Page 299 - The Black Box Book

C U R AT O R O F TA S T E various representations of fauna and flora. It also included a set of por-
traits and allegorical works reinforcing the emperor’s status, as well as
In a recent experiment, we demonstrated that neural networks can detect a number of paintings with erotic overtones. On the other hand, depictions
and categorize artistic techniques. However, could this technology be uti- of religious motifs were not often found in the imperial halls, although the
lized to classify images into less recognizable categories? For example, to “Rudolf artists” themselves also executed a number of commissions for
estimate its association with a particular art school, group of artists, exhi- church clients (Horská, 2018, p. 22).
bition, show, or individual collection? Is it possible to go even further and The painters (and artists in general) who worked for Rudolf II were
create an algorithm capable of replicating curatorial taste? If today’s gen- seasoned masters from many artistic centers in Europe who settled in
erative neural networks produce music, writings, or paintings based on Prague. The nature of art at Rudolf’s court was therefore initially very di-
databases of works by long-dead artists, why couldn’t we also resurrect verse. Thus, It is therefore only conceivable to speak of the “Rudolphine
legendary collectors like Lorenzo de’ Medici or Daniel-Henry Kahnweiler style” in the context of the period at the turn of the 16th and 17th centu-
and ask them whether they would select this or that piece to their collec- ries, when artists had been working in one place for a long time and were
tion? As an alternative, how about looking for paintings that are formally influencing each other (Zlatohlávek, 2012).
comparable to these collections and determining whether these similarities The sample dataset of Rudolphine works was derived from Thomas
point to any deeper connections between the pieces of art? DaCosta Kaufmann’s book The School of Prague: Painting at the Court
There is a tendency for major styles as well as some smaller ensem- of Rudolf II, which is based on contemporary inventories and archival
bles of works to have a similar aesthetic. When we speak of workshops, sources and is the most exhaustive source on painting collections of
schools or circuits in the visual arts, we frequently refer to formal simi- the time. The resulting database contained 328 paintings from 29 artists.
larities between individual artifacts. Other groups, however, were formed The range of artists included well-known names such as Hans von Aachen,
through a variety of external circumstances, and their composition can Giuseppe Arcimboldo, Bartolomeo Spranger and Joseph Heintz, as well
be visually quite disparate. Artists were admitted to aristocratic courts as lesser-known artists such as Daniel Fröschl, Paul Vredemann de Vries
on the basis of recommendations from masters or professors; they joined and Johann Geminger.
societies or groups on the basis of shared opinions and attitudes; they If we want to train a neural network to recognize a certain phenom-
may have been included in an exhibition through the efforts of gallerists enon, we must frequently prepare not just an example dataset containing
or art dealers, and so forth. The machine vision algorithm cannot take images of the phenomenon but also a negative set in which the observed
into consideration these (not only) social settings because they are not element does not exist. For example, if the goal is to create an algorithm
inherent in the visible layer of the artwork. The absence of context can that recognizes a cat, we need to collect and label photographs that
be both advantageous and disadvantageous, on the one hand, historical include the animal, as well as those that do not include the animal at
knowledge is omitted, but on the other hand, the works can be compared all. What exactly should a picture with a “non-cat” look like? Simply put,
with the “innocent eye” of a program unburdened by a predefined context. there should be a wide variety of photographs showing potential settings
As foreshadowed in the introduction, this experiment focuses on in which the cat might be present but is not. Adopting a similar approach,
the tastes recorded in the collections of Emperor Rudolf II. These col- we also require pieces by “non-Rudolphines,” or painters unrelated to
lections were intended to encompass the total diversity of the world at the Prague court, in order to develop a program that reflects the aesthetic
the turn of the 16th and 17th centuries and to be a representative manifes- preferences of the Rudolphine painting collections.
tation of the metaphorical concept of the Theatrum Mundi. The collections How to define these “non-Rudolfers”? If the taste of the imperial
included objects from every known continent, representing a wide range curators is a difficult-to-grasp abstract phenomenon, how can one pick
of human interests and diverse components of animate and inanimate works that do not include this phenomenon without committing biases
nature. They included minerals, semi-precious and precious stones, based on one’s own preconceptions and preferences? For the time being,
fossils of plants and animals, stuffed animals, skeletons, skins, antlers, it seems most prudent to identify as non-Rudolphine all works that were
bezoars, technical devices, clocks, astronomical instruments, historical made before Rudolf’s adulthood and hence could not have been influenced
artifacts, copies of ancient works, and contemporary (late Renaissance and by his collecting tastes at the time they were created. However, the nega-
Mannerist) works of art in the fields of glass, jewelry, sculpture, drawing, tive dataset did not include any works produced after the Emperor’s death.
printmaking, and painting. The rationale behind this is the belief that the themes, style, and taste
From the perspective of this experiment, the most noteworthy is of Rudolphine painters persisted long after the monarch’s death. Indeed,
Rudolf’s painting collection, which represents the source of paintings for finding works that remained influential subsequent to the Monarch’s reign
the sample dataset. This collection included allegorical and mythological was one of the aims of this experiment.
works, genre paintings, especially landscapes, still lifes with flowers and
298 299