“…Despite digital media’s obsession with newness, machine learning is largely about engagement with, and the extraction of value from, historical artefacts. According to Schneider (2018), the register of remains opens up to a response that ‘weaves past and future in intervallic resonance’ (p. 90) and creates a foundation for a ‘response-ability’ (Schneider and Ruprecht, 2017), in the sense both of calling ‘the past to appear for account’ and of being called by ‘the past to respond with account’ (p. 90). While seemingly straightforward, such an engagement often turns out to be complex, partly because ‘archives represent scenes of unbearable historical weight’ (Okwui, 2008) and partly because the methods by which we encounter these traces risk reproducing, in addition to subverting, the exposure of vulnerable archival subjects (Sutherland, 2017).…”