Research Date: December 2024 Purpose: Foundation research for Eternal Game Go-like positional play mode
1. Go Territory Scoring and Influence Concepts#
Scoring Systems#
Japanese/Korean (Territory Scoring):
- Count empty points controlled
- Plus captured stones
- More restrictive gameplay
Chinese (Area Scoring):
- Count controlled points AND stones on board
- Mathematically different, usually same winner
- Encourages filling in own territory
Territory vs. Influence Dichotomy#
Territory:
- Directly enclosed empty intersections
- Secured points
- Example: Corner plays creating closed-off areas
Influence:
- Building walls/groups for indirect control
- Forces opponent into suboptimal moves
- More valuable in center of board
- "Thickness" radiates control across board
Key Insight:
- Thickness is better used from a distance
- Building territory in front of thick position is inefficient
- You'll likely get that territory anyway
Strategic Principles#
Corner, Side, Center:
- Corners easiest (two board edges as boundaries)
- Sides next (one edge as boundary)
- Center hardest (must surround on all four sides)
Invasion vs. Reduction:
- Invasion: New living group inside opponent's influence
- Reduction: Stone far enough to reduce territory, close enough to survive
Sente and Gote:
- Sente: Moves that compel response (initiative)
- Gote: Forced responses (losing initiative)
- Player with sente controls game flow
Balance Concepts#
Multiple Level Balance:
- Close together = secure area
- Spread out = cover more area but leave weaknesses
- Too low (near edge) = insufficient territory
- Too high (far from edge) = invadable
Overconcentration:
- Wastes potential influence
- Spread forces for multiple area control
- Flexibility is key
2. Othello/Reversi Flip Mechanics#
Core Flip Mechanic#
How Flipping Works:
- Place disc with your color up
- Any opponent discs in straight line (H/V/D) between placed disc and existing disc flip
- Must flip at least one disc for valid move
Unique Property:
- Captured pieces switch sides, not removed
- Board becomes increasingly full
- Late-game massive swings possible
Positional Strategy#
Corner Control:
- Corners can NEVER be flipped once placed
- "Stable" discs - permanent advantage
- More corners = higher win probability
X-Squares and C-Squares:
- X-Square: Diagonal to corner - DANGEROUS (gives corner to opponent)
- C-Square: Adjacent (H/V) to corner - risky but less so
- Avoid X-squares unless you already have that corner
Edge Control:
- Stable edges build from corners
- Losing both corners on an edge = disaster
- Balanced edge presence important
Mobility Strategy#
Paradoxical Principle:
- Flip as FEW discs as possible in early/mid game
- Restricts opponent's options
- Makes them more predictable
- Late game: massive flips to win
Stability:
- Stable discs can never be flipped
- Build stable positions from corners outward
- Quality over quantity early game
3. Area Control in Inis and El Grande#
El Grande - The Defining Area Control Game#
Majority Scoring System:
- 1st place in region = most points
- 2nd place = fewer points
- 3rd place = even fewer
- Multiple players can benefit from same region
Strategic Implications:
- Sometimes fighting for 2nd is better than challenging 1st
- Resource allocation across multiple regions
- Timing of scoring rounds crucial
Key Innovation:
- Defined the majority-control eurogame genre
- Influenced countless subsequent designs
- Balance between concentration and spreading out
Inis - Deceptive Simplicity#
Core Mechanic:
- Celtic clans vie for island control
- Area control with card drafting
- Tight rule set with subtle implications
Strategic Depth:
- Single pawn can cost the game
- Conflicts can end in peace
- Constant counting and recounting
- Every action and movement matters
Victory Conditions:
- Multiple paths to victory
- Presence vs. sovereignty vs. sanctuaries
- Creates interesting strategic choices
Area Control vs. Area Influence#
| Aspect | Area Control | Area Influence |
|---|---|---|
| Dominance | Complete territorial | Partial presence |
| Scoring | Binary (control or not) | Graduated (majority rewards) |
| Example | Risk | El Grande |
| Feel | Aggressive | Strategic |
4. Influence Projection in Abstract Strategy#
Radiation Model#
Distance Decay:
- Influence strongest at source
- Weakens with distance
- Rate of decay varies by game
Blocking:
- Obstacles can block influence
- Enemy pieces may reduce influence
- Terrain effects
Go's Moyo (Framework)#
Concept:
- Potential territory, not yet secured
- Large areas of influence
- Can be invaded but with risk
Strategic Use:
- Build moyo to control center
- Force opponent to invade
- Convert influence to territory or profit
Connectivity#
Groups and Chains:
- Connected pieces share life/death
- Isolated pieces vulnerable
- Connectivity = strength
Liberty Counting:
- Liberties = adjacent empty points
- More liberties = harder to capture
- Influence proportional to stability
5. Emergent Complexity from Simple Rules#
Definition#
Emergent Gameplay:
- Complex situations arise from simple mechanics
- Effect of whole system not deducible from components
- Interactions between rules create complexity
Go as Prime Example#
Simple Rules:
- Place stone on intersection
- Capture by surrounding
- Territory wins
Immeasurable Depth:
- ~10^170 possible game states
- No computer solved it through brute force
- Deep strategic patterns emerge
Depth-to-Complexity Ratio#
Goal: High depth, low complexity
Achieve Through:
- Strip away non-essential rules
- Remove excess theme/terminology
- Focus on meaningful decisions
- Let interactions create complexity
Types of Complexity#
Mechanical/Rules Complexity:
- Understanding rules and interactions
- Baseline needed for strategic depth
- Example: Chess > Checkers
Emergent Complexity:
- Decision complexity from too many options
- Go's open board = endless possibility
- Natural pruning as game progresses
Strategic Complexity:
- Meaningful decisions
- Multiple viable approaches
- Depth without cognitive overload
Design Principles#
Before Adding Mechanics, Ask:
- Does it improve existing gameplay?
- Or is it just a clever idea?
- First adds depth proportionally
- Second adds mostly complexity
6. Ko Rule and Superko Variations#
Basic Ko Rule#
Problem: Infinite capture-recapture loop Solution: Cannot immediately recapture ko
How It Works:
- Player A captures one stone
- Creates position where B could immediately recapture
- B must play elsewhere first ("ko threat")
- Then can recapture
Ko Fights#
Strategic Element:
- Ko threats as currency
- Larger ko threats force responses
- Can trade ko for other board positions
- Complex evaluation required
Superko Variations#
Positional Superko (PSK):
- No board position can repeat
- Ignores whose turn it is
- Most restrictive
Situational Superko (SSK):
- Position + whose turn can't repeat
- Same position with different player allowed
- American Go Association uses this
Natural Situational Superko (NSSK):
- Position + who created it can't repeat
- Based on who placed last stone
- Subtle distinction from SSK
Application to Eternal Game#
Recommended: Custom Resonance Ko
- Position + Resonance state can't repeat
- Different Resonances create different "situations"
- Allows for more interesting repetition dynamics
7. AlphaGo Evaluation Functions#
The Challenge#
Why Go is Hard for AI:
- Strategic/aesthetic nature
- No clear evaluation heuristic
- Branching factor too large for tree search
- Territory and influence are fluid concepts
Neural Network Solution#
Value Network:
- Evaluates board positions
- Estimates probability of winning
- Learned, not designed
- Solves "designer skill ceiling" problem
Policy Network:
- Selects promising moves
- Trained on expert games
- Guides search to relevant moves
How AlphaGo Handles Influence#
Learned Evaluation:
- Territory control
- Influence over key areas
- Potential for future expansion
- All learned from self-play
Maximize Win Probability:
- Favors higher chance of winning by less
- Over lower chance of winning by more
- Different from human "maximize territory" tendency
AlphaGo Zero Advancement#
Single Network:
- Combined policy and value
- More efficient training
No Rollouts:
- Relies purely on neural network evaluation
- High quality network replaces Monte Carlo simulation
Lessons for Eternal Game#
AI Evaluation:
- Don't hard-code influence formulas
- Let evaluation emerge from training
- Multiple factors combine in learned function
- Win probability > point margin
8. Design Recommendations for Eternal Game#
Stone System by Resonance#
| Stone Type | Characteristic | Influence | Capture Difficulty |
|---|---|---|---|
| Clarity | Reveals hidden | Low range, high accuracy | Easy |
| Flow | Extended reach | Slippery, mobile | Medium |
| Presence | Local power | Strong, attractive | Medium |
| Root | Immobile anchor | Low range, stable | Very Hard |
| Void | Hidden threat | Destructive | Hidden until adjacent |
Continuous Influence Model#
Not Binary Control:
- Influence percentage per territory
- Multiple players can have influence
- Majority determines control
- Ties require resolution rules
Distance Decay:
influence(d) = base_strength * (1 - decay_rate)^distance
Type Interactions:
- Clarity reveals Void
- Flow extends through gaps
- Presence amplifies allies
- Root anchors chains
- Void disrupts all
Capture Mechanics#
Influence-Based Capture:
- Not pure surrounding (Go-style)
- Capture when influence exceeds threshold
- Resonance vulnerabilities apply
Cascade Effects:
- Captured stones may trigger adjacent captures
- Chain reactions possible
- Creates dramatic swings
Ko/Repetition Rule#
Resonance Superko:
- Position + Resonance state + turn
- Allows more strategic repetition play
- Resonance changes create new situations
Scoring System#
Hybrid Territory/Influence:
- Territory points: Controlled empty spaces
- Influence points: Partial presence in contested areas
- Capture points: Stones captured
- Chain bonuses: Connected group strength
AI Opponent Design#
Learned Evaluation:
- Train neural network on self-play
- Value position, not hard-coded rules
- Multiple difficulty levels via depth/quality
Evaluation Factors:
- Stone count by type
- Controlled territory
- Influence projection
- Chain strength
- Potential vs. secured points
References#
- Go Strategy and Tactics - Wikipedia
- Othello/Reversi Strategy - Various Sources
- El Grande - Wolfgang Kramer & Richard Ulrich
- Inis - Christian Martinez (Matagot)
- "From Rules to Emergence" - rct AI
- "Managing Complexity" - Pine Island Games
- "Mastering Go with Deep Neural Networks" - Nature (AlphaGo paper)
- BoardGameGeek - Area Majority/Influence Mechanic
- Go Rules Variations - debigare.com