Absence and anxiety
An extract from an essay I wrote at Indiana University on connoisseurship, computation and the black box. A revised version, "Absence and Anxiety: Critical Reflections on AI", is in progress.
Why are we anxious about using tools such as computer vision algorithms that might be employed to detect patterns in pottery? What exactly causes us to be suspicious of a method that purports to recognise patterns and to find associations between images? If we are anxious, what kind of dialogue could we create to help alleviate that anxiety? In researching the life and methods of Sir John Beazley, I found that many of the issues common to connoisseurship accumulate equally around computer vision. Simply stated, there is an anxiety that surrounds the method itself and any work in proximity to it.
[…]
When cybernetic language was developing, mechanisms such as Norbert Wiener's anti-aircraft predictor also opened a new universe of black boxes. A black box can be described as a thing into which inputs go and out of which outputs stream. You can observe things before and after it but not trace them into the box. Such opacity is empty and full of absence. In this image, there is no human, only the machine.
Connoisseurship, likewise, has been criticised for being equally opaque. Connoisseurs like Morelli analysed the painting as a whole, and the method could often seem mysterious to the viewer. In other words, connoisseurship can be a black box. Within a construct that does not disclose its rules or how it obtains its results, how can we observe, let alone interpret, the past? Whether we accept or reject the ontology of connoisseurship or of computational methods, we find ourselves at the intersection of a robust set of cultural meanings. The anxiety is well founded, and the question remains: is there a way through the darkness?
[…]
One way is intellectual and programmatic transparency, which makes our methodological bias less absent. Another is to assault the black box directly and engage actively in the dialogue. One anxiety that several art historians have expressed is that computer vision might replace them. To this, I always reply that it is humans who build the models, humans who maintain them, and humans who interpret the results.
Much of the anxiety surrounding computational methodology stems from a lack of transparency about the ideologies, context and intellectual history that surround it. The stakes are high, and one of the primary dangers I foresee is that we will program ideas like connoisseurship. Ultimately, we do not want a difficult black box to revise and critique. We want open programs with designers who are transparent about the motives, biases and potential pitfalls of their tools. Art historians are some of the scholars best placed to engage thoughtfully with this discussion. Who better than an art historian to tell a programmer about their unconscious biases? I hope we can delve deep into the dialogue rather than only be anxious about it.
Works drawn on include Richard Neer, "Connoisseurship and the Stakes of Style" (2005); Michael Ann Holly, "Mourning and Method" (2002); and Peter Galison, "The Ontology of the Enemy" (1994) and "Computer Simulations and the Trading Zone" (2011).