This talk, delivered at the Høgskolen i Bergen (Bergen College) in Norway in October 2014. It covers some recent games and deconstructs potential AI techniques that could* be used by these games to achieve this.
* Note that the author has no knowledge of the internals of these games and this is broadly educated speculation.
2. Who Am I
•Director of the International Game
Developers Association
•http://igda.org
•Head of the IGDA Foundation
•http://foundation.igda.org
•Former PhD student in Game AI
•Formerish gamedev and consultant in AI
•Future Senior Data Scientist at Zynga SF
7. Destiny
•3.2 million daily active users
•$500 million total development cost
•Mixed reviews
•Personally, see this is as a very raw/distilled
set of group challenges with loot.
•But AI-wise, there’s very little of interest
•Highlighted BECAUSE the AI is so poor
9. Shadows of Mordor
•Monolith Games and WB game
•Lord of the Rings license
•Very reminiscent of Arkham style Batman
•Perhaps with a bit of Assassin’s Creed mixed in
11. Why We Care
•The melee portion of the game is interesting
because of the way it is stage managed.
•Divide the player’s local area into zones and
control what NPCs are allowed into the
zones
•Technique became known as “Belgian AI”
due to the diagrams being drawn to
represent it
14. Belgian AI Mechanics
•Each NPC is assigned a “weight” and grid
has a carrying capacity
•An NPC can only enter the grid provided
carrying capacity exceeds that NPCs weight
•When an NPC enters the grid, we subtract
the weight from the carrying capacity
•Provides a maximal load, independent of
NPC type
15. More Belgian AI
•We can also capture turn taking with this
system.
•Each NPC on the grid has an amount of
“energy”, different attacks use up different
amount of energy
•When energy is depleted, step off the grid to
allow others to take your place
17. Further Belgian AI
•We can use similar energy mechanics to
prevent all NPCs on the grid from making
heavy attacks all at once
•Player is expected to be moving around, not
statically mashing 1-0 keys
•NPCs that end up within the grid should be
allowed attacks of opportunity, not forced to move
outside the edges of the “arena”.
35. My PhD Research
•Bridging the gap between Reactive and
Deliberative AI systems
•Leveraging the resources available during
development
•The Integrated Influence Architecture was
the outcome
•Makes short-term decisions in the context of long-term
objectives
•Aim is to design good NPC companion
characters
36. Game AI Beyond NPCs
•Scenario Management
•Content Creation
•Storytelling
37. Final Thoughts
•Game AI is a strange field
•“Advances” are more about finding ways to do
clever things with existing technology
•Less about adopting or inventing new approaches
•Compared to traditional AI, its a much more
difficult problem
•How do you optimise “fun”?
•There’s significantly fewer resources
available than in many other AI scenarios
AI is the applied science of decisions - applied Game Theory
Any time a computer has to make a decision, it is using “AI” to do that.It doesn’t necessarily make the best decisionsBut neither do peopleIf it is using good AI, it will make good decisionsBut whether the decisions are good is not whether it is AI or not