In plain English
This page explains how AI systems can change over time through updates, tests, retraining, memory, and approvals even when no single model rewrites itself.
- Why this matters: AI risk can come from the whole arrangement, not one obvious model.
- What to look for: data, memory, routes, adapters, tools, evaluators, updates, and rollback paths.
- Technical version below: the expert terminology remains available and is linked through the glossary.
Evolution
AI systems can change over time even when no single model rewrites itself. Updates, tests, retraining, memory, and human approval processes can still create a selection loop where some behaviors survive and others disappear. Feed means the system receives data or feedback. Fork means a candidate is created. Fight means candidates are tested. Flee means failed candidates are removed or bypassed. No-opThe decision not to change the system. Open glossary definition means the system correctly chooses not to change.
The diagram shows a bounded evolutionary loop. Feed supplies evidence and resources. Fork creates candidates. Fight evaluates candidates. Flee removes unsafe or wasteful states. No-op is shown at the center to preserve non-growth as a legitimate outcome.
The unsafe unit can be the transition graph.
The relevant safety boundary includes every permitted transformation, not only the current model artifact.
Adaptive model ecologiesA changing AI system made from many connected parts, not just one model. Open glossary definition can behave evolutionarily even when every model artifact is immutable at runtime and every release requires human approval.
Variation can be created by an external pipeline. Evaluation can be performed by independent gates. Selection can preserve the artifacts that score well. Inheritance can occur through fine-tuning, merging, distillation, synthetic data, or routing rules. Succession can replace a carrier while retaining a behavior.
Read the flagship page: Nothing Has to Reproduce Itself.
Added evolution guides
- Teleodynamic Reproduction Control
- No-op as Reproductive Control
- Functional Replication Versus Autonomous Replication
New evolution expansion
v1.8.0 report-driven pages
- Endogenous Yardstick Drift
- Death by a Thousand Edits
- Deprecation as Apoptosis
- Selection Pressure in the Report Corpus
- Open-Ended Evolution Without Autonomy
v1.21.4 controlled-evolution additions
- Fitness, Novelty, and Selection — how utility, cost, novelty, and no-op decisions should be made explicit before promotion.
- ModelBreeder Controlled Evolution Synthesis — report-derived vocabulary for genomes, fitness vectors, novelty archives, and controlled release.