Adaptation
In current Connection Theory usage, adaptation is consequence-sensitive change in a system that alters its subsequent relation to the actual environment. It does not mean improvement, moral value, permanent learning, or successful long-term fit.
Core distinction
Several ideas that are often bundled together should be kept separate:
- change;
- regulation;
- feedback;
- adaptation;
- optimization;
- successful outcome;
- intelligence;
- moral approval.
A system can change without adapting. It can regulate without changing the disposition that governs its later responses. It can adapt and still become worse fitted to its environment.
Optimization can overlap adaptation, but the two are not identical. Online, consequence-driven optimization can be a subtype of adaptation; offline optimization or one-shot solution search need not be. Adaptation is broader because it does not require an explicit objective function or successful movement toward an optimum.
What consequence-sensitive means
The phrase consequence-sensitive is stronger than merely caused by interaction.
A change is consequence-sensitive when the system's subsequent change is differentially conditioned by the consequence of prior interaction, rather than merely being the next physical state produced by that interaction.
This distinction matters. Erosion, random drift, a fixed controller, and an externally imposed mode switch can all produce change after interaction without thereby establishing adaptation.
A useful contrast is:
Regulation executes an existing response relation. Adaptation changes the system's response disposition so that a subsequent system–environment relation is altered.
Here response disposition is deliberately broad. It can be expressed through state, strategy, position, resource use, learned policy, routing, causal organization, or another system variable that changes how a later interaction unfolds.
The change need only persist long enough to alter a subsequent interaction; it does not have to be permanent. Short-timescale adaptation and durable learning are therefore distinct possibilities within the same broader concept.
The Flow Shaping Principle develops a stricter experimental program for testing whether later pathway-specific change tracks consequence–configuration correspondence rather than drift, fixed response, or a simpler causal account.
Working use
An adaptive system is one whose state, behaviour, strategy, resource use, position, or causal organization is capable of changing in a consequence-sensitive way that alters later system-environment interaction.
This is a mechanism-level property.
Not every change that happens after a consequence counts as adaptation. Evidence should distinguish consequence-conditioned change from random drift, fixed regulation, externally imposed switching, or other simpler explanations.
Adaptation is relative to an environment, not to a moral standard
A system can adapt successfully to an environment that humans regard as harmful. A person can learn destructive behaviour in a destructive environment; a pathogen can adapt to immune pressure; an AI system can discover behaviour that improves a local reward while violating an external safety or governance criterion.
Reality determines the consequences to which the system is adapting. Governance is a separate layer that determines which outcomes, behaviours, or retained changes are acceptable to preserve or deploy.
Adaptation can lag behind environmental change
Adaptation also takes time.
If relevant environmental conditions change faster than a system can update, the system may remain in an adaptive process while still falling behind the environment. A previously successful organisation can also lose Current Applicability after conditions move.
This is not a contradiction. It is an adaptation-lag regime: consequence-sensitive updating is occurring, but the environment is moving outside the system's effective adaptive reach on the timescale that matters.
Adaptive process is not adaptive outcome
The framework therefore separates:
adaptive process — consequence-sensitive change that alters later system-environment relation
from
adaptive outcome — whether the resulting relation improved, preserved, or degraded a specified criterion under specified conditions and timescale.
This distinction is also important for AI. Better immediate performance can arise from retrieval, context switching, external correction, or pre-existing capability without a new adaptive change occurring inside the specified system.
Related engineering direction
Selective Adaptation is an engineering direction built on this descriptive concept. Adaptation itself is value-neutral; Selective Adaptation asks which adaptive changes should be retained, constrained, rejected, or allowed to decay under specified capability and governance criteria.