Cognitive Modeling of KnowledgeGuided Information Acquisition in Games

1
Cognitive Modeling of Knowledge-Guided
Information Acquisition in Games

• Reijer Grimbergen
• Department of Informatics
• Yamagata University, Japan

2
Outline

• Where do we want to go?
• A cognitive model for perception in games
• A reproduction experiment in shogi
• Experimental results
• Conclusions and future work

3
Where do we want to go?

• Using Marvin Minskys Society of Mind theory for
  game playing
• What are the primitive agents?
• Input agents perception of board and pieces
• Output agents playing moves

First step Studying perception to understand
input agents
4
A cognitive model for perception in games

• The three-stage memory model (Atkinson and
  Shiffrin, 1968)

Environment
Information from the outside world
Sensory memory
Selective attention
Encoding
Long-term memory
Short-term memory
Retrieval
Output
5
A cognitive model for perception in games

• Perception guided by knowledge

6
A cognitive model for perception in games

• Feedback from long-term memory

Environment
Information from the outside world
Sensory memory
Selective attention
Encoding
Long-term memory
Short-term memory
Retrieval
Output
7
Perception in games

• Acquiring the necessary knowledge about a board
  position to make a decision about the next move
• Perception is strongly related to chunks
• Stronger players have bigger chunks of game
  knowledge, so are better at information
  extraction
• Chunks agencies?

Reproduction experiment to understand the
fundamental agents dealing with perception
8
Hypotheses about perception

• Hypothesis 1 It is easier to perceive ones own
  pieces than the pieces of the opponent

9
Hypotheses about perception

• Hypothesis 2 It is easier to perceive pieces
  that are promoted than pieces that are not
  promoted

10
Hypotheses about perception

• Hypothesis 3 Pieces closer to oneself are easier
  to perceive than pieces further away

11
Hypotheses about perception

• Hypothesis 4 Bigger pieces are easier to
  perceive than smaller pieces

12
Reproduction experiment
13
Reproduction experiment

• 10 positions
• Generated randomly
• 5 positions with pieces in hand
• 11 subjects
• All in their early twenties
• 9 subjects were absolute beginners
• 2 subjects had played in elementary school

14
Experimental results

• Reproduction time

15
Experimental results

• Correctness

16
Experimental results

• Hypothesis 1 It is easier to perceive ones own
  pieces than the pieces of the opponent

17
Experimental results

• Hypothesis 2 It is easier to perceive pieces
  that are promoted than pieces that are not
  promoted

18
Experimental results

• Hypothesis 3 Pieces closer to oneself are easier
  to perceive than pieces further away

19
Experimental results

• Hypothesis 4 Bigger pieces are easier to
  perceive than smaller pieces

20
Conclusions and future work

• Conclusions
• Perceptual clues in board and pieces do not guide
  the knowledge stored in memory
• Perceptual clues only trigger this knowledge
• Agents and agencies should be built around
  primitive concepts like board, piece and king
• Future work
• Build it!
• Further reproduction experiments with players of
  different playing strength