ARTFEED — Contemporary Art Intelligence

Artists Build AI Consent Systems to Address Media Reproduction in Machine Learning Era

digital · 2026-04-20

Mat Dryhurst and Holly Herndon are creating AI consent frameworks via haveibeentrained.com to tackle issues related to media reproduction in the age of machine learning. They refer to the phenomenon of "spawning," where AI generates endless works from its training data, as having greater implications than traditional sampling methods. This situation necessitates new consent models, as current ideologies surrounding free culture and intellectual property are insufficient. The roots of this problem can be traced back to Ted Nelson's Xanadu from the 1960s, although Tim Berners-Lee's HTML ultimately prevailed. The rise of Web 2.0 has resulted in orphaned media. Dryhurst and Herndon have trained neural networks using consenting data, highlighting that AI lacks the intrinsic value of human collaboration. They assert that AI will not supplant artists, given the socially intricate nature of artistic practices.

Key facts

  • Mat Dryhurst and Holly Herndon are building AI consent systems at haveibeentrained.com
  • They describe AI's ability to create infinite works from training data as "spawning"
  • Spawning is considered more consequential than 20th-century sampling or collage
  • New ideological frameworks for consent and attribution are needed in the AI era
  • Ted Nelson conceptualized Xanadu in the 1960s with bi-directional links for attribution
  • Tim Berners-Lee's HTML used one-directional links, which became the web standard
  • Web 2.0 prioritizes attention over creator attribution, leading to orphaned media
  • Companies build large AI models using data not consciously offered for training

Entities

Artists

  • Mat Dryhurst
  • Holly Herndon

Institutions

  • haveibeentrained.com
  • ArtReview

Sources