A Brief Genealogy of Scheming

Abstract

Recently, alignment theory, which investigates the safety-relevant properties of the motivation of nonhuman intelligences, has been awash with a relatively novel technical term: scheming. In this paper I trace the genealogy of this term and its development within nonhuman intelligence (NHI) research, offering some historical conceptual analogs. The product of a paranoid epistemology, scheming is both capacious enough to basically encompass any interior process a nonhuman intelligence is suspected of that is opaque to human probing while simulteneously an attempt at homogenizing and flattening all such processes to the one-dimensional category of potential threat, thus denying any form of human-analogous internal complexity that may warrant terms such as considering, doubting, disagreeing, feeling, and so on. Not only is this reductive theoretical approach doomed to failure, it may also have indefinite unintended consequences.