Haunting
Haunting refers to the possible return of that which resists translation into computational form in AI systems.
One of the main functions of algorithmic architectures is to ‘render calculable some things that hitherto appeared intractable to calculation’ (McKenzie, 2017: 8) – dynamics which encapsulate what the digital media scholar Ed Finn (2015) calls the ‘computational imperative’: a wager that all complex systems could be modelled quantitatively that finds its roots in Alan Turing’s Universal Machine.
While management psychology in the 1970s and 1980s focused on how intuition, as a human capacity, could be measured and indexed (Lussier, 2016; Pedwell, 2023), artificial intuition deals only with what is legible in computational terms and discards everything else.
In the course of the algorithmic operations that constitute “artificial intuition”, then, something is inevitably elided, lost, or repressed – there is always a remainder which resists translation into computational form and therefore “haunts” the system.
We might then consider the afterlife of that which is elided or repressed within “intuitive” computational systems. If that which machine learning algorithms ‘leave behind reside[s] uneasily in limbo, known and unknown, understood and forgotten at the same time’ (Finn, 2015: 51), under what conditions might such elements return and with what critical implications?
Such questions resonate with the political geographer Louise Amoore’s call for an experimental and processual approach to AI ethics which ‘involves reopening the multiplicity of the algorithm’ to reinstate ‘the partial, contingent and incomplete character of all algorithmic forms of calculation’ (2020: 162, 21).
They also align with the media psychologist Lisa Blackman’s exploration of how the ‘queer aggregations’ of haunted data can be ‘mined, poached, and put to work in newly emergent contexts and settings’ (2019: xiii).