Article Analysis: Can machines do art historY?

This article, "Can machines do art history?", details both the aspects where machines succeed, and where they fail in analyzing art. This type of AI uses image processing software, and in particular, a CNN (convolutional neural network). This type of software isolates specific groups of pixels in the image together. When trained, the algorithm can then recognize patterns and brushstrokes, then attribute paintings and styles to artists. However, it has been said that these algorithms should not be used for attribution. Whether they should never be used or are only currently inadequate is left to be seen. Algorithms can also only analyze objectively and visually, without any other background knowledge. There is also the subject that much of the algorithm's process is unknown; we do not know how it is able to reach conclusions. I think that while AI will inevitably grow stronger at attribution and visual analysis, there is no way for it to perform analysis on the meaning of a painting. AI relies on the visual aspects of a painting, and analysis requires other context and knowledge that the visuals do not cover. There will be aspects it excels in, however, ultimately there will also be tasks it can't do.

Brushstroke extraction on Van Gogh's 'Red Cabbages and Onions', preformed at Pennsylvania State University, 2012