The Society of Shogi A Research Agenda

1
The Society of Shogi- A Research Agenda -

• Reijer Grimbergen
• Department of Informatics
• Yamagata University

2
Outline

• Introduction
• The Society of Mind
• The Society of Shogi
• Building shogi agencies
• Conclusions and future work

3
Introduction

• Marvin Minskys Society of Mind
• A general theory about the workings of the human
  mind
• Few attempts to implement the theory

Research question Is a Society of Mind for games
possible?
4
The Society of Mind

• Important concepts of Minskys theory
• Agents
• Agencies
• Problem solving
• Growth

5
The Society of MindAgents

• Agent
• Any part or process of the mind that by itself
  is simple enough to understand
• K-lines
• Used to turn on a particular set of agents
• General classes Nemes and Nomes

6
The Society of MindNemes

• Nemes
• Represent aspects of the world
• Polynemes invoke partial states in multiple
  agencies
• Micronemes refer to aspects of a situation that
  are difficult to attach to any particular thing

7
Color agency
Green
Red
Blue
Yellow
Shape agency
Cost agency
Oval
Square
Round
234 \
100 \
198 \
Triangle
1000 \
Taste agency
Salty
Foul
Delicious
Sweet
8
The Society of MindAgencies

• Frames
• Used to construct agencies
• Represent a thing and all the other things or
  properties that relate to it in a certain way

9
The Society of MindProblem solving

• Difference-engines
• Reduce or eliminate the important differences
  between the current state and some desired state
• Censors and suppressors
• Censors suppress mental activity leading to
  unproductive actions
• Suppressors suppress the unproductive actions
• A-brains and B-brains
• B-brains monitor the activity of the A-brain

10
The Society of MindGrowth

• Different forms of learning
• Accumulating remembering each separate case
• Uniframing finding a general description
• Transframing bridging different representations
• Reformulation new ways to describe existing
  knowledge
• Attachment figures
• Goal learning instead of skill learning

11
The Society of Shogi

• The Society of Shogi
• Shogi agents
• Shogi agencies
• Problem solving in shogi
• Growth in shogi

12
The Society of Shogi Shogi agents

• The basic shogi agents
• Input agents location of pieces
• Output agents knowing what a piece can do
• Basic K-lines
• Change the location of pieces
• Change the abilities of pieces

13
The Society of Shogi Nemes in shogi

• Recognize castle agency

• Build castle agency

14




15
The Society of Shogi Problem solving in shogi

• Difference engines
• Playing towards the optimal position
• Censors and suppressors
• Good players are unable to see bad moves
• A-brains and B-brains
• To avoid looping

16
The Society of Shogi Growth in shogi

• Different types of learning in shogi
• Accumulating Storing of the games that have been
  played
• Uniframing Ibisha vs. Furibisha
• Transframing Joining pawn attack
• Reformulation Wall silver
• Attachment figures in shogi
• Professional players

17
Building shogi agencies

• The three shogi agencies
• Recognition agency
• Lookahead agency
• Learning agency

18
Building shogi agenciesRecognition agency

• Recognition agency
• Context agency
• Strategy, castle, attack strategy, piece attack,
  etc.
• Evaluation agency
• Material, king danger, mobility, etc.

19
Defend Piece


Pawn drop
Lance drop
Knight drop
Pawn move
Lance move
Knight move
Generate pawn drop on square
Generate pawn move on square
Generate lance drop on square
Generate knight drop on square
Generate lance move on square
Generate knight move on square
20
Building shogi agenciesLookahead agency

• The lookahead agency
• Move generator agency
• Capture piece, move attacked piece, attack king,
  etc.
• Search agency
• Attach priority to moves
• Satisficing
• Feedback

21
Building shogi agenciesLearning agency

• The learning agency
• Overlooking a move
• New agents need to be created
• Underestimating a move
• Changing the activity level of the agents
  involved
• Might need new agents

22
Conclusions and future work

• Building a Society of Shogi seems possible
• Building agents and agencies will be
  time-consuming
• Focusing on learning might be more efficient
• Problem Minskys theory is relatively vague
  about learning
• Research plan
• Year 1 Recognizer agents and agencies
• Year 2 Lookahead agents and agencies
• Year 3 Learning agents and agencies