Blog Archive

Wednesday, March 10, 2010

live notes GDC AI summit, rant

Quick notes from the rant session

SPEAKER/S: Dave Mark (Intrinsic Algorithm), John Funge (Netflix), Borut Pfeifer (Plush Apocalypse Productions), Steve Rabin (Nintendo of America), Kevin Dill (Lockheed Martin) and Adam Russell (University of Derby)


//went out

mby first speaker,
missed talk, only saw ending with a pic of a sexdoll

2nd speaker
working with designers: red pill - blue pill
bottom up design...creativity, emergence...feel open enden...
may argue it is all variation of the same thing. boring. hard to control. frustrates designers.
wall climbing...
top-down design.
complex scenes. narrative. supports cinematic style. allows delivering more unique experiences.
reduces content re-use. does not scale well. reduces re-play value
not enough. not bottum up either.
must support free expressing.
do both. Side-ways design. Pierre Bourdieux
A system of dispositions 1 peception 2 thought 3 action
esternally driven
freedom within structural constraings (fascade pic in bg)
should foucs on multiple npcs. make content declarative, not imperative.
speaker has the most ironic tone.
Adam Russel

3rd. John Funge.
not rich, bottom of food-chain.
try to get middleware into platform SDK. Hard.
protect code. Patent.
developers won't want the name of middleware maker in credits (cool thing done by someone else)

4th (Utility person)
basic premises, things we hope to see
why not boolean... floats giving shades of gray. //good point, where is the spectrum
lack of subtlety and nuance, and predictability

utility. Slow. Why would it be slow? Computers good at computation but bad at branching. Don't have to rethink every frame. React to certain even, duck from granade, not ponder going for a beer.

custom script languages

harder to author logic.
with custom scripting language: no tools, intellinsense etc
not stable.
no breakpoints, no stepping, risky to reload. // omg yes the printf, argh.
language design is hard! c++ - 30 years, LUA for 17.
tools hard to make too! 27 years into debuggers (MS)
who makes it? this one guy. Needs to be a megalomaniac. The language taking over it all. They have a degree! The combination is deadly!
2 years on pain working with a scripting language.
hope. Lua.
make games not languages!

6th Dave Mark
annoyed by media. How game ai is reviewed.
and how the reviews are responded to.
"these people hate us"
bad, weak, retarded,...
no-one says anything nice!
nothing noted about the good stuff! If good its not noted, as if not existing.
...what IS cool? need to tell: Fear and sims3.
The more natural the less noticed.
catch 22
#1 dont let the agent do anything stupid...
maybe if:
- more challenging ai
- more life-like
expect more than 'dont be dumb'?


Some quick notes jotted down about the talk

SPEAKER/S: Michael Mateas (University of California, Santa Cruz), Daniel Kline (Crystal Dynamics) and Emily Short (Independent)


official DESCRIPTION: For some time, the industry has been exploring how to effectively manage the complexity of multiple story arcs, contextually appropriate character behavior, and yet still maintain an over-arching ebb and flow of tension and drama. In their quest, writers and designers have started looking to AI for solutions to these problems. Additionally, many techniques that are already being used in interactive drama can be used to augment traditional games. Through three short lectures, this session gives examples of ways that AI can enable the design and implementation of branching narratives, dynamic adaptive dialog, interactive storytelling, and drama management.

Daniel Klime starting

8 use cases.
"storytelling is searching.

A practical taxonomy of games.

gaming vs story...



RC - Experiment - Challenge

pace control


there are off the shelf tools that can be used now for this

drama management tecniques.
policy for 'story move' selection
aternative to explicitely conded links.
action selection problem - we have worked for that for years.
selection policy - as a function of the history.
story policy - exponential unwinding - computers are good at that.

- story moves: beats
- policy desired tension arc, beat preconditions.
FarCry used similar method

Far Cry 2
- story moves: missions
- policy: desired infamy, mission preconditions

Policies can use 'look ahead'

chess-like heuristics can be used for drama management

dramamangare: looking at past history, picking best story moves maximising expected goodness and utility.
future expected goodness.

next step is to move from drama management to story generation.
what if the nodes become virtual?
generate nodes.

classical approaces to story generation. symbolic ... remimplemted in eis classic story generators. in past - limited by CPU.

story modeling - grammars
world and character modelling
author modelling

tale-spin: world and character simulation
tale spin, first. simulation. HTM planning (term BT not invented then)

cavazza and mead - taking talespin model, new HTM. (well, BT)
representational, abstract changes. Ross wanting to gain rachels affection.

Caveat: character pursuing goals and plans does not automagically lead to interesting stories.

M looking forward to Chris Heckers take on this (sims3)

second classic:
authorial goals plans,
Universe author gials and plog framgments (plans)
- author goals and plans may make no sense form character viewpoint.

example plot fragment.
2 persons married. Have parents who are not nice.

universe algo:

recusively solving author goals.

//argh camera died. Hope the slides will be online.

universe model - graphical interface for creating and running universe domains
//was pesented at DAC, james was there

m giving a quick overview.

library of stroy actions. Looking at past. Adapt to future situations.

implemented new version, for quest generation.
pointing to Noah's book - one chapter for each.

take home message
-3 points

Emily Short

ideal conversation
- characteries npc

"Dialog trees suck. "

expressiveness of NPC difficult to convey via dialog trees.

- Quip- an actual line of dialog spoken.
- subject: a representation of what the dialog is about. "weather"
- scene: a sequence of dialogue exchanges that has some narrative unity

game: Glass - a fractured fairy tale
twisted cinderella story
in stepmothers livingroom
player is in a cage, as the family pet, a parrot
player can squeak out words - by so affecting the dialog.

subject-structured. links between subjects, guiding associations for conversations.
goal seeking behaviour of NPCs- scene specific.
Quips as connectors

example where the parrot is silent. leads to trying shoe on stepsister.

parrot goes grazy: narrative can go the same way as a silent parrot, stepsister + shoe

or parrot affecting so cinderlla gets to try shoe (?)

game: Alabaster
stepmother asks for the princess' heart in a box.
player is the hunters "should i be killing her or not?"
this game is more in the players control


//suddenly the jet-lag hits me

npc has quip agenda
narrative purpose,
game drawing back to what is narratively interesting, but is still free to explore. "Not stuck in a corner in a dialg tree"

narrative purpose = NPC purpose

no time for questions :(

... Now time for rant-session.