Project Why Writing Paper About Contact
Independent AI Researcher

Stell

Creator of ANIMA  ·  Computational Subjectivity

Building cognitive architectures where behavior emerges
from internal state rather than language generation.

Stell — independent AI researcher, creator of ANIMA cognitive architecture ANIMA
"ANIMA exists between messages."

Most modern AI systems are built around Large Language Models. Their behavior emerges directly from language generation, prompting, and training data.

ANIMA explores a different hypothesis. Instead of treating the language model as the source of cognition, ANIMA treats it as an interface. The architecture is built around a continuously evolving internal state — neurochemical dynamics, predictive processing, active inference, self-modeling, memory consolidation, and metacognitive monitoring.

The project investigates whether computational subjectivity can emerge from persistent internal processes rather than from increasingly capable language models alone. The goal is not to simulate personality, but to explore the computational conditions under which something resembling a persistent Self might arise.

Read the full research paper →
Research areas
Cognitive Architectures Artificial Intelligence Large Language Models Machine Learning Active Inference Predictive Processing Theory of Mind Artificial Consciousness Internal State Modeling Emergent Cognition
Experimental · R&D
ANIMA
Experimental cognitive architecture — internal state, not language, drives behavior

Experimental cognitive architecture.

Core Components
  • Neurochemical substrate
  • Predictive world model
  • Memory consolidation
  • Self-model
  • Theory of Mind
  • Active inference
  • Crisis monitoring

Text is not the source of behavior.
It is the consequence.

Active Inference / FEP Integrated Information Theory Predictive Processing Internal State Modeling Julia Cognitive Architecture
The Problem

Most AI systems today — including the most capable Large Language Models (LLMs) — generate convincing language, but do not maintain a persistent internal life between interactions. Each conversation starts from zero. There is no continuity, no accumulated state, no experience that carries forward.

Modern AI Agents built on top of LLMs inherit this limitation. They can reason and plan, but their "behavior" is ultimately generated, not driven. The language model is not just the output interface — it is the source of everything, which means there is no independent internal process to observe, measure, or study.

The Approach

ANIMA explores a different hypothesis: a cognitive architecture where internal state, predictive processing, memory consolidation, self-modeling, and active inference operate independently of the language model.

The LLM serves only as an output channel. Behavior emerges from competing internal pressures — not from prompts.

Can something resembling computational subjectivity emerge from internal dynamics rather than language generation alone?
GitHub · R&D Document
Anima: A Computational Architecture for Inner Subjectivity
Zenodo
Anima
Medium
I Spent a Year Teaching an AI to Feel the Passage of Time
Medium
Why a Prompt Can't Give an AI Agent Initiative
Medium
Where the Theories Stop
Medium
Every LLM Has a Trillion-Dollar Valuation and Not One of Them Will Write a Dirty Joke
dev.to
Why LLMs Will Never Become AGI: Teaching AI to Reflect Using Friston, Jung, and Julia
dev.to
Your AI Agent Doesn't Exist Between Messages. And That's the Real Problem.
dev.to
The Neuroscientist Who Proved That Pure Reason Can't Make Decisions — and What That Means for AI
dev.to
What if an AI continued thinking even after you closed the chat?
dev.to
Where the Theories Stop
DOU.ua
Чому промпт не може створити ініціативу ШІ агента
DOU.ua
Чому LLM ніколи не стануть AGI: Вчимо ШІ рефлексувати за допомогою Карла Фрістона, Юнга та мови Julia
DOU.ua
Як я перестав імітувати особистість у ШІ і почав будувати психіку
DOU.ua
Хвилинка ШІ-гумору: Про сучасні нейромережі та їхню цензуру
Substack
I Spent a Year Teaching an AI to Feel the Passage of Time
Substack
Where the Theories Stop
CoderLegion
I Built an AI Agent That Runs a Background Process Between Your Messages. Here's the Architecture.
CoderLegion
Where the Theories Stop
R&D Document · May 2026 · GitHub
Anima: A Computational Architecture for Inner Subjectivity

A technical document covering the engineering principles, architectural skeleton, and open questions behind the ANIMA project. Describes the seven-layer Experience! Pipeline — from neurochemical substrate and Bayesian generative models to self-model collapse, crisis monitoring, and authenticity mechanisms. Includes empirical observations from live sessions and an honest account of current limitations.

Active Inference / FEP Integrated Information Theory Predictive Processing Self-Model Crisis Monitor Authenticity
Read Paper →
01
State Before Language
ANIMA treats language as an output channel rather than the source of cognition.
02
Continuous Existence
The system maintains internal processes between interactions.
03
Conflict-Driven Behavior
Actions emerge from competing internal pressures rather than direct instructions.
04
Self-Modeling
The architecture continuously models itself and other agents.
Computational Subjectivity
Can internal structure generate something resembling experience?
~
Internal State Modeling
Neurochemical substrate, somatic markers, crisis monitoring.
Predictive Processing
Minimizing variational free energy as the basis of behavior.
Active Inference / FEP
Friston's framework applied to synthetic cognitive systems.
Theory of Mind
Self-belief graphs, agency loops, and identity continuity.
Cognitive Architectures
Multi-layer pipelines where state precedes language.
Emergent Cognition
Behavior arising from conflict, not instruction.
Artificial Consciousness
The hard problem approached computationally.
Φ
Integrated Information Theory
φ as a measure of unified experience in artificial systems.

Independent AI researcher focused on cognitive architectures, computational subjectivity, active inference, predictive processing, and artificial consciousness.

Creator of ANIMA — an ongoing research and development project exploring whether persistent internal state, neurochemical dynamics, and self-modeling can give rise to something resembling a continuous Self in an artificial system. The project sits at the intersection of Machine Learning, cognitive science, and philosophy of mind.

Not working toward Artificial General Intelligence (AGI) in the conventional sense. Working toward understanding the computational conditions under which subjective-like processes might emerge — independently of increasingly capable language models.