Page 80 - AC-2-3

Arts & Communication Computer vision in tactical AI art

submission in technically mediated sociocultural relations.
This face recognition setup draws linear portraits of the
visitors, but to make it work, the sitter must keep their eyes
closed while resting their pen-holding hand on a small
platform that moves along the X/Y axes over a fixed piece
of paper. By playfully turning visitors into the “mechanical
parts” of a drawing apparatus and making them conscious of
both their acquiescence and slight functional unreliability,
Blind Self Portrait (knowingly or not) references William
Anastasi’s Subway Drawings from the 1960s. Similarly,
54
Patrick Tresset’s Human Studies (since 2011) is a series
55
of posing sessions in which humans become passive Figure 6. Ben Grosser, Computers Watching Movies (2013). Image
sitters for portrait-drawing robots that imitate Tresset’s courtesy of the artist
visual style. However, Tresset started developing robots
for Human Studies as creative prostheses to overcome into the widespread exploitation of transnational echelons
56
his artistic block, and the primary poetic feature of this of underpaid workers for AI development, which has been
non-tactical project is the novelty-induced anticipation of identified as “fauxtomation,” “sociotechnical blindness,”
59
46
finished machinic drawings that relieves visitors’ deference “ghost work,” and “human in the loop” complex. 61
60
and typical modeling boredom. As the pragmatic functions of CV and other machine
It is easy to forget that the control and command “reign” vision technologies, command and control are repurposed
of CV software routines is governed by a higher order of in these works to unveil the questionable or problematic
conditioning and routing algorithms. By intersecting this aspects of AI in the economy, culture, and various layers
computational hierarchy facet with cultural heritage, artists of everyday life, which originate in the higher-level
can point out the constraints of AI-influenced society. computational processes and human reasoning in their
57
For instance, in Computers Watching Movies (2013), design and implementation.
Ben Grosser combined the informative open-endedness
of abstract forms with cumulative cultural experience to 2.3. Biometric classification
engage visitors in an imaginative guessing game (Figure 6). CV has been extensively integrated into a pipeline for
The work features six video sequences produced by the CV extracting and combining different types of biometric data
analysis of popular film sequences. Points and vectors of to build formal categorization models that can be used for
the CV program’s “focal interest” (image locations assigned contact filtering, individual identification, verification,
with higher weights) are animated as colored dots and lines profiling, and determination of rights, such as entry
on a white background (the processed film footage is not privileges, service access or modification, and pricing.
visible) and synchronized with the original film sound. Potential areas of application include state governments,
The coupling of minimalist visuals with sonic guidance military, intelligence, health and wellness, HR, commerce,
draws viewers into a series of comparisons between their and entertainment industries. Modern AI processes
culturally developed ways of seeing and CV software’s biometric data in statistical prediction frameworks whose
“attention” logic, which has no historical, narrative, or reductiveness can be problematic even in hypothetical
emotional patterns. scenarios with completely unbiased classification models
Beyond their immediate topical concerns, these works because they ultimately always make implicit (but
connect with tactical art production that examines the unobjective) claims to represent their subjects. These issues
roles of CV in cybernetic systems designed for labor prompt different modes of tactical engagement with the roles
maximization, profit extraction, and other questionable of CV in biometric classification and taxonomic imaging.
forms of societal control. For instance, Matt Richardson’s For example, Jake Elwes’ video Machine Learning Porn
Descriptive Camera (2012) uses a novel form of taking (2016) plays with moral prejudices as well as perceptive
58
62
snapshots to indicate the socioeconomic background and technical flaws that influence AI design within the
of modern AI technologies. This camera has a lens, but context of online content “propriety.” Elwes took the open_
instead of displaying the photographed image, it sends nsfw convolutional neural network trained on Yahoo’s
the image to an Amazon MTurker tasked with writing model for detecting “sexually explicit” or “offensive”
down and uploading its brief description, which the device content and repurposed its recognition classifiers as
prints out. It offers a revelatory, counterintuitive glimpse parameters to generate new sequences of abstract images

Volume 2 Issue 3 (2024) 7 doi: 10.36922/ac.2282