Intuition Machines

Abstract


 
 
 The urgency of environmental, security, economic and political crises in the early twenty-first century has propelled the use of machine vision to aid human decision-making. These developments have led to strategies in which functions of human intuitive processing have been externalized to ‘vision machines’ in the hope of optimized and objective insights. I argue that we should approach these replacements of human nonconscious functions as ‘intuition machines.’ I apply this approach through a close reading of artworks which expose the hid- den labour required to train a machine. These artworks demonstrate how human agency shapes the ways that machines perceive the world and reveal how values and biases are hardcoded into nonconscious cognitive machine vision systems. Thus, my analysis suggests that decisions made by such systems cannot be considered fundamentally objective or true. Nevertheless, artworks also exemplify how externalized intuitive processing can still be helpful as long as we refrain from blindly taking the results as a go-signal to take immediate action.
 
 

Machine vision situations: Tracing distributed agency

“…The machine learning model then operates within these parameters. Hayles discusses how human non-conscious processes “feed forward intuitions to conscious awareness” ( 2017 :41); likewise, we can understand the technical non-conscious of machine learning models as generating a type of “technical intuitions” ( Kronman, 2020 ). In Mosaic Virus, technical intuitions are expressed as visual hallucinations of impossible tulips.…”

“…The data collected on digital games formed the basis for three studies: one on the use of surveillance cameras as an interface in digital games, proposing the term ‘cyborg vision’ to account for the experience of embodied surveillance that these games offer to the player ( Solberg, 2022a ); a second on how holograms mediate between human and non-human actors in games ( Solberg, 2021 ); and a third on enhanced vision in games and its relation to ideas of domination and power ( Solberg, 2022b ). The data collected on artworks formed the basis for Kronman’s (2023) analysis of different approaches to hacking machine vision in art, revealing how art is used to expose bias in machine vision, as well as papers on non-conscious cognition and agency ( Kronman, 2020 ), and aerial perspective and prediction in machine vision assemblages ( Kronman, 2019 ). Gunderson (2021) drew on the database to analyse the representation of augmented reality in popular culture.…”

“…While each of our papers grew out of the quantitative work that went into making the database, several have turned to qualitative, close readings of works and situations as their main analytical method ( Gunderson, 2021 ; Kronman, 2020 ; Kronman, 2023 ; Solberg, 2021 ; Solberg, 2022a ; Solberg, 2022b ). Based on this experience, this paper demonstrates how the concept of machine vision situation, and the formal data structure we developed for it, can serve as a tool to identify the compelling and relevant moments for analysis within a work, and function as a framework for not just quantitative, but also qualitative analysis of distributed agency.…”

The ethics and politics of data sets in the age of machine learning: deleting traces and encountering remains

“…Artistic research on data sets has played a crucial role in drawing attention to the politics of data sets and communicating this attention to a broader audience (Crawford and Paglen, 2019; Harvey and LaPlace, 2021a; Kronman, 2020; Onuoha, 2018; Pipkin, 2020; Schmitt, 2019). Computer scientists Buolamwini and Gebru (2018) explored the racial bias of data sets in the academic project ‘Gender shades’, which also became an exhibition at the Barbican.…”

Algorithmic failure as a humanities methodology: Machine learning's mispredictions identify rich cases for qualitative analysis

“…The failed predictions of machines let us use machine learning as a collaborator, using algorithmic failures as ‘technical intuitions (that) function as an interface between technical and human cognizers’ (Kronman 2020). To some extent, this aligns with human-in-the-loop approaches to machine learning.…”