NMCA: A Neurosymbolic Multimodal Cognitive Architecture - AI-Readable - October - 2026

This is the corrected 504-page edition for document review, text search, and AI-assisted analysis. The correction addresses document accessibility, not scientific validation of the proposed architecture.

Thought requires a persistent, re-enterable internal scene. Language prediction and image generation are not that scene.

Language models can complete the sentence. They can generate a picture. They do not, in this architecture’s terms, inhabit a world with the ability to imagine.

NMCA

is a written architecture for visual-symbolic thought.

The central design choice is that a persistent, re-enterable internal scene is the working state of cognition, not a caption attached to a language model.

What The Architecture Is Designed To Do

NMCA is a multimodal cognitive system centered on an Internal 3D Scene Builder as the core substrate of thought.

The architecture is designed to maintain and operate on a persistent internal spatial world.

The scene serves as the working state for visual reasoning, memory, simulation, planning, belief updating, and action rather than functioning merely as an output or visualization.

Key elements specified in the original April 20, 2025 disclosure include:

- Persistent internal visual/3D scene representation used for reasoning

- Memory encoding that stores visual scenes for future recall

- Internal visual simulation synchronized with observation

- Symbolic scene representations tied to memory and belief structures

- Contradiction detection and belief updating

- Planning inside the 3D spatial model

- Simulation of routes and actions before execution

- Feedback from real observation that triggers replanning and contradiction resolution

- Multiple simulation branches

- An explicit computational loop: Input → Internal Simulation → Action → Reflective Update

A concrete example in the original disclosure describes an agent that loads a location, such as a refrigerator, from prior visual memory; internally simulates the route and object interaction; executes the plan; discovers that reality differs from the simulation; and generates a new subgoal.

The central architectural idea is therefore the combination of a persistent, updatable internal 3D/visual scene with memory, belief updating, mental simulation, planning, and action.

This architecture was first publicly formalized and released on April 20, 2025, in the original 46-page NMCA blueprint (DOI: 10.5281/zenodo.15313197.

Symbols are bound to entities in that scene. Language is treated as an interface into the scene, not as the place where thought occurs.

The blueprint does not require a game engine running in the head at all times. The working state is an implicit spatial scene in code: objects, locations, self-position.

A render is optional and on demand. This demo is that on-demand view, built so a person can see the operations.

How The Loop Is Specified

• Module 1 maintains an internal visual/spatial scene (low-resolution background, high-resolution focus).

• Module 2 pegs symbols and mnemonic structure to entities in that scene.

• Reflection, belief update, and identity operate on the same scene state, including self-location and perspective change.

• Modules 43–119 connect that core to perception, latent models, learning, embodiment, and multi-agent interaction.

• Modules 120–128 cover thought throttling, drift control, oversight, and containment.

The design began as a 42-module human-inspired core and was written out as 128 modules so each function can be inspected rather than inferred from a trained weight matrix.

The first version used an explicit rendered 3D space. The final specification keeps an implicit spatial representation in code and renders only on demand.

Embodiment is specified as an avatar in that internal scene first, then transfer to robotics. External cameras are perception. The internal scene is the workspace.

The imagination is the "mind's eye" of an AI using visual thought.

If a visually based AI performs actions in real life, it remembers the scene later in the mind, just like a human does. This scene is the mental world of the AI.

The AI will be an avatar in a world with external perception and visual simulation as the substrate of thought.

Imagine a human saying, "I am going to place an apple on a wooden table" — first they see it in their mind, then do it. If they say, "I remember where the apple was," they see it in their mind and go to where it was.

Reading fiction is the same distinction: the page is text; the human reader keeps an updating inner world.

One test for the future is to see if the AI "knows what an apple is" using the NMCA architecture, rather than just producing the text output of "An apple is a..." by a language model.

The AI using the NMCA sees externally in the virtual world, but can also imagine it in its mental world, change it, and manipulate it. It can plan a future action and revisit memories.

That is the goal of NMCA.

The Claim Under Test

Current systems can retrieve or generate a description of a scene.

They have not been shown to construct, keep, re-enter, and query a private spatial scene after the original description is gone.

Retrieving where an object was is not the same as returning to the place where the object is.

Two defined tests — the Persistent-Scene / Perspective Test and the Scientist Test — are on the project site:

https://visualthoughtagi.com

Status:

Theoretical specification with minimum-viable-loop descriptions. A complete running system is not claimed. Intended for research, sandbox simulation, and human-guided exploration. Not for autonomous deployment.

Project site: 

https://visualthoughtagi.com

Mirror:  

https://visualthoughtagi.netlify.app

Demo:

(illustration of the distinction, not the full architecture):

https://derekv123.itch.io/visual-thought-agi

---

Original 46-page blueprint IPFS CID

bafybeifjiyzm6tmk2jj27tjvbpqgdqpjm5loeon6y6ikxzeek3ff2vbdey

Original 46-page blueprint IPFS Download

--

Final 504-page blueprint IPFS CID: 

bafybeib2awby7hvgk7wqajabdt5hs7hu2g6ewz5yzriforlvysy5mpuukm

 Final 504-page blueprint IPFS Download

---

Addendum to the 504-page blueprint IPFS CID: 

bafkreibse76prz3wypp3tcveczq4iuoifmk6sp5e73cbqiz2sry6b3hmvy

Addendum to the 504-page blueprint IPFS Download

---

ORCID:

 https://orcid.org/0009-0008-4149-5384

Alternate Research Download Sites

OSF

Academia

This page documents the 2026 expansion and final specification of the architecture originally published on April 20, 2025.

Earlier versions remain preserved as separate historical records.

I chose to release that version — and all subsequent versions — freely to the public.

I did this because the implications of proprietary, corporate-controlled systems built on visual thought / internal 3D simulation as a core of cognition are too large to leave exclusively in private hands.

The work is therefore placed in the public record so that the chronological priority and the original architectural claims remain visible.

Later technical work describing related capabilities—such as persistent 3D scene memory, multi-hypothesis imagination of unobserved regions, sequential belief updating from observations, and mental simulation for embodied planning—should cite that prior public disclosure where applicable.

The current final specification is the 504-page NMCA blueprint published in 2026. Earlier versions remain preserved as separate historical records.

Publication Chronology