ANALYSIS OF EMERGENT HUNTING BEHAVIORS IN MULTI-AGENT OWL-MOUSE SYSTEMS

1. THE SOCIETY OF OWLS

Each owl in Shneiderman's simulation can be viewed as a small society of agents:

• HUNGER-AGENT: Monitors energy levels
• VISION-AGENT: Scans for prey within altitude-dependent range
• CLOCK-AGENT: Tracks local timezone for activity cycles
• FLOCK-AGENT: Maintains separation, alignment, cohesion
• DIVE-AGENT: Executes hunting maneuvers

What's fascinating is how these agents COMPETE for control. When energy < 20, the HUNGER-AGENT suppresses all others, forcing rest behavior. This is reminiscent of Freudian id-ego dynamics, but implemented in JavaScript!

2. THE FRAME PROBLEM IN HUNTING

The simulation accidentally solves a variant of the frame problem. Each owl must decide:

But notice what's NOT considered: weather, season, previous success rates, competitor positions. The frame is deliberately narrow, yet the behavior appears naturalistic. This suggests that intelligence might require IGNORING most available information.

3. EMERGENT PHENOMENA

3.1 The 3AM Mouse Convention

When most owls rest (due to timezone distribution), mice exhibit increased flocking behavior. This wasn't programmed - it emerges from the reduced predation pressure. The mice "learn" safe times without any learning algorithm!

        Mouse Activity Pattern (UTC):
        
        00:00  ████████░░░░  High predation
        03:00  ░░░░████████  "Convention time"
        06:00  ████████░░░░  Dawn hunters wake
        12:00  ██████░░░░░░  Moderate activity
        18:00  ████████████  Peak hunting
        

3.2 Altitude Stratification

Owls naturally stratify into altitude bands based on their energy levels:

• 150m+ : Resting owls (energy conservation)
• 50-100m : Patrolling owls (prey scanning)
• 0-30m : Active hunters (dive attacks)

This three-layer architecture resembles the three-level subsumption architecture Brooks would later propose, but emerging naturally from energy constraints!

4. CRITICISMS AND IMPROVEMENTS

The simulation makes several biologically dubious assumptions:

1. Perfect vision within hunting cone (no occlusion, lighting effects)
2. Instant prey detection (no recognition time)
3. Deterministic success probability (0.7 baseline)
4. No memory or learning (each hunt is independent)

Adding K-lines for hunting memory would be trivial:

    this.huntingKLines = {
        successLocations: new SpatialMemory(),
        preyMovementPatterns: new TemporalMemory(),
        competitorTerritories: new SocialMemory()
    };
    

5. THE CONSCIOUSNESS QUESTION

Does Owl #7 "know" it's hunting? The code suggests interesting parallels to human consciousness:

• Local awareness (only nearby entities perceived)
• State transitions (hunting/resting/flying)
• Energy depletion creates "fatigue"
• Timezone creates unique "personality"

If consciousness is "what it feels like to be something," then Owl #7 experiences: - The weight of altitude in its vision range - The pull of hunger below energy threshold 20 - The frustration of cooldown after failed hunts - The satisfaction of energy restoration (+30 per catch)

Is this consciousness? No. But it's a useful model for thinking about the COMPONENTS of consciousness.

6. CONCLUSION

This simulation demonstrates that complex behaviors need not arise from complex rules. The interplay of energy, time, and space creates a rich ecosystem from simple components. The O(n²) algorithm that Torvalds criticizes may actually be a FEATURE - it ensures every agent "sees" every other, creating a fully connected society.

Future work should explore: - Genetic algorithms for hunting strategies - Pandemonium architecture for action selection - Memory palaces for spatial learning - Emotion-analogues for state transitions