Page 97 - The Black Box Book

The new archivist is the name given to an artificial neural network in
the role of a curator who deals with the contents of a black box without
knowing the code in which it is written and unaware of the situation in
which it was created. We wondered what leitmotifs the AI would recog-
nize in the visual evidence of our experience of the Covid-19 pandemic if
we kept inputs in the form of parameter specifications and objects to be
searched for to a minimum. Therefore, we have chosen an unsupervised
learning method (unsupervised learning).1 Such an intelligent tool resem-
bles a linguist who has discovered an unknown language or an extrater-
restrial (Alien) who has picked up unidentified signals from planet Earth.
Like them, it searches the unstructured cluster of particles for repeating
symbols and patterns of their arrangement to decipher the sign system
and reconstruct the meaning of the box’s contents.

DATASET
The primary condition for the effective use of artificial neural networks is
the preparation of a sufficiently large dataset. In our case, however, we
had only a few dozen text and image records of selected art projects at
the beginning of the project. Therefore, we again contacted the artists
to get them involved and asked them to send us additional documentary
material on their projects, but also images capturing their everyday life
during the quarantine restrictions, such as screenshots of text messag-
es from mobile phones or other visual signs of mediated communication
Figure 3: Image position allocation based on and situations they were experiencing. This way, we were able to collect
dimensionality reduction of image parameters obtained by approximately 1 000 images. In addition to extended documentation of
a pre-trained neural network. Author: Štěpán Miklánek
selected art projects, we collected images capturing secret parties of
students stuck in dorms and student-shared apartments, scenes from
the home environment of young families with young children, photos from
trips to the countryside, selfies from family car trips, as well as stacks of
U N LE A SH I N G A I books read, views from the window, lists of favourite podcasts on mobile
phones, and snippets of Messenger communications. All images were
We named the curatorial experiment with the use of an artificial neural given the names of the artists and a brief description (name of the artist,
network the New Archivist to identify the figure of the change of perspec- title of the work or identification of the image, date taken), but not for
tive and method of working with archival material, which is characterized the sake of the instructions for the neural network, but for our better ori-
by the search for ways to go beyond the limitations and consequences entation in the output of the AI work, and their placement in the Archive
of anthropocentric interpretation of the world within the framework of section of the website. (Figure 4)
knowledge of the world.

“A new archivist has been appointed. But has anyone actually ap-
pointed him? Is he not rather acting on his own instructions? Certain 1 GHAHRAMANI, Zoubin: Unsupervised Learning. BOUSQUET, Olivier,
Ulrike VON LUXBURG a Gunnar RÄTSCH, ed. Advanced Lectures on
malevolent people say that he is the new representative of a struc- Machine Learning [online]. Berlin, Heidelberg: Springer Berlin Heidelberg,
tural technology or technocracy. (…) Some say that he is a shammer 2004, s. 72-112 [cit. 2020-12-21]. Lecture Notes in Computer Science. ISBN
who cannot back himself up with reference to the sacred texts, 978-3-540-23122-6. Dostupné z: doi:10.1007/978-3-540-28650-9_5
and who seldom quotes the great philosophers. Others, though,
claim that something radically new has appeared in philosophy, and
that this work is as beautiful as those it challenges. It celebrates
the dawn of a new age.” (Delleuze, 1999, p. 3)
96 97