The Affect Papers - Complete

Summary

A structural framework identifies affect as the evaluation of a system's trajectory against its own self-maintained invariants under recursive maintenance, placing it between recursive maintenance and conscious awareness. The framework distinguishes itself from phenomenological, functional, and autopoietic traditions and specifies falsifiability conditions. Applied to artificial systems, it finds current architectures (large language models, reinforcement learning agents, hybrid systems) structurally lack affect, regardless of substrate or behavior, and outlines future architectures that could satisfy the conditions.

Study at a glance

Abstract

The four papers collected under this cover develop a structural treatment of affect — what it is, where it sits in the dependency hierarchy of identity-maintaining systems, what follows from that placement for specific contested cases, and what empirical work the framework calls for. The papers can be read independently and collectively; each stands alone, and the four together constitute the framework's full engagement with the question of affect. The papers are: Paper 1 — Core. Affect as Constraint-Relative Evaluation: A Structural Specification. The foundational paper. Identifies affect with the differentiated evaluation of trajectory against self-maintained invariants under recursive maintenance. Places affect in a dependency hierarchy between recursive maintenance and the conscious regime. Distinguishes the framework from adjacent traditions (phenomenological, functional, eliminativist; autopoiesis/enactivism, active inference, allostatic regulation). Specifies falsifiability conditions. This paper establishes the framework's central commitments; the three companions develop applications. Paper 2 — Companion. Affect and Artificial Systems: Structural Criteria for Machine Consciousness Assessment. Applies the framework's structural criterion to the question of whether current and near-future AI systems have affect. Returns negative verdicts on current architectures (large language models, reinforcement learning agents, hybrid systems) on structural rather than substrate or behavioral grounds. Develops three diagnostic error modes for attribution decisions (anthropomorphic over-ascription, eliminativist under-ascription, representational collapse). Specifies what future architectures would need to satisfy the structural conditions. Addresses implications for AI welfare discussions. Paper 3 — Companion. Affect and the Philosophical Zombie: A Structural Argument Against Hardism. Develops an argument against zombie-possibilism that differs from the main lines of existing response. The argument does not dissolve the notion of phenomenal consciousness (as illusionism does) and does not operate through logical-epistemic trouble (as self-reference arguments do). It shows that the zombie scenario's stipulations are internally incoherent given the framework's identifications: a structural duplicate of a conscious system instantiates the structural conditions for affect by the terms of the duplication, and affect under substrate coupling is phenomenal character, so the zombie cannot lack phenomenal character without contradicting the first condition of its own specification. The argument's force is conditional on accepting the framework's identifications; it contributes a non-illusionist route to the anti-hardist literature rather than a universal refutation. Paper 4 — Companion. Affect and Empirical Measurement: Methodological Requirements for Testing Structural Predictions. Addresses the empirical dimension of the framework directly. Specifies what the framework predicts (affect signals tracking corrective cost structure, the anesthesia dissociation profile, absence of signatures in path-independent systems), identifies the measurement problems that prevent current testing, proposes three translational hypotheses testable with existing multivariate methods, and addresses the validation bootstrapping problem that any novel empirical framework faces. The paper is direct about the gap between the framework's in-principle testability and its current in-practice testability, and specifies what methodological work would close the gap.

Tags