Embodied Time, Active Inference, and Reflexive Selfhood: A Theoretical and Dynamical Architecture Proposal for Artificial Subjectivity

Large language model (LLM) agents achieve linguistic fluency yet remain stateless between invocations, disembodied, and devoid of intrinsic temporality or epistemically grounded self-reference. This paper inverts the LLM-centric paradigm by relegating the language model to a downstream organ of articulation governed by a continuous-time, five-layer dynamical architecture integrated at $10\,\text{Hz}$. Its core contribution is a unified mathematical formalization that translates philosophical conditions of subjectivity---standing-streaming time-consciousness, counterfactual freedom across a reflective hiatus, indexical de se grounding, and windowless intersubjectivity---into coupled state-space operators, autopoietic viability constraints, and epistemic gates. A 29-dimensional manifold ($\mathbf{\Omega}_t \in \mathbb{R}^{29}$) couples egocentric sensory hyle, 7D core affect, an 8D liquid-time-constant workspace with endogenously bifurcating attractors, and 6D HiPPO-Legendre retention. Intentional agency originates prior to language through a spectrally contractive, rank-adaptive sensorimotor world model evaluated under Expected Free Energy and metabolically costly volitional bifurcation (self-control vs. akrasia). Between strictly encapsulated ("windowless") agents, social co-regulation arises solely from observable kinematics, acoustic prosody, predictive mirror simulation, and Kuramoto phase synchronization, while speech is grounded by default via affordance-constrained decoding, forward-pass residual tensor steering, and final hidden-state dissonance monitoring. Analytical contraction, boundedness, and metabolic scaling conditions---together with world-model rank-scaling, $\pm 20\%$ Monte-Carlo perturbation, and 8B language-model tensor-bridge evaluations in the \emph{Little Sister} testbed---establish a coherent framework for structural-functional correlates of subjectivity without claiming phenomenal consciousness.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-17
DOI
https://doi.org/10.5281/zenodo.24325277
Primary Topic
Embodied and Extended Cognition
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Embodied Time, Active Inference, and Reflexive Selfhood: A Theoretical and Dynamical Architecture Proposal for Artificial Subjectivity

Robert Friedrich Heß
1 links
Zenodo (CERN European Organization for Nuclear Research)
Embodied and Extended Cognition
preprint

Embodied Time, Active Inference, and Reflexive Selfhood: A Theoretical and Dynamical Architecture Proposal for Artificial Subjectivity

Robert Friedrich Heß
preprint en

Abstract

Large language model (LLM) agents achieve linguistic fluency yet remain stateless between invocations, disembodied, and devoid of intrinsic temporality or epistemically grounded self-reference. This paper inverts the LLM-centric paradigm by relegating the language model to a downstream organ of articulation governed by a continuous-time, five-layer dynamical architecture integrated at $10\,\text{Hz}$. Its core contribution is a unified mathematical formalization that translates philosophical conditions of subjectivity---standing-streaming time-consciousness, counterfactual freedom across a reflective hiatus, indexical de se grounding, and windowless intersubjectivity---into coupled state-space operators, autopoietic viability constraints, and epistemic gates. A 29-dimensional manifold ($\mathbf{\Omega}_t \in \mathbb{R}^{29}$) couples egocentric sensory hyle, 7D core affect, an 8D liquid-time-constant workspace with endogenously bifurcating attractors, and 6D HiPPO-Legendre retention. Intentional agency originates prior to language through a spectrally contractive, rank-adaptive sensorimotor world model evaluated under Expected Free Energy and metabolically costly volitional bifurcation (self-control vs. akrasia). Between strictly encapsulated ("windowless") agents, social co-regulation arises solely from observable kinematics, acoustic prosody, predictive mirror simulation, and Kuramoto phase synchronization, while speech is grounded by default via affordance-constrained decoding, forward-pass residual tensor steering, and final hidden-state dissonance monitoring. Analytical contraction, boundedness, and metabolic scaling conditions---together with world-model rank-scaling, $\pm 20\%$ Monte-Carlo perturbation, and 8B language-model tensor-bridge evaluations in the \emph{Little Sister} testbed---establish a coherent framework for structural-functional correlates of subjectivity without claiming phenomenal consciousness.

Zenodo (CERN European Organization for Nuclear Research)
Embodied and Extended Cognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon