Learning Curves

Modeling how agents acquire, retain, and lose capability over time#

Learning in EcoEchoSystem is not linear improvement.
It is a time‑bound, stress‑sensitive, socially mediated process shaped by attention, identity, and feedback.

Agents do not “optimize.”
They adapt imperfectly, often too late, sometimes in the wrong direction.

Purpose#

This module exists to:

model non‑linear learning trajectories
capture plateaus, regressions, and overfitting
explain delayed adaptation and institutional lag
support regime transitions and collapse dynamics
prevent agents from learning unrealistically fast

Learning is constrained by time, identity, and cost.

Learning as Substrate Expression (S / E / R)#

Structure (S)#

skill representations
mental models
institutional procedures
training pathways

Activation (E)#

stress and urgency
feedback intensity
reward and punishment signals
crisis pressure

Relational Time (R)#

learning rate
forgetting half‑life
habituation
generational transfer

Learning curves are time‑asymmetric.

Core Learning Phases#

Agents typically pass through these phases.

1. Exposure#

initial contact with new information
low confidence
high error rate

2. Rapid Acquisition#

steep improvement
pattern recognition
fragile competence

3. Plateau#

diminishing returns
proceduralization
resistance to change

4. Stress Testing#

performance under pressure
reveals hidden gaps
may trigger regression

5. Consolidation or Collapse#

integration into identity
institutionalization
or abandonment

Learning Curve Shapes#

Common curve archetypes include:

Classic S‑curve — slow start, rapid gain, plateau
Crisis‑accelerated curve — sudden learning under threat
Overfit curve — rapid gain, poor generalization
Delayed curve — late but durable learning
False mastery curve — confidence exceeds competence

Different agents may follow different curves simultaneously.

Forgetting and Decay#

Learning decays through:

disuse
overload
narrative replacement
institutional drift

Forgetting is default, not failure.

Learning and Identity Coupling#

Learning is constrained by identity:

identity‑consistent learning accelerates
identity‑threatening learning resists
crisis may override identity filters

Agents often learn what they can accept, not what is true.

Agents learn via:

imitation
prestige bias
norm enforcement
misinformation diffusion

Social learning can:

accelerate adaptation
entrench error
synchronize failure

Institutional Learning#

Institutions learn slower than individuals due to:

procedural inertia
legitimacy constraints
coordination cost

Institutional learning often arrives after crisis.

Learning Metrics (Simulation Hooks)#

Trackable indicators include:

learning rate
error persistence
transfer effectiveness
decay rate
stress sensitivity

These metrics inform regime stability.

Failure Modes#

Learning modeling fails when:

agents learn instantly
learning is always beneficial
forgetting is ignored
identity is bypassed

Learning must be costly and uneven.

Integration Notes#

Learning curves:

feed into identity development
shape belief dynamics
constrain agent loops
explain delayed adaptation

This module explains why systems fail to learn in time.

Status#

Canonical learning curve framework for cognitive agent simulation.
Designed for individual, institutional, and civilizational agents.