The Ouroboros effect and heterodox domains
摘要
The Ouroboros Effect refers to the fact that generative artificial intelligences (AIs) are increasingly being trained on AI-generated data. The effect results in the production of “boilerplate” or “biased” output, which does not pose any significant problem for what we call bias- and boilerplate-friendly domains. However, it does limit the applicability of generative AI to what we call heterodox domains: domains that build on yet depart from traditions, value novel and original contributions, and are context-dependent. Heterodox domains may therefore be resistant to AI-integration. And, significant problems are created when AI is applied to them. This has important implications for how we understand the application of AI to domains such as art, ethics, and education, as well as other domains that count as heterodox.