The Elias Thorne Effect: AI Inbreeding and the Quiet Degradation of Generated Stories
Arwa Mahdawi’s Guardian column examines how a single fictional character, Elias Thorne, keeps appearing in AI-generated stories. Cornell researchers found Elias in about 26.5% of 20,000 prompts and that most stories reuse a small set of names and motifs, suggesting models trained on large, shared datasets copy from each other and may avoid copyrighted material. This “virus-like” inbreeding could degrade future AI outputs and fuel a broader spread of AI-generated nonsense—serving as a caution about how the AI ecosystem’s own feedback loops might shape what we read and trust online.