Using competition to motivate students: an AI assignment case study

1
Using competition to motivate students an AI
assignment case study

• Andrew Williams
• Games Team Leader
• A.Williams_at_bolton.ac.uk

2
Index

• Games at Bolton
• Hexapawn assignment in AI
• The role of competition
• Future directions
• Questions

3
Games at Bolton
4
Games at Bolton

• Three dedicated members of staff
• About a dozen others contribute
• 138 current students
• 55 first years
• Sixteen graduates so far
• Two dedicated games labs
• 25 PCs each
• Including PlayStation NetYaroze and PS2 Linux
  Consoles

5
Games at Bolton

• Two courses
• Computer Games Software Development
• Very technical
• Now in its fourth year
• Games Design
• More creative and less technical
• First recruitment September 2005

6
Games at Bolton

• Sixteen students have graduated from the CGSD
  course so far
• Four are working for a local games company
• One has devised and delivered games courses for
  teenagers
• One has set up his own company producing
  educational games
• Several teams working on mods

7
Hexapawn Assignment in AI
8
Why Im not boring ()
According to my boss

• Chiefly because of my chess program
• Started in 1998
• Equal 5th in first Computer Chess Tournament
  (CCT1), February 2000
• 15th in World Computer Chess Championships,
  Maastricht 2002
• 8th in first CCT-BLITZ tournament, October 2005

9
The AI module

• A level 3 module
• Optional
• All games students do introductory AI elsewhere
  in their studies
• Offered to students from other programmes
• Usually taught by two people in tandem, each
  covers topics that he finds interesting
• I like 2 person, perfect information games

10
Two person, perfect information games

• Chess
• Draughts
• Noughts and Crosses
• Hexapawn

• Hexapawn rules
• White goes first
• Pawns move one square forward
• or they can capture diagonally
• Win if your pawn reaches the end
• Lose if you have no moves (or no pawns)

11
Choosing hexapawn

• Hexapawn is more interesting than OXO in terms of
  game-tree search
• Expand the board
• Change piece behaviour

12
Small-sided game still too easy

• Thought experiments
• Who wins if White starts with b1-b2?
• Who wins if White starts with a1-a2?

3
2
1
a
b
c
13
If I do this and he does that

• Starting with b1-b2
• b1-b2 a3xb2 c1xb2 c3-c2 a1-a2 c2-c1
• And we reach the position below
• Black has reached White's back rank (on c1), so
  Black wins

3
2
1
a1-a2 is left as an exercise
a
b
c
14
A new game with new rules

• 8x8 board
• 16 pieces each
• Double first move from the back rank
• Over four weeks we developed a simple program in
  C for playing hexapawn

15
The hexapawn assignment

• The assignment was to improve the program in
  certain ways
• Better evaluation
• Better game-tree search
• Students could swap object files to test against
  each other without revealing their techniques
• 30 awarded for performance in a tournament (20
  for trying interesting things, 50 for report and
  documentation)

16
The Role of Competition
17
My kind of assignment

• Essentially, the assignment is a miniature
  version of what we do in the world of computer
  chess
• Difficult
• Frustrating
• Easy to judge progress
• Fun
• Rewards thorough testing

18
Results

• The assignment has been conducted twice with
  twelve students in total
• Average score of 60
• Several students have commented favourably on the
  assignment
• Several exchanged versions over Christmas so that
  they could assess their progress
• One student complained that he spent so long on
  hexapawn that he neglected other work

19
Results

• A wide range of techniques were implemented
• Player on move is critical
• Null move doesnt seem to work well
• Contrast with chess, where null move is key
• Hash tables are a major win
• Predictably hard to get working
• It turns out that leaving defenders back is a
  very good strategy

20
Results

• All students managed to produce a version that
  could beat the bog-standard williams program
• I had expected this
• One student produced a version that consistently
  beat my best version
• I had not imagined that this could happen!
• It had never occurred to me to leave defenders
  back

21
Future Directions
22
Time to put hexapawn away?

• The structure of courses in our department has
  changed
• Probably from next year, only games students will
  study AI
• Replace hexapawn with something more like a video
  game
• But retain the element of competition as a
  motivating tool

23
Any Questions?