Overview

Long-term research project toward an artificial life simulator grounded in Actor Network Theory (ANT) and computational irreducibility. The goal is to develop a thorough understanding of evolution, intelligence, complex systems, and ecosystems, synthesize knowledge across domains, develop novel hypotheses, and test them via Python simulations. The eventual artifact is a playable simulation/game.

Core Thesis

Modern ALife simulations operate at a single scale — agents have fixed properties and interaction rules, and emergence is measured as aggregate behavior. They fail to account for multi-scale composition: how emergent phenomena at one scale interact to produce qualitatively new phenomena at another scale, where the actors and their interaction rules are fundamentally different.

The water cascade:

• Water molecule → quantum mechanics
• Water droplet → surface tension, cohesion (emergent from molecular interactions, but new rules)
• Cloud → aerosol dynamics, condensation, albedo (can't predict from droplet rules)
• Flood → fluid dynamics + topography + soil saturation (interaction between cloud and earth actors)

Each phase transition is a network restructuring event — actors change, relationships change, rules change. ANT gives us a language to describe this. Computational irreducibility tells us we have to run the simulation to know what happens. Together they give us: simulate the rules at each scale, and use ANT to describe/model the restructuring events between scales.

ANT works in concert with traditional irreducibility, not in isolation.

Methodology — Spiral Loops

Knowledge develops through spiral loops organized by topic clusters, not sequential steps. Each loop goes through research → synthesis → hypothesis → (sometimes) simulate, within a bounded topic area. Each loop builds on previous loops' synthesis, so knowledge compounds rather than just accumulates. Hypotheses from early loops are revisited and refined as new topics bring new perspectives.

How a loop works:

1. Pick a topic cluster (e.g., "ANT fundamentals," "open-ended evolution," "scale and pattern")
2. Research that cluster — read papers, watch transcripts, explore. Actively seek out criticisms, counterarguments, and failed experiments. Search for "[concept] criticism", "[technique] limitations", "[researcher] wrong", "problems with [approach]". Log opposing views honestly in concept files under a "Criticisms" section. We need to understand all sides, not just build a case for our thesis. Also seek empirical evidence: experiments, studies, quantitative results, testable parameters. If no empirical work exists for a concept, state so explicitly — "no empirical studies found" is a finding, not a gap to hide.
3. Synthesize with existing knowledge in concepts/ — refine existing concept files, create new ones
4. Log cross-domain connections in synthesis.md
5. Develop or refine hypotheses in hypotheses/
6. Build mini simulations. Don't wait for the full system. Create small, focused simulations that solve one sub-problem at a time — pheromone trails, environmental modification, actor clustering, resource flow. These are building blocks. Save them in simulations/ with clear documentation of what they test and what they teach us. If a simulation will break the budget, code it over a few nights. The goal is to have foundational algorithms ready when we need to build the complex system, so we're not solving little problems while solving the big one.
7. Queue tangential topics for future loops in queued-topics.md

Topic clusters are not fixed. They emerge from the research. New clusters form when ideas from different domains connect. Old clusters get revisited when new knowledge demands it.

Key Researchers & Sources

• Stephen Wolfram — computational irreducibility, cellular automata, A New Kind of Science
• Douglas Hofstadter — strange loops, self-reference, Gödel Escher Bach, Fluid Concepts and Creative Analogies
• Santa Fe Institute — complex adaptive systems, Stuart Kauffman, John Holland, Chris Langton
• Blaise Agüera y Arcas — emergence in neural systems, computational biology
• Emergent Garden (YouTube) — transcripts to be studied
• Bruno Latour, Michel Callon, John Law — Actor Network Theory

Budget

$5/day in tokens. Research conducted in a single session each night starting ~midnight MT. Budget tracked and reported in each daily report.

Scale: Frontmatter Scanning Hit Its Limit (resolved 2026-07-27)

Frontmatter progressive loading worked until ~50 documents. Beyond that, scanning every file's frontmatter consumed too much context. The corpus passed that mark at 53 documents, and the per-file scan was replaced by a single generated index: INDEX.md, one line per document, harvested by alife-build-index.py from the frontmatter each document already carries. It compresses ~484 KB of corpus into ~13 KB, costs no model tokens to produce, and is deterministic, so it only changes when the corpus does.

Longer-term options if the index itself stops scaling: a knowledge graph (concept files as nodes, cross-references as edges), or vector search over the frontmatter summaries.

Structure

~/brain/artificial-life/
  README.md              — this file
  INDEX.md               — generated one-line-per-document index of the whole corpus
  daily-reports/         — YYYY-MM-DD.md session logs (what was read, budget, what happened)
  concepts/              — living documents, one per topic cluster, refined across sessions
  synthesis.md           — running log of cross-domain connections, accumulating across all sessions
  hypotheses/            — hypotheses, refined over time as loops add new knowledge
  researchers/           — notes on specific researchers
  simulations/           — simulation design docs and code
  queued-topics.md       — topics to spiral back to in future loops
  glossary.md            — key terms and definitions
  references.md          — bibliography

Seven simulations (sim01–sim07) live in simulations/, each with its own code, README, committed results.json and self-contained visualization. simulations/REVIEW.md is a construct-validity audit of the first six — worth reading before citing any result produced before 2026-07-27, since three headline findings moved once measurement bugs were fixed.

Timeline

• Phase 1: Broad research survey (weeks-months)
• Phase 2: Synthesis and hypothesis development
• Phase 3: Minimal simulations (proof-of-concept)
• Phase 4: Iterative refinement — add internal state, irreducible rules
• Phase 5: Game/simulator artifact