Page 37 - The Black Box Book

AI AS A NOOSCOPE
Pasquinelli and Joler highlight the dangers of accepting the common
rhetoric that AI is a magical tool capable of perfectly mimicking human
creativity and improving it in many ways. “The actual problem is the black
box rhetoric, which is closely tied to conspiracy theory sentiments in which
AI is an occult power that cannot be situated, known, or politically con-
trolled.” (Pasquinelli – Joler, 2021, p. 1265) According to them, contempo-
rary artificial intelligence is primarily a result of the automation of human
perception and, at the same time, a means of knowledge extractivism
and epistemic colonialism of human work and social behaviour through
their algorithmization. (Pasquinelli, 2019) To free the current debate on
the artificial intelligence of the haze of the mythical narrative of the mag-
ic box, they replaced the term AI or machine learning with Nooscope:
A I CU RA T I N G
“The Nooscope is a cartography of the limits of artificial intelligence,
The use of artificial intelligence in curating art collections has expanded intended as a provocation to both computer science and the hu-
considerably in recent years due to the systematic digitization of cultural manities. Any map is a partial perspective, a way to provoke debate.
heritage. In the multitude of such research and application projects, we Similarly, this map is a manifesto—of AI dissidents. Its main purpose
can identify two basic approaches, which can be characterized as archival is to challenge the mystifications of artificial intelligence. First, as
and curatorial-artistic. Artificial intelligence is gradually becoming a legit- a technical definition of intelligence and, second, as a political form
imate actor in digital online curatorial practice. that would be autonomous from society and the human.”
(Pasquinelli – Joler, 2021, p. 1263)
DATASET
A prerequisite for the use and development of current AI is the availability By renaming the AI as Nooscope, they are turning attention to AI as a tool
of datasets for training intelligent software: “Mass digitalisation, which for cognition and, therefore, control of our behaviour and aesthetic pref-
expanded with the internet in the 1990s and escalated with datacen- erences. “Nooscope is described as a machine that operates on three
tres in the 2000s, has made available vast resources of data that, for modalities: training, classification, and prediction. In more intuitive terms,
the first time in a history, are free and unregulated.” (Pasquinelli – Joler, 2021, these modalities can be called: pattern extraction, pattern recognition,
p. 1266). Developers of AI software focused on visual object recognition and pattern generation.” (Pasquinelli – Joler, 2021, p. 1268) To demystify
most often use ImageNet, which has been in development since 2010 AI, they dismantle the machine learning process into its fundamental
by a research team led by Fei-Fei Li. ImageNet is a database containing components, which they study separately and in coordination with each
14 million manually annotated images designed for learning AI software. other. These are “object[s] to be observed (training dataset), an instrument
To do this, the research team used a crowdsourcing method of work of observation (learning algorithm) and a final representation (statistical
aggregated through Amazon Mechanical Turk. The creators of ImageNet model).” (Pasquinelli – Joler, 2021, p. 1265) In addition, they emphasize
state that while they do not hold the copyright to the images contained the distortions, biases and simplifications that affect the output of ma-
in the dataset, they are making it available for non-commercial use by chine learning, whether these are limitations imposed by the limits of
research teams and for educational purposes: ImageNet does not own the technology used or constraints written into the output of AI work
the copyright of the images. ImageNet only compiles an accurate list of based on biases written into the datasets by the people involved in their
web images for each synset of WordNet. For researchers and educators creation. To illustrate the gradual distortion of information as an integral
who wish to use the images for non-commercial research and/or educa- part of machine perception and machine learning, they created a precise
tional purposes, we can provide access through our site under certain diagram that makes their argument comprehensible and applicable to
conditions and terms. (ImageNet) Pasquinelli and Joler, however, see various specific AI use cases (Joler – Pasquinelli, 2020).
the free-for-all database content, in the context of the current machine
learning boom, as the next phase in the evolution of the knowledge econ-
omy, which has transformed into a new forms of capitalism, referred to
as cognitive capitalism and surveillance capitalism. (Pasquinelli – Joler,
2020, p. 1266)
36 37