Page 288 - The Black Box Book

production, but only an incomplete sample representing past and present
curatorial preferences. In addition to this bias, the dataset also reflected
the (un)willingness of individual institutions to share their collections, their
priorities in digitisation, and the effects of time. It is a well-known fact that
the deeper into history our interest goes, the fewer artifacts we have avail-
able. But how great is this disparity? Just 1.3% of the Collection’s assets,
as determined by the Digital Curator dataset, are made up of all works
of art that date back to 1500 or before. Respectively, the database has
36,658 pieces from the first half of the 20th century compared to just
314 from the first half of the 15th century.
Looking at these numbers and the accompanying graph (Figure 1),
we can easily see how much the amount of artifacts varies between peri-
ods. This variation causes a partial problem when comparing phenomena
across periods. In order to track their decline, stagnation or growth, it is
necessary to harmonize the values and use proportional representation.
Thus, when the Digital Curator makes a chronological comparison, he is not
interested in the absolute number of artifacts with a given characteristic,
but in their share of the total number of objects in a particular period. For
example, the following histogram comparing the frequency of depictions
of men and women across history shows that around 1400, the machine Figure 1: Composition of the Digital Curator database by year of exhibit creation.
vision algorithm recognized a woman in almost 60% of the images, and The horizontal axis shows the median year of (presumed) creation of the work, the vertical
the opposite sex was present in about 25% of the works (Figure 2). axis the number of works in the database. Author: Lukáš Pilka
For this period, the sample examined consisted of only about three hun-
dred works, whereas for the period around 1900 it already contained tens
of thousands of images.

Figure 2: Proportion of men and women depicted across the centuries. The chart
shows the proportion of images in which computer vision has identified a male and female
figure. Author: Lukáš Pilka
288 289