# Emergent Wisdom

> Research on AI alignment, the architecture of machine cognition, and
> multi-agent coordination. Six papers from March 2026 by Henrik
> Westerberg, each pairing an idea with a working prototype. The site
> is a front door to PDFs, source code, and the understanding-graph
> visualization that shows how the reasoning behind one of the papers
> was actually built up.

The work is organized as a stack of six layers, each layer answering
a different question about how to build intelligence that understands
rather than just predicts:

- **The Language** — how agents agree on meaning without coordination.
- **The Curriculum** — how models learn to reason rather than retrieve.
- **The Memory** — how to persist thinking, not just conclusions.
- **The Conscience** — how safety can be the substrate, not a coat.
- **The Spark** — how to make models produce structurally novel ideas.
- **The Architecture** — how to make intelligence compound through reuse.

## Papers

- [Sema — When the Hash Is the Word](https://emergentwisdom.org/papers/sema.pdf): Content-addressed semantics where each anchored term is simultaneously a word and a cryptographic proof. Companion site: https://semahash.org. Code: https://github.com/emergent-wisdom/sema.
- [Temporal Hindsight Learning — Blindness as Teacher, Hindsight as Curriculum](https://emergentwisdom.org/papers/temporal-hindsight-learning.pdf): The knowledge cutoff as a pedagogical tool. A 70B model matches much larger frontier systems on unseen events after fine-tuning on hindsight-generated curricula. Code: https://github.com/emergent-wisdom/temporal-hindsight-learning. Model: https://huggingface.co/emergent-wisdom/thl-llama-3.3-70b-lora.
- [Understanding Graph — Persisting the Invisible Thinking](https://emergentwisdom.org/papers/understanding-graph.pdf): A typed cognitive graph for capturing the shift from confusion to clarity — tensions, hypotheses, dead ends, belief revisions. Not what the AI concluded, but how it understood. Code: https://github.com/emergent-wisdom/understanding-graph. Live example: https://emergentwisdom.org/entangled-alignment/?project=metamorphosis.
- [Entangled Alignment — When Safety Is the Substrate](https://emergentwisdom.org/papers/entangled-alignment.pdf): Safety woven into the training distribution itself, so the model never sees thinking without values. There is no unaligned substrate to revert to. Code: https://github.com/emergent-wisdom/entangled-alignment.
- [The Ontology of the Alien — Escaping the Median Trap](https://emergentwisdom.org/papers/ontology-of-the-alien.pdf): Hard cognitive boundaries (e.g. alien physics) force LLMs out of their default archetypes and into structurally novel mechanisms. The boundary does the creative work. Code: https://github.com/emergent-wisdom/ontology-of-the-alien.
- [Fractal Intelligence — Infrastructure for Solving Any Problem](https://emergentwisdom.org/papers/fractal-intelligence.pdf): Decomposing concepts rather than tasks. A prototype across 100 problems in 20 domains shows the same conceptual nodes recurring, a single solver serving 17 domains, 65% reasoning reuse. Code: https://github.com/emergent-wisdom/fractal-intelligence. Prototype: https://emergentwisdom.org/fractal-intelligence/.

## Interactive

- [Understanding Graph visualization](https://emergentwisdom.org/entangled-alignment/): A 3D graph viz of the actual reasoning graphs built while writing two of the papers above (Metamorphosis case study and LLaDA). Each node is a typed concept (foundation, tension, surprise, decision, …); each edge carries a semantic relation (supersedes, refines, contradicts, …). It is a frozen snapshot of how a human researcher plus Claude thought through a problem together. Browse by clicking.
- [Fractal Intelligence prototype](https://emergentwisdom.org/fractal-intelligence/): The running prototype described in the Fractal Intelligence paper, showing concept reuse across problem domains.

## About

- Author: Henrik Westerberg
- Contact: henrik.westerberg@emergentwisdom.org
- License: papers CC-BY-4.0, code MIT
- Organization: https://github.com/emergent-wisdom

## Companion sites

- [semahash.org](https://semahash.org): Interactive browser for the Sema pattern library (~450 content-addressed cognitive patterns). Companion to the Sema paper.