Archive: April 17th, 2009

Summarizing research

[Research] (04.17.09, 8:06 pm)

This documents some of the general notes on my research that have been floating around in my head. Important elements to be discerned later are:

  • Define the problem space, “Here is the problem that I have defined”
  • Define methodology, resources, approach, “Here are my tools to address the problem”
  • Define what results and conclusions have been reached, “Here is the deliverable”

I apologise for the extremely rough nature of these notes, but I want to put them up somewhere rather than having them disappear.

Summarizing research:

* project goal: adapting novels into games.
* focus: Pride and Prejudice

* Narratives, and novels in particular are already simulations
* Focus for adaptation should be WORLD, not plot
* world works according to model (mechanics)
* model is tied to the work, makes the work what it is (functionally)
* interpreting a model from a work is intrinsically creative

* games and simulations about mechanics (rules)
* games can simulate mechanics of a system (adaptation)
* vocabulary of mechanics used commonly in games is a set of tools

* adaptation involves 2 parts: carry-over, adaptation (analogous extension)
* can involve moving tropes/conventions from one medium to another (mystery novel ->

adventure; action film -> action game)
* specific thing to look at is conflict resolution

* need a critical aesthetics for game adaptations
* we do not want fidelity criticism, but there are games that are better and worse
* believe better adaptations will adapt the underlying mechanics well
* what mechanics are necessary and sufficient for the adaptation to work?

* technology is not a barrier for making game adaptations
* technology may enable better representation of characters

* there is barrier in games that disinclines adaptation of other types of fiction
* we must be open, as a culture, to recieving new types of games

* concern is not reproducing emotion exactly as in original
* concern is not reproducing plot of original
* plot is up to player, player has tools to drive own experience

* perspective is integral to perception of a work
* perspective is tied to how we understand systems
* adaptation of systems gives power to expose new perspectives

* many adaptation ideas lead to percpetions of game mechanics that depart from mainstream
* potential for advancement of reception, understanding of games

* meaning in novel comes from understanding novel as social world, reader simulates
* reader understands (or can understand) characters as deep, with inner lives
* novel is a kind of social laboratory (Lukacs)

>>>> more specific to P&P

* world is social; mechanics are social
* mechanics must be built from social rules, hence sociology
* social mechanics and games reveal character

* for simulation, characters must have autonomy
* we depart from conventions of planning in AI (conflicting goals)
* characters act according to situation and according to social context (thus, cannot be

transplanted without addressing context)

* thus, to simulate P&P, necessary to find how to describe social contexts, and how they

* development of AI is thus not in character, but in social world itself
* this method can be applied to other texts

* focus should not be to derive supremely general schema for human behavior in any context
* focus is about specific focused contexts, and representing those clearly
* avoid AI complete pitfall

AI papers galore!

[Readings] (04.17.09, 7:55 pm)

I decided to read a bunch of AI papers to make sure that my representations and critiques of AI in games is accurate. Over the course of this, I’ve realized something important. This is that I am not trying to develop a simulation system that will replace AI in games in social environments. I am not trying to depose planners completely. Rather, I want to make a more subtle argument. The success of an AI artifact is dependent on how closey knit the underlying technology is to the domain it represents. So, in some circumstances, planning is extremely effective, but in other circumstances, it is much less so. Planning is also not a uniform and monolithic infrastructure, it is a relatively loose system of algorithms. Planning frameworks have been developed that incorporate (or attempt to incorporate) situated and reactive reasoning, action repair, emotional models, and interaction with other systems and agents. While it is not monolithic, it still represents a perspective of how agents act and think, and thus conveys a particular model.

Included below is a list of a bunch of the papers that I have read, with notes describing some of the relevant take-aways from the papers. This is generally particular to my work, and may be of limited use to other readers. The papers are not in a particular order.

Mao and Gratch: Social Judgment in Multiagent Interactions (2004)
paper is about judgement of attribution of responsibility in social settings
tied into military system of authority

McCoy and Mateas: The Computation of Self in Everyday Life (2009)
looks at applying Goffman to character simulation, manifests as social games
adaptation target: Sex and the City

Perlin and Goldberg: Improv (1996)
early work, involves framework for animation and behavior
scripted behavior systems

Geib: The Intentional Planning System: ItPlanS (1994)
builds from STRIPS action model (with preconditions and postconditions)
attempts to exchange preconditions with intentions, done via simulation
still about robot control

Magerko, Laird, Assanie, Kerfoot, Stokes: AI Characters and Directors for Interactive Computer Games (2004)
describes goals of setting up interactive drama:
computer games with nonviolent, plot-driven stories
focus is in author centric model, with working around players
target is newly authored artifact: Haunt 2

Cavazza, Charles, Mead: Characters in Search of an Author (2001)
model is character-centric approach
addresses issues of narrative and authorial and user control
adaptation target is “Friends” scenario
system is built from model of Barthes S/Z
planning model is consistent with sitcom genre

Pizzi and Cavazza: Affective Storytelling based on Characters’ Feelings (2007)
attempting to develop computational character system based on emotional theory
this is based on appraisal and coping
adaptation domain is Madame Bovary

Cavazza, Pizzi, Charles, Vogt, Andre: Emotional Input for Character-based Interactive Storytelling (2009)
adaptation domain is Madame Bovary
about using emotional voice input to interact with the program

Si, Marsella, Pynadath: Thespian: Modeling Socially Normative Behavior in a Decision-Theoretic Framework (2006)
focus is interactive drama
built around modeling social norms
military goal, norms are essentially a means to an end of uncovering information

Magerko: A Proposal for an Interactive Drama Architecture (2002)
proposes a model for interactive drama that uses a director
structure of drama itself is composed of scenes
director is attempt to resolve conflict between authored plot and user agency

Cavazza, Charles, Mead: Emergent Situations in Interactive Storytelling (2002)
applies planning model to sitcom genre
recognizes need for situated reasoning and action repair within planning model

Magerko and Laird: Mediating the Tension between Plot and Interaction (2005)
describes director model wherin director makes predictive planning
director does simulation of world based on player model
involves reconciling errant player behavior

Peinado, Cavazza, Pizzi: Revisiting Character-Based Affective Storytelling under a Narrative BDI Framework (2008)
uses Madame Bovary domain
develops alternative model of BDI (belief, desire, intention) as Narrative BDI
first looks at Shakespearian model

Gratch: Why You Should Buy an Emotional Planner (1999)
Applies emotional models and appraisal theory to planning formalism
aim is to reconcile areas where planning has trouble: conflicting goals, limited resources, imperfect information
primary focus is construal and assignment of blame/responsibility
odd examples with conflict over moving a car
but leads to model with system of personalities in agents

Geib, Webber: A Consequence of Incorporating Intentions in Means-end Planning (1993)
incorporates situated reasoning into planning model
this means replacing preconditions with alternative approaches
preconditions use generation conditions and execution conditions
alternatives are robust failure, replanning, and action repair

Bates: The Role of Emotion in Believable Agents (1994)
concept of believability, comes from animation (esp Disney)
describes Woggles world
emotion necessary for recognition of personality and believability
uses emotion theory of Ortony, clore, and Collins
characters in most video games show no reaction to violent world around them

Goldberg: Avatars and Agents, or Life Among the Indigenous Peoples of Cyberspace (1998)
poses model of animation and behavior that comes from Perlin’s Improv
heavy importance of scripted behaviors, but scripts are extended to have robust extensibility
model comes from numeric properties, and then matched with importance values, allowing weighted selection

Pizzi, charles, Lugrin, Cavazza: Interactive Storytelling with Literary Feelings (2007)
initially outlines focus of constructing interactive storytelling experience with focus on literary feelings
introduces Madame Bovary domain
domain requires focus on characters’ feelings, and planning is about long term objectives — this is HUGE change from traditional models
uses clear system of adaptation, from feelings, to literary analysis, to computational model

Christian and Young: Comparing Cognitive and Computational Models of Narrative Structure (2004)
uses traditional cognitive science system of models to look at user’s understanding of narrative in virtual world
cognitive models are very abstracted, very structural
domain is Unreal spatial puzzles (levers, lifts, bridges, etc)
user study to match model of knowledge structures to model of computational system

More Cellular Automata

[Genetic Image,Toys] (04.17.09, 6:56 pm)

I’ve been very interested in doing experiments with cellular automata and other soft of image generation work, and amid reading AI papers, I’ve done some miscellaneous code experiments. Right now I’ve built a nifty little system that is able to handle many types of CAs, and can represent in space in several ways, represent their contents in several ways, and render them in a variety of ways as well.

I’ve included a little demo applet which handles a small diffusion-like CA, and is hopefully a sign of some things potentially to come.

Looks like applets don’t work for you