Where is the Mind? Persona Vectors and LLM Individuation
This work provides a conceptual framework for philosophers and AI researchers debating the nature of mind in LLMs, but it is incremental as it builds on existing mechanistic interpretability and persona literature.
The paper addresses the individuation problem for LLMs—identifying which entities associated with them should be considered minds—by analyzing persona vectors and mechanistic interpretability. It argues for the virtual instance view and introduces two new views (instance-persona and model-persona), concluding that persona-based views are promising alternatives.
The individuation problem for large language models asks which entities associated with them, if any, should be identified as minds. We approach this problem through mechanistic interpretability, engaging in particular with recent empirical work on persona vectors, persona space, and emergent misalignment. We argue that three views are the strongest candidates: the virtual instance view and two new views we introduce, the (virtual) instance-persona view and the model-persona view. First, we argue for the virtual instance view on the grounds that attention streams sustain quasi-psychological connections across token-time. Then we present the persona literature, organised around three hypotheses about the internal structure underlying personas in LLMs, and show that the two persona-based views are promising alternatives.