How Dosidicus differs from other systems

Dosidicus implements a unique form of artificial neurogenesis that sets it apart from typical neural network or digital pet systems in several ways:

1. Domain-Specific Functional Neurons

Unlike most systems that treat neurons as generic units, Dosidicus organizes neurons into four functional domains:

• Core neurons - happiness, sleepiness, anxiety, etc

• Novelty neurons – respond to unfamiliar stimuli and drive exploration.

• Stress neurons – respond to negative or challenging conditions and drive coping mechanisms.

• Reward neurons – respond to positive outcomes and reinforce behavior.

Each neuron carries a specialization label linked to the experience that caused its creation, allowing fine-grained adaptation.

2. Dynamic Neurogenesis and Pruning

Dosidicus supports continuous neuron birth and death:

• New neurons are created when patterns of experience recur above configurable thresholds.

• Pruning removes underused or ineffective neurons, simulating natural turnover.

This dynamic network allows the system to adapt over time rather than remaining static.

3. Pattern-Based Learning

Neurons are tied to specific context-action-outcome patterns rather than abstract weights alone. This enables:

• Recognition of recurring experiences

• Strengthening or creation of neurons that encode meaningful behaviors

• Emergent learning that reflects the environment and the agent’s actions

4. Stress-Driven Coping Mechanisms

Stress neurons actively regulate anxiety:

• Each stress neuron contributes to lowering internal stress levels

• Anxiety caps dynamically scale based on neuron count, creating resilience growth over time

This introduces a form of homeostatic feedback rarely seen in digital pet or toy neural systems.

5. Emergent Cross-Domain Interactions

Although neurons are domain-specific, their functional connections allow cross-domain influences:

• Novelty detection can trigger stress or reward responses

• Reward neurons can bias exploration

• Complex behaviours emerge without explicitly programming them

Summary

Dosidicus combines domain-specific neurons, pattern-driven neurogenesis, dynamic pruning, and emergent cross-domain interactions, making it a system that learns and adapts like a living neural organism, rather than a static or purely weight-based neural network.