Page 96 - The Black Box Book

The new archivist is the name given to an artificial neural network in
Figure 3: Image position allocation based on
dimensionality reduction of image parameters obtained by
a pre-trained neural network. Author: Štěpán Miklánek
U N LE A SH I N G A I
We named the curatorial experiment with the use of an artificial neural
network the New Archivist to identify the figure of the change of perspec-
tive and method of working with archival material, which is characterized
by the search for ways to go beyond the limitations and consequences
of anthropocentric interpretation of the world within the framework of
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
malevolent people say that he is the new representative of a struc-
tural technology or technocracy. (…) Some say that he is a shammer
who cannot back himself up with reference to the sacred texts,
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
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
and situations they were experiencing. This way, we were able to collect
approximately 1 000 images. In addition to extended documentation of
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
books read, views from the window, lists of favourite podcasts on mobile
phones, and snippets of Messenger communications. All images were
given the names of the artists and a brief description (name of the artist,
title of the work or identification of the image, date taken), but not for
the sake of the instructions for the neural network, but for our better ori-
entation in the output of the AI work, and their placement in the Archive
section of the website. (Figure 4)
1 GHAHRAMANI, Zoubin: Unsupervised Learning. BOUSQUET, Olivier,
Ulrike VON LUXBURG a Gunnar RÄTSCH, ed. Advanced Lectures on
Machine Learning [online]. Berlin, Heidelberg: Springer Berlin Heidelberg,
2004, s. 72-112 [cit. 2020-12-21]. Lecture Notes in Computer Science. ISBN
978-3-540-23122-6. Dostupné z: doi:10.1007/978-3-540-28650-9_5
97