Experience Buffer

The Experience Buffer is a core component in the system's learning and neurogenesis (neuron creation) module that functions as a memory for recent, significant events. It maintains a time-ordered log of the squid's context and uses this data to identify recurring patterns, which helps the system decide when and how to create new, functionally specialized neurons.

TIP: View the experience buffer using the Debug menu

How the Experience Buffer Works

The experience buffer is technically implemented as the ExperienceBuffer class, which operates in two main ways: maintaining a rolling log and tracking pattern recurrence.

1. Rolling Log (deque): The buffer uses a deque (double-ended queue) to store a fixed, limited number of recent experiences (default maximum is 50 experiences). When a new experience is added, the oldest one is automatically discarded, ensuring the buffer only contains the latest, most relevant context.

2. Pattern Tracking: When an experience is added, it is processed to generate three levels of pattern signatures, which are counted to track how often similar events occur:

• Specific Pattern: The most detailed signature, identifying a precise combination of trigger, outcome, and primary motivational neuron state.

• Parent Pattern: A broader pattern used for hierarchical grouping.

• Core Pattern: A minimal pattern used for "fuzzy matching" or identifying basic event categories.

The system analyses these counts to determine if a situation is a novel event or a recurring pattern, which heavily influences whether a new neuron is created, or if an existing neuron is strengthened.

---

Example Experiences (ExperienceContext)

Each recorded experience is captured as an ExperienceContext object, which is a snapshot of the squid's state and environment at the moment a significant event (the trigger) occurs.

• trigger_type - The general category of the event: novelty, stress, or reward.
• outcome - The result of the experience: positive, negative, or neutral.
• active_neurons A dictionary of all current neuron activations (e.g., {hunger: 20, anxiety: 85}).
• recent_actions A list of recent actions taken by the squid (e.g., [approach_plant, hide]).
• environmental_state Key external facts at the time (e.g., {food_count: 0, has_rock: True}).