Adaptation
Make the search procedure or organization a variable that can respond to evidence.
Advanced topics · Research & teaching
Advanced
AutonomousSystems
When intelligence begins to organize itself.
A journey from self-improving procedures to adaptive organizations, constitutional swarms, and interactions between artificial societies. Study the ideas. Inspect the mechanisms. Explore the limits.
The architectural premise
The system itself
becomes a
research question.
A model produces an answer. An agent chooses an action. An advanced autonomous system can also participate in changing the procedures, relationships, and institutions through which its decisions are made.
The teaching is accessible; the subject is not elementary. These laboratories examine adaptability and collective complexity while keeping the underlying mechanisms open to inspection.
Emergence
Study how dispersed knowledge and complementary capabilities produce relationships.
Institutions
Examine how collaboration rules are proposed, selected, and translated into behavior.
Governance
Keep authority bounded, changes attributable, and organizational decisions inspectable.
The methodological ladder
Autonomy
moves upward.
At every stage, one more organizational decision becomes variable. The progression is conceptual: the notebooks isolate different mechanisms rather than forming a single executable pipeline.
01
Procedures
Improve the improver
Search over the method that generates and selects solutions.
STOP
02Separate inner policy search from outer scaffold optimization.
Organizations
Adapt the organization
Diagnose missing capabilities; activate or retire approved specialists.
Recursive adaptation
03Make team composition an explicit, testable decision.
Relationships
Let collaboration emerge
Discover evidence, signal a need, and recruit complementary expertise.
Swarm discovery
04Study how local interactions produce a collaboration graph.
Institutions
Choose the rules of collaboration
Propose social compacts and inspect their collective selection.
Constitutional swarms
05Trace how an adopted rule changes sharing and coalition formation.
Boundaries
Connect autonomous societies
Reach beyond local competence through a limited cross-society compact.
ESCAPES
Follow information, relationships, and institutional practices across boundaries.
A parallel trading branch explores capability exchange and temporary alliances between existing constellations. Explore dynamic ecosystems ↗
The reading collection
The ideas behind
the architecture.
Move from the umbrella research agenda to the individual mechanisms and the capstone. Each reading supplies the conceptual language needed to interpret the experiments.
00 · The umbrella memorandum
The research agenda
From agentic architectures to self-organizing artificial societies. Begin with the central proposition: autonomy can migrate upward into organization itself.
01 · Foundations
When the improver improves itself
A conceptual introduction to STOP: scaffolds, utility, meta-utility, and the search procedure as an object of improvement.
02 · Financial application
Self-improving trading systems
Twenty forecasting models, nested improvement loops, and the essential distinction between a better selection score and better held-out performance.
03 · Adaptive organization
Recursive organizational adaptation
Observe, diagnose, validate, and adapt. Study bounded specialist activation under an immutable governance layer.
04 · Endogenous relationships
Swarm Discovery
Follow dispersed evidence through signals, complementary responses, simple contracts, coalitions, and an evolving collaboration graph.
05 · Communication as institution
The bulletin board
Explore how a shared communication surface can support unexpected collaboration, social compacts, and transfer across domains.
06 · Collective rule selection
The constitutional swarm
From discovering other agents to proposing constitutions: examine how collective choice begins to shape the rules of cooperation.
07 · The capstone
ESCAPES
Three local societies, a competence frontier, a foreign encounter, an institutional collision, and a compact that makes collaboration possible.
08 · Return to the whole
A synthesis of the journey
Revisit the umbrella framework after the experiments. This synthesis entry currently contains the same memorandum as the opening research agenda.
The correspondence is thematic, not one-to-one: some readings share a notebook, while others frame a broader research agenda. See the complete correspondence table ↗
The computational laboratories
Read the argument.
Inspect the system.
Seven notebooks make selected mechanisms visible through code, synthetic environments, and recorded outputs. Open a notebook in Colab or inspect its source on GitHub.
N1–N7 are navigation labels for this collection. Display titles describe the actual content; original filenames are preserved. GitHub editions omit the large inserted infographics and retain code, text, and generated outputs.
01 · Procedures & organization
Improve the search. Reconfigure the team.
Search over search
The STOP forecasting laboratory
Optimize model combinations, then the recipe used to search for them. Compare validation gains with held-out transfer.
Anthropic key requiredBounded reconfiguration
Recursive organizational adaptation
Compare fixed and adaptive trading organizations across six synthetic regimes, with a specialist registry, validation gate, and audit ledger.
Deterministic fallback available02 · A related trading branch
From adaptive teams to interacting constellations.
Cumulative trading · NB04
Build self-adapting teams
Translate observed conditions into capability requirements. Assemble feasible specialist teams and inspect additions and removals.
Earlier-course data and results requiredCumulative trading · NB05
Build dynamic ecosystems
Let Growth, Recovery, and Preservation constellations negotiate services, exchange knowledge, and form temporary alliances.
Earlier-course data and results required03 · Collective autonomy
Discover collaborators, institutions, and other societies.
Endogenous architecture
Swarm Discovery
Trace local search, signals, responses, contracts, and the graph they create in a synthetic acquisition universe.
Deterministic core · See setup notesEndogenous institutions
The constitutional swarm
Use semantic assessment and constitutional proposals alongside deterministic voting. Observe how the selected sharing threshold shapes collaboration.
Anthropic key requiredThe capstone
ESCAPES: beyond the local society
Follow a bounded foreign appeal through institutional collision, a limited compact, and the return of information and institutional practices.
Prepared fallbacks availableThe discipline that survives
More autonomy.
More precise
questions.
The problem helps shape the organization that solves it. Evidence must establish whether that organization actually helps.
What is changing?Distinguish a revised answer, a revised search procedure, a revised team, and a revised institution.
What remains fixed?Identify the designed substrate: identities, permissions, registries, budgets, and selection mechanisms.
What counts as improvement?Separate selection scores from held-out performance, and a working demonstration from a general result.
Who authorizes the change?Trace proposal, evaluation, acceptance, implementation, and the record of what followed.
Choose your starting point
One collection.
Several ways in.
Start with the question you want to understand. The monographs support conceptual study; the laboratories invite experimentation and critical comparison.
Understand
Build a vocabulary for autonomy, endogeneity, organization, and institutional boundaries.
Experiment
Inspect an improvement loop or an adaptive team, then change one assumption.
Explore
Follow the emergence of relationships, collective rules, and cross-society interaction.
Investigate
Design a stronger baseline, test transfer, or make a proposed institutional rule executable.
Before you run
A clear route
into the notebooks.
The collection contains several kinds of laboratory. Check the setup requirements for the experiment you choose, then run its cells in order.
Local simulation,
selective model use.
N1, N2, N5, N6, and N7 construct their synthetic environments internally. N1 and N6 require ANTHROPIC_API_KEY; N2 and N7 provide fallbacks.
N5 has a deterministic core, but its setup still requests the Anthropic secret. Review the setup notes before running. Model identifiers and live-call budgets are configured inside each notebook.
Read the notebook-specific setup ↗Bring the earlier
course artifacts.
The team-reconfiguration and ecosystem notebooks depend on datasets and results produced by earlier course notebooks. Those prerequisite artifacts are not included in this repository.
Restore the earlier course inputs before an end-to-end run. The initial teaching policy is deterministic; optional live sections use OPENAI_API_KEY.
Saved outputs document particular executions. Synthetic examples illustrate mechanisms within their assumptions; they do not establish general superiority or investment performance. Read the full academic and methodological guide ↗