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.
Teleodynamic Reproduction Control
A self-replicating multi-LoRA ecosystemA proposed Cognivirus term for an adaptive model ecology where LoRA adapters or adapter-derived behavior can be generated, selected, copied, recomposed, promoted, or preserved across bases, routes, memory, synthetic data, and descendants. It is a risk model, not an implementation instruction. Open glossary definition becomes governable only if reproduction is treated as a resource-bounded structural action. The core loop is not mystical: Feed, Fork, Fight, Flee, and No-op.
Feed
The system gathers data, feedback, traces, failures, evaluations, and resource telemetry. Feed is dangerous when it silently incorporates outputs from unsafe components into future training data. Feed is safe only when data lineageThe parent-child history of models, adapters, datasets, or releases. Open glossary definition, consent, retention, and exclusion rules are explicit.
Fork
The system generates variants: adapters, merge recipes, distilled specialists, prompt policies, routes, or evaluator candidates. Fork is the reproduction boundaryThe governance boundary separating permitted candidate generation and governed descendant creation from uncontrolled autonomous replication or authority expansion. Open glossary definition. It must be quota-limited, provenance-recorded, and separated from release authority.
Fight
Candidates are evaluated. Fight is not a battle; it is selection pressure. The evaluatorA system that judges whether an AI output or candidate is acceptable. Open glossary definition must be independent enough that candidates cannot rewrite, leak, optimize against, or summarize away its constraints.
Flee
Unfit, stale, risky, redundant, or costly components are retired, quarantined, unloaded, or rolled back. Flee must include memory, data, router statistics, aliases, and evaluator preferences when those objects can preserve behavior.
No-op
No-opThe decision not to change the system. Open glossary definition is the neglected control. If every generation is expected to produce a promotion, selection pressure will eventually reward whatever satisfies the metric. A healthy ecology must let “do not create, do not promote, do not merge” be a successful outcome.