The I Hate AI License

Original Article ↗ Hacker News Discussion ↗

A tongue‑in‑cheek “I Hate AI” content license that bans any use of works with AI tools has triggered debate over whether such terms are legally meaningful or even coherent. Commenters question if copyright can actually restrict training data use, how fair use and text‑and‑data‑mining exceptions apply, and whether the broad definition of “AI” would unintentionally block search engines, accessibility tech, and other routine processing. Many see the license as a symbolic protest or art piece—especially since it was generated by an LLM—rather than a viable legal instrument, and argue that bespoke “anti‑AI” licenses risk being both unenforceable and harmful to openness.

Intent and Nature of the “I Hate AI License”

  • License aims to forbid use of content in AI systems, especially for training models.
  • Many see it more as a protest or art piece than a practical legal instrument.
  • Some think it’s outright satire or trolling, especially because the license text itself was generated by an LLM.

Legal Enforceability and Copyright/Fair Use

  • Repeated theme: a license can only restrict uses that copyright law already reserves to the rightsholder.
  • If AI training is ultimately ruled fair use / covered by text-and-data-mining (TDM) exceptions, a license cannot stop it.
  • Others argue the opposite: if courts find training is not fair use, such a license could help create a clean test case by forcing AI companies to rely solely on fair use, not implied licenses.
  • Several note there are already TDM exceptions in some jurisdictions (EU with opt-out, Japan, others), so training may already be largely lawful there.
  • View that either training is covered by copyright (license relevant) or it isn’t (license irrelevant).

Definition of “AI” and Unintended Consequences

  • Definition in the license (“technologies designed to simulate human intelligence”) is seen as vague and overbroad.
  • Concerns it could:
    • Block search engines and common ranking algorithms.
    • Forbid screen readers, speech recognition, translation, and accessibility tools.
    • Ban intermediate AI-based processing (e.g., summarization, spelling correction).
  • Some suggest narrowing scope to “machine learning training sets” rather than “AI” generally.

Open / Free Culture and “No-AI” Clauses

  • Debate over whether anti-AI or noncommercial clauses are compatible with “open” or “free cultural works.”
  • One side: any use restriction (NC, no-AI) disqualifies a license from being truly free/libre.
  • Other side: these licenses are still far more open than default “all rights reserved,” and useful as intermediate options.

Ethical and Economic Views on AI Training

  • Strong sentiment that training on uncompensated copyrighted content is “theft” or disrespect of creators’ labor.
  • Counterview: humans and existing tools already “learn from” public content; AI should be treated similarly if not copying verbatim.
  • Ongoing lawsuits (e.g., about verbatim reproduction of news articles) are cited as pivotal; outcomes are seen as uncertain.

Alternatives and Proposed Approaches

  • Suggestions include:
    • CC-style licenses with explicit “no AI training” clauses.
    • Licenses that allow training only if resulting models are open.
    • “Copyleft for AI” or “derivatives only” concepts.
  • Many warn against rolling your own license and advocate starting from established ones plus carefully drafted AI-specific addenda.