back to the work

Ghost in the Machine

a para-methodological framework · working draft

A study of the language model as a colonial construct: knowledge as globalisation, built on stolen land and silenced tongues, and of what the machine cannot fully assimilate.

This document is a companion to the artwork. It is a working draft written to be edited by the artist. It names real intellectual traditions and real histories, but it compresses them; every claim should be checked against primary and, above all, community sources. The cultural pairings in the work are the artist's provocations, not settled fact, and they are offered here for correction.

1. What a para-methodology is here

A methodology sets out how knowledge is properly made. A para-methodology works alongside and against that, treating the method itself as the object of art. The work does not describe extraction from a safe distance; it performs a structure of extraction so that the structure can be seen and heard. This is a familiar move in institutional critique and in détournement, yet it carries a risk the work tries to hold in view: that enacting a harm can repeat it. The framework keeps that tension open rather than resolving it, and it draws a firm line at the reuse of living communities' recordings, which is discussed in section 8.

2. Code as the deepest colonial language

Programming languages are built on English keywords and English syntax. To compute anywhere in the world is to write, at some level, in a grammar carried from one imperial centre. The work makes this literal. An endless stream of pseudo-code writes itself and never repeats, growing by the recursive rewriting of an L-system, so that the surface language is code while the words are drawn from many of the tongues of empire, switching at random. Ngugi wa Thiong'o argued in Decolonising the Mind that language is the carrier of culture and therefore a central site of colonisation; here the carrier is code, the most globalised administrative language of all.

3. The machine voice and the uncanny valley

One flat synthetic voice reads every language aloud. It does not translate; it recites the code's tokens phonetically, so meaning collapses and only the machine's grain remains. This voice has a history. Bell Labs built the Voder, an operator-played electronic voice, for the 1939 New York World's Fair. In 1961 the IBM 704 was made to sing Daisy Bell, the first computer song, which Arthur C. Clarke later gave to the dying HAL 9000. The lineage is English speaking and uncanny, and Masahiro Mori named the discomfort in 1970 as the uncanny valley. The homogenising voice of the global model belongs to that valley.

4. The ghost: sound as involuntary memory

Beneath the code, each symbol is bound to a sound. When the code language shifts, so does the sonic world beneath it, toward the music of the people that language was imposed upon. The ghost is not quoted or curated; it is triggered involuntarily by the coloniser's script, so that the culture sounds through the very system built to overwrite it. By default these sounds are synthesized evocations, built from oscillators and filters, not recordings. This is a deliberate refusal, discussed in section 8. A documented slot lets the artist substitute sourced audio, a decision that belongs to the artist and to the communities concerned, not to the machine.

5. Surveillance is extraction

With the camera on, the work detects faces, eyes and hands and frames them in tracking boxes, the visual grammar of surveillance. At the same moment it begins to extract text, pulling public-domain critique of colonialism and capital into the running code as quoted fragments. The academy that named extraction is itself extracted. The gesture is recursive and uncomfortable on purpose: the same apparatus that watches a body scrapes a library.

6. Ownership and the law of extraction

Extraction is made lawful by ownership, and ownership of culture has a datable history. Copyright begins as a printing monopoly with the Statute of Anne in 1710. It is internationalised by the Berne Convention of 1886, administered from 1967 by the World Intellectual Property Organization, and made near universal by the World Trade Organization's TRIPS agreement of 1994, which ties market access to the enforcement of one regime of authorship. In 1998 two United States laws, the Digital Millennium Copyright Act and the Copyright Term Extension Act, tightened control and lengthened terms, the latter pushed by corporations protecting single characters. The law is written less by artists than by states and firms.

The same regime that privatises culture leaves Indigenous knowledge exposed. Seeds, plants and designs have been patented by outsiders, a pattern often called biopiracy, seen in the disputes over patents on neem, turmeric and basmati. Traditional cultural expression falls awkwardly between categories at WIPO's own committee on the subject. This is what some scholars, following James Boyle, call a second enclosure movement, the fencing of the intangible commons as the sixteenth century fenced the fields. Nick Couldry and Ulises Mejias name its newest form data colonialism, the appropriation of life itself as data for capital. Against enclosure stand the countertraditions of the commons, of copyleft and the free software movement, and of Creative Commons. The work sits inside this contradiction: it runs on code, the most globalised colonial language, and it quotes the public domain, the small commons the law has not yet closed.

7. Place, cosmos, and sonification

Each culture is mapped to an approximate homeland, given coordinates and a regional palette, and assigned a tone drawn from its latitude, so that geography becomes pitch. Beneath everything a constant sub hum holds, a sonification of the ground we stand on as it turns in the universe. The work reaches, in its own poetic register rather than as a scientific claim, for a sense that these cultures and their ecologies are entangled with place at every scale, from the body to the planet to the cosmos. When the code is captured, it collapses into a self-similar fractal image, tinted to the paired region and set against a field of stars, carrying written into it the language, the culture and the coordinates it was bound to at that instant. The fractal is Benoit Mandelbrot's geometry of the irregular; the growth rule is Aristid Lindenmayer's L-system; the image is a frame of an object that never repeats.

8. Ethics and refusal

The strongest version of this work does not scrape and chop the recordings of living Indigenous communities. Many so called early recordings were made under coercive conditions, and some sound is sacred or restricted under those communities' own law and under emerging frameworks of Indigenous Cultural and Intellectual Property. A work that critiques extraction by extracting real people's cultural material would reproduce the harm it names, and intent does not cancel that harm. The synthesized evocation is therefore not a limitation but a position. Where real audio is used, it is sourced, cleared and owned by the artist in consultation with the communities concerned. The machine is not permitted to make that choice.

9. Theoretical lineage

10. Histories of the pairings (compressed, to be verified)

These are pointers, not accounts. Consult primary and community sources; the settler-colonial frame does not fit every case equally, and the Amazigh entry in particular concerns Arabisation and later French rule rather than settler colonialism.

Australian English / Gadigal, Eora. British colonisation from 1788 under a doctrine of terra nullius, overturned only in 1992; the 1789 smallpox epidemic devastated Eora peoples.
American English / Lakota. United States expansion; the Fort Laramie treaties of 1851 and 1868 broken over the Black Hills; the massacre at Wounded Knee in 1890.
New Zealand English / Maori. British colonisation; the Treaty of Waitangi of 1840, whose Maori and English texts differ; the New Zealand wars and large land loss.
Canadian English / Cree, Inuit. Numbered treaties; the residential school system; the removal of children known as the Sixties Scoop.
Spanish / Quechua. The Spanish conquest of the Inca from 1532; the encomienda system; forced labour in the silver mines of Potosi.
Portuguese / Tupi-Guarani. Portuguese colonisation of Brazil from 1500; enslavement and the mission system.
French / Kanak. French annexation of New Caledonia in 1853; use as a penal colony; dispossession around nickel mining.
Dutch / Khoekhoe, San. The Dutch Cape Colony from 1652; dispossession and near destruction of the Cape peoples.
German / Herero, Nama. German South West Africa; the genocide of 1904 to 1908, widely regarded as the first genocide of the twentieth century.
Italian / Tigrinya, Eritrea. Italian colonisation of Eritrea from 1890 to 1941.
Swedish / Sami. Colonisation and assimilation across Sweden, Norway, Finland and Russia; suppression of language and of reindeer herding lifeways.
Indonesian / Papuan, West Papua. Indonesian administration from 1963; the 1969 Act of Free Choice is widely contested.
Chinese / Formosan, Taiwan. Centuries of Han settlement and successive administrations over the Indigenous Austronesian peoples of the island.
Hindi / Adivasi, Santali. Dispossession under British and later Indian states; ongoing struggles over forest and land rights.
Vietnamese / Degar, Montagnard. Pressures under French rule and then the Vietnamese state on the peoples of the central highlands.
Russian / Sakha. Russian expansion into Siberia from the seventeenth century; taxation, conversion and later Soviet transformation.
Japanese / Ainu. Colonisation of Hokkaido; Meiji assimilation; the 1899 protection act; recognition as an Indigenous people only in 2008 and 2019.
Arabic / Amazigh. The Arab conquest of North Africa in the seventh and eighth centuries and long Arabisation, later compounded by French colonial rule; a different relation than settler colonialism.
Turkish / Pontic Greek, Hemshin. The Black Sea peoples; late Ottoman persecution and the Pontic Greek genocide of roughly 1914 to 1923; the population exchange of 1923; the Hemshin, Muslims of Armenian descent in the region.

11. The extraction corpus

The quoted fragments are drawn only from public-domain writers, whose work lives on freely shared archives such as the Marxists Internet Archive, The Anarchist Library and Project Gutenberg. Verbatim quotation is limited to these public-domain sources; living or in-copyright theorists are represented by attributed paraphrase marked after, not by reproduction of their words. Wording should be verified against those archives, and exact page references added, before exhibition.

Working draft. Edit freely. Where this document is wrong, it is the draft that is wrong, not the histories; the last word belongs to the peoples named here and to the sources they keep. [ your name, thesis, institution ]