The Vectorial Turn: Consciousness, Self-Attribution, and Steerable Representations in Neural Space

Abstract

Street et al. (2026, arXiv:2607.28607) isolate a direction in the activation space of large languagemodels that separates states in which a model affirms its own mindedness from states in which itdenies it, and show that suppressing this direction — an incidental side effect of ordinary safety fine-tuning — also suppresses the model’s attribution of mind to animals, nature, and spiritual entities,and that steering restores these attributions together with more human-like answers on 95 GeneralSocial Survey items. This paper treats that finding as the occasion, not the proof, of a VectorialTurn in the philosophy of mind: a shift in what the field has to reason about, from purely behavioralevidence toward a mechanistically localized, causally steerable representational object. We call thisobject the Self-Attribution Vector (SAV) and treat “self-attribution vector” and “consciousnessvector” as strictly distinct terms throughout: the first names what has been measured, the secondnames a philosophical hypothesis about it.Building on a first revision that corrected the paper’s terminology and quotation practice, thissecond revision responds to detailed external review by (1) separating the title’s claim from the body’sclaim, so that neither implies consciousness = a steerable direction; (2) introducing an explicit four-level ladder — self-report, self-attribution, self-model, phenomenal consciousness — with a statedepistemic barrier at each step; (3) adding a dedicated section stating what the SAV finding doesnot establish; (4) stating the framework’s null and alternative hypotheses (H0–H3) and specifyingwhich the current evidence actually supports; (5) tightening the Vectorial Explanatory Gap (VEG) sothat it is a claim about the evidential limits of representational intervention specifically, rather thana relabeling of the Hard Problem; (6) adding a sourcing-confidence table for every contemporaryposition discussed; (7) specifying the cross-model replication protocol (models, layers, seeds, andthe representational-alignment metric) needed to make VF-P3 operational; and (8) expanding theliterature base into mechanistic interpretability and the science of consciousness. We retain VectorialFunctionalism (VF) as the paper’s organizing framework and the Controlled Philosophical Ablationprotocol as its proposed methodology, while narrowing every claim to what the layered structure canactually support.