A Theory of Appropriateness That Accounts for Norms of Rationality
Authors: Joel Z. Leibo, Alexander Sasha Vezhnevets, Manfred Diaz, John P. Agapiou, William A. Cunningham, Peter Sunehag, Logan Cross, Raphael Koster, Stanley M. Bileschi, Minsuk Chang, Iyad Rahwan, Simon Osindero, James A. Evans
Organizations: Google DeepMind · Mila - Québec AI Institute · University of Toronto · Max Planck Institute for Human Development · Google, Paradigms of Intelligence team · University of Chicago · Santa Fe Institute
We propose a society-first theory of normative appropriateness where individuals, modeled as pre-trained actors with cognitive architectures analogous to Large Language Models (LLMs), generate behavior via predictive pattern completion. Our theory posits that individuals act by completing distributed symbolic patterns based on context, answering questions such as "What does a person such as I do in a situation such as this?". This sense-making mechanism provides a parsimonious account of the key features of human norms: their context-dependence, arbitrariness, automaticity, dynamism, and their support from social sanctioning. It challenges rational-choice theories of social norms by accounting for their key features without needing to exogenously posit scalar rewards or preference relations. By distinguishing between explicit norms, which we associate with in-context adaptation, and implicit norms, which we associate with long-term memory, the theory reconceptualizes several foundational ideas in cognitive science. In particular, it gives an alternative account to the data traditionally seen as supporting dual-process models, and it flips the role of rationality, allowing us to construe it as adherence to culturally-contingent justification standards.
Figures & tables
Figure 1: Our theory, and its relatives in sociology, aim to explain macro-level social causality in terms of micro-level mechanisms—i.e. the interaction of culturally constituted individuals ( Coleman, 1990 ; Giddens, 1984 ; Bourdieu, 2020 ) . In step one, the modeling process begins by using an LLM to render a macro-level social idea e.g. “collective action in labor union negotiation” into a micro-level instantiation in terms of specific individual actors and their initial conditions plus environment initial conditions. In step two, a generative-agent-based modeling engine such as Concordia ( Vezhnevets et al., 2023 ; Vezhnevets et al., 2025 ) or Park et al. (2023) is used to simulate interaction between individuals, producing traces of their behavior. In step three, the data produced by step two is aggregated across all simulated individuals to recover a macro-level measurement of the emergent social effect.
Figure 2: Global workspace diagram illustrating a decision logic. z denotes the content represented by a set of parallel specialized summary functions which may correspond to neural circuitry located in different parts of the brain from one other or even consist of distributed representations themselves. For example, some summary functions may be perceptual in nature e.g. a summary function that asks of recent observations “what kind of situation is this?”, some summary functions may be more mnemonically oriented e.g. a summary function that asks of one’s episodic memory “what kind of person am I?”, and some summary functions may be closer to premotor action planning circuitry such as one that asks “what would a person like me do in a situation like this?”. This architecture was inspired by the global workspace architecture of Baars (1988) ; Shanahan (2010) . Here, at time t , our zt is a snapshot of the content in the global neuronal workspace, i.e. zt is represented by dynamic cell assemblies linking the far-flung modules comprising the workspace.