There’s a drive in modern culture to collect culture, create definitive collections of specific historical artefacts, assemble a complete historical record of past events. That’s the drive of a museum or an archive. There’s an equal drive to create idiosyncratic collections reflecting networks of relata, readers, readings, resemblances, reuses and recontextualizations of expressions, forms and symbols. That’s a cabinet of curiosities, a Mnemosyne Atlas, a UbuWeb of one sort or another.
These procedures of collection ground, situate and sometimes blind our understanding of history. However, the collections they create cannot be comprehended by attention we muster in our media diet and do not fall under the strictures of attention economy. They rather require methods, even if idiosyncratic ones. They require the capacity to study, to draw out what lies in hiding between the artefacts.
Machine learning algorithms are trained to recognize patterns in large sets of data. They are engineered to draw out what lies in hiding. Thus, they’ve been trained in cultural collections. They are extracting potentialities that lie in the historical records and converting them into generated material, with their developers and owners seeking to find applications – and sometimes generate value under the strictures of the attention economy.