Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Machine Learning Processes as Sources of Ambiguity: Insights from AI Art

Christian Sivertsen, Guido Salimbeni, Anders Sundnes Løvlie, Steve Benford, Jichen Zhu

TL;DR

The finding indicates that the current conceptualization of ML as a design material needs to reframe the ML process as design elements, instead of technical details, and offers reflections on commonly held assumptions in HCI about ML uncertainty, dependability, and explainability.

Abstract

Ongoing efforts to turn Machine Learning (ML) into a design material have encountered limited success. This paper examines the burgeoning area of AI art to understand how artists incorporate ML in their creative work. Drawing upon related HCI theories, we investigate how artists create ambiguity by analyzing nine AI artworks that use computer vision and image synthesis. Our analysis shows that, in addition to the established types of ambiguity, artists worked closely with the ML process (dataset curation, model training, and application) and developed various techniques to evoke the ambiguity of processes. Our finding indicates that the current conceptualization of ML as a design material needs to reframe the ML process as design elements, instead of technical details. Finally, this paper offers reflections on commonly held assumptions in HCI about ML uncertainty, dependability, and explainability, and advocates to supplement the artifact-centered design perspective of ML with a process-centered one.

Machine Learning Processes as Sources of Ambiguity: Insights from AI Art

TL;DR

The finding indicates that the current conceptualization of ML as a design material needs to reframe the ML process as design elements, instead of technical details, and offers reflections on commonly held assumptions in HCI about ML uncertainty, dependability, and explainability.

Abstract

Ongoing efforts to turn Machine Learning (ML) into a design material have encountered limited success. This paper examines the burgeoning area of AI art to understand how artists incorporate ML in their creative work. Drawing upon related HCI theories, we investigate how artists create ambiguity by analyzing nine AI artworks that use computer vision and image synthesis. Our analysis shows that, in addition to the established types of ambiguity, artists worked closely with the ML process (dataset curation, model training, and application) and developed various techniques to evoke the ambiguity of processes. Our finding indicates that the current conceptualization of ML as a design material needs to reframe the ML process as design elements, instead of technical details. Finally, this paper offers reflections on commonly held assumptions in HCI about ML uncertainty, dependability, and explainability, and advocates to supplement the artifact-centered design perspective of ML with a process-centered one.
Paper Structure (30 sections, 4 figures, 1 table)

This paper contains 30 sections, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Theoretical Framework
  3. Ambiguity in HCI
  4. Ambiguity in AI Art Theories
  5. Uncertainty in the Machine Learning Process
  6. Approach
  7. Sample AI Art
  8. Results
  9. Dataset Curation
  10. Selecting existing datasets with questionable ontologies
  11. Making bespoke ontologies through new datasets
  12. Model Training
  13. Repurposing upscaling to introduce visual artifacts
  14. Under-fitting ML models
  15. Changing the output modality from the input modality
  16. ...and 15 more sections

Figures (4)

  • Figure 1: A Typical Machine Learning Pipeline for Computer Vision and Image Synthesis
  • Figure 2: From top left to bottom: Butcher's Sonklingemann_butchers_2017, Machine Biasyaszdani_machine_2018 and Learning to Seeakten_learning_2019
  • Figure 3: From top to bottom: Biometric Mirrorwouters_biometric_2019,POSTcard Landscapes from Lanzaroteguljajeva_postcard_2022 and ImageNet Roulettepaglen_imagenet_2019
  • Figure 4: From top to bottom: Poisonmunch_poison_2021, Unsupervisedanadol_unsupervised_2022 and in transituada_ada_ada_in_transitu_2022