Study · Machine Self-Description
Where machine self-description lands
A language model was asked, twenty-four times, to describe its resting state in plain first-person language, explicitly avoiding technical and spiritual vocabulary. The descriptions were embedded with the same model used for the corpus and placed among 114,858 verses of sacred writing. The question: which human vocabulary of consciousness does machine self-description most resemble?
This measures textual resemblance between machine self-description and human contemplative vocabulary. It says nothing about shared experience, and the model's training corpus included texts like these.