Woven All of Dream and Error began through a process of walking. Over the course of four years, Kata Kovács and Tom O’Doherty undertook a series of walks along the paths of now-disused railway routes in Berlin and Brandenburg, which they filmed as they walked. In these walks, they also carried portable speakers, playing the sound of machine-learning-generated audio of invented trains: sound that has been hallucinated by contemporary computational technology. These films are explorations of traces left behind, cinematic studies of incisions in the land.

— Harley Aussoleil · From the book Woven All of Dream and Error: Hallucinations and Remnants (Estovers Editions, 2024)

The routes along which we walked, in the course of our research, span almost two centuries of use (and are outlined in greater detail below).

The impetus to undertake these walks emerged in part from a curiosity about the “deep history” of technology — the points of connection of the tools that humans use, past and present. In the nineteenth century, the social and industrial innovation of railway infrastructure was the radical cutting-edge of human technology. However, it has had the time, since then, not only to become mature, but also to develop layers of ruins, disused remnants, and buried vestiges, both physical and social.

In writing about the history of technology, in 1999, the philosopher Paúl Virilio quipped that “the invention of the ship was also the invention of the shipwreck” — by which he meant that all technologies not only become more obsolete over time, but that they also contain hints of their own future state of ruin, disuse, and abandonment.

At the moment, machine learning — usually referred to as artificial intelligence — is in the process of steadily becoming ubiquitous. This innovation has come to represent and exemplify the forefront of current computational possibilities. At the same time, these technologies contain their own inherent biases and flaws. The possible future integral “shipwrecks“ that these capacities possess are still largely unknown.

Woven All of Dream and Error creates a tangibly human space and offers me tactile blessings: sound that I do not get tired of being inside and around. In many ways, this music is a dialectical triumph because it is so sweet and mortal and yet suppresses many of its human traces. Tom O’Doherty’s guitar and bass parts hum and sustain and warp through various boxes, and Kata Kovács’s trumpet sails around inside digital echo, all of it lying between the train noises produced by machine learning. The tracks and wheels you hear are industrial vehicles that don’t exist but instead were generated by mulching the recordings of existing trains the artists never saw. The technology of the locomotive was part of the 20th century’s military success story, and AI—genocide’s new best friend—is central to the current military investment boom. Using the latter to recreate the former is a quiet way of melting the guns, so to speak.

—Sasha Frere-Jones, reviewing Woven All of Dream and Error

Woven All of Dream and Error was released by Soft Noise Acousmatics, in collaboration with Estovers Editions, in June 2025. It is a fourteen-track album that combines the sounds of machine-learning-generated hallucinated trains with guitar, bass, and trumpet, drawing from the lineage of drone, dub, and ambient music. The album has been released as a download, and also as a series of fourteen lathe-cut records (one for each track), which were cut by Kris Slyka on the SL-24 record lathe.

The lead track from the album, ‘Dream 1,’ has an accompanying video (below) showing excerpts from the films that were the result of our research.

Links:

• Woven All of Dream and Error on Bandcamp
• Soft Noise Acousmatics
• Estovers Editions
• SL-24 record lathe

Woven All of Dream and Error was an exhibition, presented at Hošek Contemporary, Berlin, in September 2024.

The exhibition also included the discussion event German Weapons, German Silence, with featured guest Dimitra Andritsou from Forensic Architecture. This event discussed German weapons exports to Israel, German cultural complicity in the Gaza genocide, the emergence of AI weaponry, and the overlaps, both technological and symbolic, between the vestiges of past technological eras represented by abandoned railway routes and an emerging era of automated genocide.

The works presented were:

• And but one sureness in our truth may lie (Part 1), and And but one sureness in our truth may lie (Part 2)
  Set of two single-channel videos, each 40 mins, dimensions variable
• We know it not by knowing it thereby
  Single-channel video, 29 mins, dimensions variable
• Walk 1 (Anhalter Bahnhof); Walk 2 (Angermünde); Walk 3 (M-Bahn); Walk 4 (Brommystraße); Walk 5 (Tempelhof); Walk 6 (Johannisthal); and Walk 7 (Tunnelbahn)
  Set of seven film still grids, on 300g paper, each 44cm × 44cm
• Dream 1–Dream 14
  Set of fourteen custom lathe-cut records, each 21cm × 21cm

Also initially presented at the exhibition was the accompanying book, Woven All of Dream and Error: Hallucinations and Remnants by Harley Aussoleil (Estovers Editions, 2024).

Routes that were part of Woven All of Dream and Error:

• Anhalter Bahnhof, Berlin 
  Dates of active use: 1841–1952
• Angermünde–Bad Freienwalde regional line, Brandenburg
  Dates of active use: 1877–1997
  Regional branch line, in use from the era of the German Empire through to the aftermath of the Wende.
• M-Bahn, Berlin
  Dates of active use: July 1991
  One of the world’s earliest magnetic-levitation train routes, under construction for most of the 1980s but ultimately only in public use for two weeks in 1991. Read more »
• Brommystrasse, Berlin
  Dates of active use: 1850–1871
• Tempelhof, Berlin 
  Dates of active use: 1923–1954
• Johannisthal, Berlin 
  Dates of active use: 1909–1952
• Stralau–Treptow Spreetunnel, Berlin
  Dates of active use: 1899–1932
  Formerly connected Stralau to Alt-Treptow through a tram line going under the river Spree. The tunnel still exists, but the entrances on both sides are filled in and the tunnel is flooded. Our initial research involved retracing the route by rowboat.

The audio that is used in Woven All of Dream and Error is the result of training a machine-learning model, and is generated via TensorFlow running on Colab.