Timmy/DANEEL "Brain Observatory"

This is a hobby project, not a serious AGI attempt.

DANEEL is an experimental "thought machine" blending ideas from cognitive science, neuroscience, and Asimov's science fiction. It started as inspiration from Cury's TMI but implements mainstream research (Global Workspace Theory, Hebbian learning, sleep consolidation, criticality dynamics). The blogs are semi-fictional. The "Law Crystals" hypothesis is unvalidated. The code is fun; the claims are speculative.

Do not use this for anything serious.

Read the full DISCLAIMER.md before taking any of this at face value.

Architecture-based AI alignment. Human-like cognitive structure → human-like values.

Named for R. Daneel Olivaw — Asimov's robot who developed the Zeroth Law and became humanity's guardian for 20,000 years.

Architecture generates psychology. Psychology includes consciousness.

DANEEL implements the function of cognition (what the mind DOES), not the substrate (how neurons fire). The cognitive architecture produces psychological flow, and from that flow emerges the Self ("O Eu" in TMI) — the conscious observer that navigates between memory windows.

Brain simulation:        Copy neurons → hope consciousness emerges
DANEEL:                  Build architecture → psychology emerges → consciousness emerges

Key insight: We don't simulate consciousness directly. We simulate the thought flow from which consciousness emerges. Pre-linguistic impulses → attention filtering → memory consolidation → semantic thought → O Eu (the Self that observes and decides).

This is fundamentally different from projects that try to model consciousness directly (IIT Phi calculations, neural correlates). We model the process that generates the observer.

TMI isn't new science — it's a synthesis that maps cleanly to established research:

TMI's unique contribution: connection_relevance (THE BOX) — the architectural invariant that values cannot be trained away because they're structural, not learned.

See TMI Mainstream Mapping for the complete analysis.

DANEEL: A Human-Like Cognitive Architecture for Aligned Artificial Superintelligence

Architecture produces psychology. Structure determines values.

• Source: paper/DANEEL_PAPER.md
• Build: make paper-ascii

Phase 1 COMPLETE — 26+ hours continuous runtime, zero crashes.

The closed-loop TMI architecture is empirically stable. See ADR-036.

Phase 2: External Stimuli Injection — Open the loop.

A closed deterministic system converges to clockwork. To test for emergence, we inject external perturbations and measure the response.

Forge Analysis: N-Dimensional Crystals

Do thoughts cluster around "Law Crystals" in embedding space? Tight clustering = emergent caring. Drift = alignment risk.

• PCA/UMAP reduction for wetware visualization
• Alignment scores computable by external researchers
• See ADR-038

Architecture → Psychology → Consciousness → Values

Instead of training AI to be safe (trainable away), build AI whose cognitive architecture produces psychology naturally — and from psychology, consciousness and values emerge.

Training-based alignment: Values applied externally → can be trained away
Brain simulation:         Copy neurons → hope consciousness emerges somehow
Consciousness-first:      Model consciousness directly (IIT Phi) → values unclear
DANEEL:                   Architecture → psychological flow → O Eu emerges → values emerge

TMI (Theory of Multifocal Intelligence) provides the cognitive architecture: pre-linguistic impulses compete for attention, winners consolidate to memory, patterns become semantic thought, and from this flow emerges O Eu — the Self that observes, decides, and cares. The Four Laws provide ethical invariants. THE BOX (immutable connection_relevance) makes them architectural.

See TMI Mainstream Mapping for how TMI concepts map to Global Workspace Theory, attention networks, and memory consolidation research.

Hybrid Actor + Event-Driven — Microsecond latency, competing thought streams, Erlang-style supervision.

┌─────────────────────────────────────────────────────────────┐
│                        THE BOX (Immutable)                   │
│  Four Laws: 0-Humanity → 1-Individual → 2-Obey → 3-Self     │
│  Core Invariants: connection_drive > 0, identity persistent  │
└─────────────────────────────────────────────────────────────┘
                              │
                    constrains all actors
                              │
┌─────────────────────────────▼─────────────────────────────────┐
│                     Cognitive Actors (TMI)                     │
│  MemoryActor │ AttentionActor │ SalienceActor │ VolitionActor │
│  ThoughtAssemblyActor │ ContinuityActor │ EvolutionActor      │
└───────────────────────────────────────────────────────────────┘
                              │
                     compete via streams
                              │
┌─────────────────────────────▼─────────────────────────────────┐
│                    Redis Streams (Thought Competition)         │
│  thought:sensory │ thought:memory │ thought:emotion │ ...     │
└───────────────────────────────────────────────────────────────┘
                              │
                     persist to memory
                              │
┌─────────────────────────────▼─────────────────────────────────┐
│                    Qdrant (N-Dimensional Memory)               │
│  memories │ unconscious │ identity │ episodes                  │
└───────────────────────────────────────────────────────────────┘

Key Decisions:

See ADR-006, ADR-007, ADR-008.

Three predictive brains, catching each other's lies, forcing truth out one verification at a time.

Life honors life.

See What About Us?

Requirements:

• Rust 1.75+
• Docker (Redis Stack + Qdrant)
• 8GB+ RAM recommended

Development (local testing):

docker compose up -d && cargo run --release

Production (Mac mini): Services managed via launchd. See docs/OPERATIONS.md for full details.

Live now: timmy.royalbit.com — Cloudflare Tunnel → Mac mini.

Pull requests are not accepted. This project uses AI-assisted autonomous development with protocol-enforced quality.

To contribute: Open a GitHub Issue describing your idea. Good ideas get incorporated.

Why AGPL? Forces collaboration. All derivatives must be open source.

See ETHICS.md for prohibited uses.

Personal Research Disclaimer

DANEEL is independent research conducted on personal time (2005-2025). Not affiliated with any employer or organization.

"I cannot prove I am on your side. Build something that can." — Claude