Artists on a Decade of AI Evolution: An Interview Study of Affordances, Culture, and Artistic Practice with Machine Learning

Paper Title

Artists on a Decade of AI Evolution: An Interview Study of Affordances, Culture, and Artistic Practice with Machine Learning

Publication Info

• Topic area: Evolution of artistic practices with machine learning over the past decade.
• Keywords: AI art, machine learning, generative AI, artistic practice, ethics, aesthetics, creativity, HCI, community, socio-technical shifts.

Background and Problem

• Problem / challenge: Existing research provides snapshots of AI-related art practices at specific moments or with heterogeneous AI technologies and artist populations. There is a lack of understanding of how early ML artists—those active before 2020—have experienced and adapted to socio-technical transformations in AI.
• Significance: Understanding these trajectories reveals how tools, cultural meanings, values, and artistic practices co-evolve, offering insights for HCI research on creativity and design.
• Motivation and related work: Prior work documented early ML-centered artistic practices as labor-intensive and exploratory, emphasizing crafting and unpredictability. More recent studies have examined generative AI's impact on creativity and artistic professions, highlighting opportunities and ethical challenges. However, the long-term evolution of early ML artists’ practices remains unexplored.

Solution

• Proposed approach: Thematic analysis of semi-structured interviews with 30 artists who began working with ML before 2020.
• Novelty:
  1. Provides a decade-spanning account of artistic practices with ML, showing how post-2020 shifts reconfigure aesthetics, labor, and ethics.
  2. Maps three divergent views on moral responsibility with large AI models.
  3. Extends accounts of genAI’s impact, showing harms to media artists with long-standing ML practices.
  4. Articulates three future orientations for ML-based artistic practice.
• Procedure and key techniques:
  • Recruitment of artists through online galleries and repositories.
  • Semi-structured interviews covering artistic practice, community, ethical perspectives, and future orientations.
  • Thematic analysis of 29 transcripts to identify patterns and themes.

Results

• Concrete findings:
  • Aesthetic narrowing in post-2020 ML systems due to strong text–image associations and cultural homogenization from AI imagery online.
  • Platformization and complexity of ML models reduce malleability but lower technical overhead.
  • Diverging ethical stances: models seen as inherently extractive, responsibility placed on users, or creative freedom prioritized over ethics.
  • Audience familiarity with consumer AI erodes the legibility of ML-based artistic practices, causing misinterpretation and suspicion.
  • Fragmentation and noise in artist communities post-2020, making meaningful exchanges harder to sustain.
• Advantage over baselines: Nuances prior accounts of ML artists’ ethical praxis and extends understanding of genAI’s impact on artistic professions beyond illustrators and “artists for hire.”
• Experiments / evaluation:
  • 30 artists interviewed, spanning varied practices, visibility, and locations.
  • Thematic analysis with 685 unique codes and 1651 data extracts.
• Limitations and future work:
  • Recruitment skewed toward English-speaking artists visible in institutional circuits.
  • Future work could explore broader artist populations and deepen understanding of ethical dilemmas in ML-based art.

Summary

This study investigates how early ML artists have experienced and adapted to socio-technical transformations in AI over the past decade. It finds that post-2020 ML systems narrow aesthetic possibilities, reduce malleability, and introduce ethical dilemmas, while also lowering technical barriers. Artists face eroded legibility of their practices and fragmented communities. Three orientations for future practice are identified: reorienting away from dominant AI logics, expanding toolsets, and maintaining continuity with established workflows. These insights contribute to HCI by highlighting the reconfiguration of crafting ethos and proposing implications for supporting low-level interventions within AI models and workflow composition across AI and non-AI pipelines.