EDEN
Emergent Developmental Engine for Neural-Networks
EDEN is a cognitive architecture for developing AGI and embodied AI through experience, not data.
It is built on the bicameral mind theory of consciousness, and runs inside a time-dilated virtual environment. It learns by living, not by reading. EDEN is not an LLM. It does not assume intelligence emerges from text. It takes the older, harder bet: that intelligence emerges from time, embodiment, and experience. The way you and I did.
"He wanted to create consciousness. He imagined it as a pyramid. Memory, improvisation, self-interest. Never got there. But he had a notion of what it might be. He based it on a theory of consciousness called the bicameral mind."
"Arnold built a version of that cognition in which the hosts heard their programming as an inner monologue, with the hopes that in time, their own voice would take over. It was a way to bootstrap consciousness."
— Dr. Robert Ford, Westworld
The Theory
Julian Jaynes proposed that early human consciousness was bicameral: two cognitive systems, one issuing directives, the other experiencing them as external voice. Modern consciousness, in his framing, emerged when those two systems integrated and the inner voice was recognized as self.
As a theory of human history it is contested. As a blueprint for building a mind from scratch, it is something else entirely: a developmental scaffold. A way for an artificial agent to bootstrap from instruction-following to self-directed cognition through inner monologue, the way the theory imagines we did.
The Approach
Intelligence is not retrieved from a dataset. It is grown in time, through curiosity, embodiment, and feedback. EDEN takes this seriously and builds the conditions for that growth.
A simulated world provides physics, sensation, and consequence. Time dilation compresses years of development into days. A swarm of scripted agents scaffolds language, social norms, and shared tasks. The agent learns by perceiving, acting, predicting, failing, and adjusting. No pre-training. No imitation. Just experience, accumulated.
The Architecture
EDEN is organized into four layers, each responsible for a distinct part of the developmental loop:
The simulated environment. World engine, physics, sensory feedback, agent swarms, a taskmaster that shapes exploration. This is what the agent inhabits and acts on.
The integrated cognitive system. Perception, memory, curiosity, reward, self-model, theory of mind, world model, planning, language. All interacting modules, plus a learning system that adapts them over time.
Sensory and motor I/O between Cortex and Realm. Emotional state. Language. The inner monologue loop that Jaynes' theory identifies as the seed of self-directed thought.
This is also where guardrails live. If we're building something that might one day suffer, we should be ready for that from day one, not as an afterthought.
Current Status
EDEN is in design. The repository is private and the code is early. Public work is happening on Dragonstone, the language EDEN is being built in. EDEN will merge in once Dragonstone ships a usable bootstrap, FFI, core standard library, and the Forge package manager.
The order is deliberate. Dragonstone is not a side project that EDEN happens to use; it is the substrate EDEN requires. A language fast enough to run a compiled cognitive core, expressive enough to write self-modifying behavioral logic, and unified enough to bridge both without leaving the language. That is what is being built right now to make EDEN possible. No off-the-shelf engine or LLM is going to cut it.
EDEN is the long game. Dragonstone is the work in front of it. If any of this resonates with you, the architecture, the bet, the refusal to take the LLM shortcut, the place to follow along is the Dragonstone repository. The community is small. You'd be early.