07 Emotions emotion system

Emotion System

Krnl-AI includes an emotional model that influences decision-making based on the agent's internal state. This is optional and can be disabled.

VAD Model

The emotional state is modeled using the Valence-Arousal-Dominance (VAD) dimensional model:

Dimension	Range	Description
Valence	-1.0 to 1.0	Pleasure (positive ↔ negative)
Arousal	-1.0 to 1.0	Intensity (calm ↔ excited)
Dominance	-1.0 to 1.0	Control (submissive ↔ dominant)

State Properties

from krnlai import VADState

state = VADState(valence=0.5, arousal=0.2, dominance=0.3)
state.is_positive   # True if valence > 0
state.is_negative   # True if valence < 0
state.is_calm       # True if |arousal| < 0.3
state.is_intense    # True if |arousal| > 0.7

How Emotions Affect Behavior

Emotional State	Effect
Negative valence	Increases perceived risk (bias up to +0.2)
High arousal	Adds risk bias (+0.1 per unit)
Positive valence	Slightly decreases perceived risk
Calm state	Neutral, unbiased evaluation

Emotional Transitions

The emotional state changes based on events during the cognitive cycle:

Event	Effect
High risk detected	Negative valence shift (-0.2), increased arousal
Successful execution	Positive valence shift (+0.05)
Natural decay	Gradual return to neutral (5% per step)

from krnlai.core.emotion.vad import VADModel

model = VADModel()
transition = model.update(
    delta_valence=-0.2,
    delta_arousal=0.3,
    trigger="high_risk_detected",
)
print(f"Previous: {transition.previous_state}")
print(f"Current: {transition.new_state}")
print(f"Delta: {transition.delta}")

# Emotional decay over time
model.decay(steps=3)

Emotional Memory

All emotional transitions are recorded and can be queried:

# Full timeline
model.history

# Search by trigger
emotional_memory = model.emotional_memory  # if available
emotional_memory.search_by_trigger("error")

Pain/Reward Model

In addition to the VAD model, a pain/reward system provides reinforcement learning signals:

from krnlai.core.emotion.pain_reward import PainRewardModel

pain_reward = PainRewardModel()
pain_reward.apply_many([
    {"type": "reward", "value": 0.5, "reason": "task_completed"},
    {"type": "pain", "value": -0.1, "reason": "high_risk"},
])

Configuration

agent = CognitiveAgent(enable_emotions=True)  # default

When disabled, emotional state is always neutral and no emotional memory is recorded.