'.18 'Self-organization'as a
phrase may smell like anarchist democratics. but it is in the choice of which
elements register as the evidence for, or as the most consequential part of, the
self-organization of this data that the political choice - closed ro scrutiny or
reinvention - as to what is important on the Web or about the Web is made. It is
a symptom of a self-reinforcing 'misplaced concreteness' that the mode of
interpretation of processes becomes the means by which they are moulded.
Goto a search engine of either sort and type in 'Jamaica'.You'llget information
on holidays coming up first - along with banner adverts for Sandals resorts.
Type in 'Africa· and you'll get predominantly wildlife and famine. Selforganization of data is organized on the basis of what ·setr is determined to be
in1portant.This, ar present, is something put into place by the demographics of
internet use. How data is interpreted and processed, how the grammar sorts
and orders the strings of symbols (whether they are hierarchically ordered. or
ordered according to emergent compositions of specified elements) has
immense importance. When most of this work is done by a closed culture of
proprietary specialists. finding ways in which this process can be opened up to
speculation and organisation by other forms of 'setr becomes even more
imperative - finding ways in which this process can be opened up to speculation
and organization by other forms of ·setr even more so. !... )
racist or have radalized connotationsare enreredas searchstrings. the NacuralSelectionsearch
enginereturns resultsthat. while Innocent-looking.drop the userinto oneof the overthirty other
siteson and off the mongreldomain.Collil.borators
in producingthe work includeHakim Bey.Byju,
Critical Art Ensemble,MervinJarman.RichardPierre-Davis,DimelaYekwai.
4
(SJ http:Jlwww.mongrel.org.uk/Projecr/Narural/BAA/hearhrow/index.html,
a site producedby
MervinJarman
5
161htrp:/fwww.mongrel.org.uk/Project/Narural/Venus/.
a sire produced in collaboration with
Dlmela Yekwai
6
1231FrederickW. Lancaster.Indexingand Abstractingin 11reoryand Pracrice,2nd ed. (London:
LondonLibrary Association.1998)313.
7
vol. 29. no. 10(October1996)
[241Tim BemersLee,'www: Past,Presentand Future·.Compucer.
69-77.
8
[251 Michel Foucault, TheAn:haeololl)Iof Knowledge,trans. A.M. Sheridan-Smith(London:
Tavistock,1972)62.
9
(26) See'A Meansof Mutation. Noteson 1/0/0 4: The WebStalker'.
10 (29) VisualThesaurus.a goodexample of a visuali2,1tionof this process,can be seenIn all its
ludicrously rigid beauty at http://www.thinkmap.com. This project is an Interface to the
WordNetlexicaldatabaseat http:/Jwww.cogsci.princeton.edu/-wn/.WordNetoperatesbasically
by constructinghierarchicallyorderedsetsof synonyms.SeeChristianeFellbaum.ed.. Wore/Net:
An ElecrronicLexicalDatabase(Cambridge.Massachuseru:The MIT Press.1998).
of cheWord (Londonand New York:
11 1301Walter J.Ong.Omliry and Uceraey:TheTechnologizing
Routledge.1982)46,
12 Ill JOn l..'ltentSemanticIndexing.seehttp://superbook.bellcore.com/-remde/lsi/
13 1321http:/fwww.searchenginewatch.com- a recommendedsite for a basic introduction to
searchengines
14 (331 http://bak.spc.org/iod. The Web Scolker.by contrast avoids these problems by its
representationof the network as a flat aggregateof Linksand nodescomposingtrself on the
screen rather than retranslaring It into an ordered list. Toe Web Sralker does not assign
importanceor greaterdegreeof relevanceto sites, merely relayingthe greater number of links
to them by Increasingluminosity of the node. Clever is more sophisticated in rerms of irs
mapping.however.simply by the analytical tools it is able 10bring to bearon the information.
15 (341http:/Jwww.almaden.ibm.com/cs/k53/clever.html
16 (35] 'Hypersearchlngthe Web:The CleverProject',ScientificAmericanUune 1999).
17 (361J. Kleinberg,'Authoritative Sourcesin a Hyper1inkedEnvironment'.Proceedingsof ACM·
1
Jfoomote2 in soureefTherearealsoarchivesof sectionsof the net - the searchengineDejaNews
beganan archive of all Usenet posrs going back to irs inception in 1979.This archive was
SIAMSymposiumon DiscreteAlgorithms ( 1998).
18 (37) Donna Haraway,Modest.Wirness@SeconcLMillenium.
FemaleManC>.Meers.OncoMouse
(Londonand NewYork: Routledge,1997)198.
subsequentlyiaken on by Google.
2
J3fSeeinfo on the Echelonsystemat h1tp://caq.com/CAQJCAQJJ.,cklssues.htm
3
(41This project coordinatedfor Mongrel by Harwoodand Matthew Fuller.essentiallypresenrs
(New York:
Manbcw Fuller. exrracts from Behind che Blip: Essays011 1he Cu/cureof S-Of1ware
lrself asa Straightsearchengine.However.when any of severalthousandwords that are directly
Au1onomedia.2003) 69-71. 80-81. 84-87.
Fuller//Break
the Law of Information//
131
132//INFORMATION'S MODALlTIES