AAAI Workshop

1
AI for Dynamic Difficulty in Games
2
Early work

• Interactive installations
• Stories that users participate in
• Interacting as expression

3
Fall 1999

• AAAI Narrative Intelligence Symposium
• Students from Parsons/NYU
• REPlay

4
The Wheel

• Set in a prison
• Rigid structure?
• Narrative monolog notes
• Factional
• Actions result in global change
• Persistence consequence
• Consequence drama!

5
Too Big

• Lots of great ideas
• Erving Goffman
• Character and Personality Disorders
• NPR for emotional impact
• Needed to graduate someday

6
2000

• GDC
• Flexbot
• Build a mod
• Fix multiplayer experience

7
Ghostbot

• Gladiator fiction
• Patrons donate items to favorites
• As long as you have one
• you can live!

8
Sucked!

• Bots Like White On Rice
• Novice players
• Live longer
• seemingly forever

9
2001

• Game Design Tuning Workshop
• MDA of Multiplayer
• Challenge
• Sensation
• Drama

10
Oops!

• Knee-capped multiplayer
• Removed challenge
• Rewarded failures
• Flattened the tension arc
• probably easy for experts to game

11

• Design Tools at Work

12
Multiplayer Revisited

• Fun because its people
• Training up part of the challenge
• Frustration is still a factor
• Why do we keep playing?

13
Challenge
High Skill
Too Difficult
Flow Channel
increasing challenge
Low Skill
Too Easy
increasing skill
M. Csikszentmihalyi
14
Training
Skill increase
New tasks
Flow Channel
increasing challenge
Beginner skills
Simple Goals
increasing skill
15
Tennis
Volley for 3 hours in hot sun?
Backhand?
Flow Channel
increasing challenge
Hit the ball?
Beat Your Boss?
increasing skill
16
Engagement Cyclical

• Each challenge new flow channel trip
• Understand and control those cycles?

17
Opponent Process Theory

• Comfort (steady state) is discouraged
• Sitting cross-legged
• Pins and Needles
• Applied to vision, emotion, body chemistry

                                             
18
                                             
Solomon Corbit
19
Success
10 sec.
10 min.
1 hr.
10 hrs.
time
Failure
20
Flux vs. Flow

• Regulating this feedback
• Engineering chaos, then calm
• Keeps the player engaged

21
Game Difficulty

• Typically static at run-time
• Flow happens but only for a niche

22
Experiencing Flow

• How can we expand the flow channel?
• Include just folks and gamers alike?
• How can AI help?

23

• Adjusting Game Difficulty

24
FPS as a Base Case
Die
Win (or retreat)
Enemy
Fight
Search
Solve (or not)
Spawn
Find
Find
Obstacle
Get (or not)
Object
25
Goal

• Avoid flailing
• flailing repeatedly edging towards a state
  where current means do not support necessary and
  immediate ends.

26
Inventory Theory
Flow Out
Flow In
Inventory Level
27
Health
28
Health
29
Health
30
Basic Steps

• Notice when a player is flailing
• Determine which changes can be made
• Execute changes as seamlessly as possible

31
In A Nutshell

• Look at
• Player Inventory
• Current obstacles
• Performance over time
• Generate projected probability of failure
• Adjust accordingly

32
AI View of this Process
observations
User Model
Game Play Session
conditions
actions
Action Planner
33
Dynamic Systems View
observations
State Estimator
Game Play Session
states
actions
Control Policy
34
Estimation

• Estimate probability of getting hit
• Average measurements over time
• Add a fudge factor to value new estimates over
  old ones

35
Control

• Regulator
• keeps system within a specified range
• low varience
• Tracking System
• keeps the system within a range that changes over
  time
• high varience

36

• And so

37
Hamlet

• Half Life/Case Closed mod
• Monitor game statistics
• Predict failure/death
• Define adjustment actions and policies
• Execute those actions and policies

38
Game Chart

• Some C and C widgets
• Display data
• Play session traces
• Knobs to Frob

39
Aesthetic Goals

• Ramping challenge
• Responsive reward
• Sense of mastery

40
Existing Dynamics

• Enemies
• increasing number
• increasing strength
• increasing accuracy
• increasingly variable behavior
• increasingly intelligent or tactical behavior
• Increased frequency of surprise attacks

41
Tweakable Mechanics

• Available health
• Available ammunition
• Enemy Type/HP/Accuracy
• Weapon Type/Strength/Accuracy
• of Targets
• Distance

42
Other Considerations

• Terrain/Lighting/Cover
• Group Movement
• Players Accuracy/Progress
• Players internal state

43
Two Simple Policies

• Novice case
• Expert case
• Scale between them?

44
Novice Policy

• Comfort Zone Regulator
• Babysitter
• Trial and error
• Steady supply, consistent demand

45
Expert Policy

• Discomfort Zone Tracking System
• Boxing coach or Drill Sergeant
• Ramping challenge
• Sporadic supply, high but erratic demand

46
Could you do better?

• Best policies?
• More robust policies?
• Other factors?

47
Even a little better is better

• 10 less work
• 10 more fun
• ... or both!

48
Thanks

• Ian NWU crew
• Michael, Phoebe and NI/CTP crew
• Looking Glass Mafia

49
Questions?