HOME       LINKS     RESEARCH     PUBLICATIONS

Can We Talk? by DeSmedt & Loritz

Virtual Intelligence from Artificial Reality: Building Stupid Agents in Smart Environments by Doyle

New Challenges for Character-Based AI for Games by Isla & Blumberg

Conversational Agents for Game-Like Virtual Environments by Morris

The Artificial Emotion Engine ^TM, Driving Emotional Behavior by Wilson

Building Characters: A Form of Knowledge Acquisition by Pisan

Cognitive Multi-character Systems for Interactive Entertainment by Funge & Shapiro

Intelligent Agents for Computer Games by Nareyek

Intelligent Agents for an Interactive Multi-media Game by Nicholson & Dutta

A Proposal for an Agent Architecture for Proactive Persistent Non Player Characters by Mac Namee & Cunningham

DeSmedt & Loritz - Can We Talk?

This article describes what is required to get Non Player Characters (NPCs) “talking”. The authors state that giving NPCs more personality is a way of meeting demands made by game players for more immersive gaming experiences. Giving an NPC more personality can mean enabling the NPC to have memories, beliefs and opinions as well as the ability to express these to the player. The question that the authors are trying to answer is what does it take generate this kind of character, which the authors call a conversational agent.

The authors discuss the computational linguistic requirements of a conversational agent (such as parsing, discourse analysis etc) and the problems that this can present when trying to create discourse. The main problem is the ambiguity in human language. The authors believe that the solution lies in Knowledge Representation (KR). KR in computer games involves a small domain of knowledge and giving the NPC some “common sense” is very important. Issues that arise include how to incorporate extensional knowledge as opposed to intensional knowledge (intensional knowledge is something that is true of a class and any subclass. Extensional knowledge is knowledge about an individual of a class that may differ from the intensional knowledge). The authors finish by cautioning that these advances won’t happen immediately. However, when they do work, existing stories won’t just be enhanced, but new avenues of storytelling in computer games will be opened.

Go to DeSmedt & Loritz's "Can We Talk?"

http://www.cs.northwestern.edu/~wolff/aicg99/desmet.htm

Reference details for DeSmedt & Loritz's "Can We Talk?"

DeSmedt, B. & Loritz, D., 1999, Can We Talk? in Papers from the AAAI 1999 Spring Symposium on Artificial Intelligence and Computer Games, Technical Report SS-99-02, AAAI Press, pp 28-31.

Doyle - Virtual Intelligence from Artificial Reality: Building Stupid Agents in Smart Environments

This article discusses what is required to create interesting interactions with players in online gaming worlds (such as Everquest and Ultima Online). Currently, players provide most of the interactions, and are themselves the source of most of the dynamic change that occurs within these worlds. However, players are generally not able to support the game environment as well as characters that have been specifically designed for that purpose. The author believes that this purpose is better served by intelligent agents in the roles of Non-Player Characters (NPCs). The problem is improving the believability of these agents.

The author believes that this can be done in two ways. The first is to improve their "believability-enhancing behaviours" and the second is to make the NPCs more intelligent about the world that they reside in (better domain behaviour). The author uses virtual affordances - annotations - embedded in the environment. There are five types of annotations: emotional, responsive, problem-solving, role and game-playing. The annotations are combined with some abstract competencies in the agent. One disadvantage is that annotations can lead to large, complex representations of the environment. The author tested NPCs in a Multi-User Dungeon or Dimension (MUD), where they displayed some interesting behaviour. There are still problems in this approach - knowledge representation issues as well as limitations about how much information can be given via annotations.

Go to Doyle's "Virtual Intelligence from Artificial Reality: Building Stupid Agents in Smart Environments"

http://www.cs.northwestern.edu/~wolff/aicg99/pdoyle.pdf

Reference details for Doyle's "Virtual Intelligence from Artificial Reality: Building Stupid Agents in Smart Environments"

Doyle, P., 1999, Virtual Intelligence from Artificial Reality: Building Stupid Agents in Smart Environments in Papers from the AAAI 1999 Spring Symposium on Artificial Intelligence and Computer Games, Technical Report SS-99-02, AAAI Press, pp 37-41.

Isla & Blumberg - New Challenges for Character-Based AI for Games

This article is a discussion of possible next steps that could be taken towards creating virtual, autonomous characters that can convey emotion and are able to appear "intelligent". The questions that need to be answered,  according to the authors is how do simulated creatures perceive their world? How are their reactions handled? How do they go about determining their goals and then satisfying them?

One area to be addressed is Sensory Honesty. What should the character be able to perceive? Perception also includes the problem of pattern recognition. The character should also be able to demonstrate some level of anticipation and also imagination or planning. Anticipation needs to incorporate the concepts of reward and priming. Planning could be affected by emotional state. The authors believe that emotional state is also something that needs to be considered, beyond looking at a character's level of happiness or sadness. The final aspect of a character is how it learns. The authors believe that the best way is learning through episodic memory. There are a number of important questions that need to be answered, including how these capabilities of characters should be incorporated into a game and if they result in more interesting and immersive gameplay?

Go to Isla & Blumberg's "New Challenges for Character-Based AI for Games"

http://www.qrg.cs.northwestern.edu/aigames.org/papers2002/DIsla02.pdf

Reference details for Isla & Blumberg's "New Challenges for Character-Based AI for Games"

Isla, D. & Blumberg, B., 2002, New Challenges for Character-Based AI for Games in Papers from the AAAI 2002 Spring Symposium on Artificial Intelligence and Interactive Entertainment, Technical Report SS-02-01, AAAI Press, pp 41-45.

Morris - Conversational Agents for Game-Like Virtual Environments

This paper describes a proposal to develop conversational agents that also contain models of personality and emotion. The aim of these agents is to create characters that can display believable conversational behaviour. An important aspect of believable behaviour is that dialogue should be modified to communicate relevant knowledge and to display unique character for different agents. The conversational agent architecture is built on a natural language generation system and incorporates information about the agents’ personality, current emotional state and it’s beliefs about social relationships.

The agents are placed in the game of Cluedo, and develop episodic memory about events that occur during a time span. The conversational agent architecture consists of the following models and managers: personality model, emotional model, emotional affect manager, temperament model, social role model, other agent model, language use manager and a generic conversational architecture. Parameters of the models are changed for different agents, to provide unique personalities that react in different ways. Other parameters of conversational behaviour that need to be tuned include: word choice, sentence form (e.g. passive, active voice), turn management and introducing new concepts and information or changing the focus of the current conversation. Future developments include developing a greater understanding of the interplay between the parameters mentioned above and how this affects conversational behaviour, and to fully implement this model in software.

Go to Morris's "Conversational Agents for Game-Like Virtual Environments"

http://www.qrg.cs.northwestern.edu/aigames.org/papers2002/TMorris02.pdf

Reference details for Morris's "Conversational Agents for Game-Like Virtual Environments"

Morris,T.W., 2002, Conversational Agents for Game-Like Virtual Environments in Papers from the AAAI 2002 Spring Symposium on Artificial Intelligence and Interactive Entertainment, Technical Report SS-02-01, AAAI Press, pp 82-86.

Wilson - The Artificial Emotion Engine ^TM, Driving Emotional Behavior

This article describes the Artificial Emotion Engine, which is designed to provide emotional behaviours for autonomous, interactive characters. The Engine is based on belief that emotion comprises two conceptual properties: cognitive emotions and innate emotions. The Engine does not provide cognitive emotions, but deals with innate emotions as a current achievable goal. The Engine provides the emotion which can be the motivation for a character.

Emotions are divided into three layers: reactions, moods and the underlying personality, each of which has a different level of prominence. Personality is defined as an area in 3D Cartesian space, the three axes of which are Extroversion, Fear and Aggression. This area defines a set of personality traits that combine to make an aggregate personality. Mood is determined by perceived signals of punishment and reward, but modulated by personality. The Artificial Emotion Engine consists of nine modules that interact to create the appearance of emotion: Emotional Reactions, Punishment and Reward, Memory, Personality, Mood, Motivations and Self State. There are also Input and Output modules. Input receives punishment or reward signals. The Output module generates output in three formats: body and facial gestures, a semantic action plan and raw emotional data. The author concludes by stating that this engine provides a solid foundation to build more complexity and personality into characters.

Go to Wilson's "The Artificial Emotion Engine ^TM, Driving Emotional Behavior"

http://www.qrg.cs.northwestern.edu/aigames.org/2000/IWilson00.pdf

Reference details for Wilson's "The Artificial Emotion Engine ^TM, Driving Emotional Behavior"

Wilson,I., 2000, The Artificial Emotion Engine ^TM, Driving Emotional Behavior in Papers from the AAAI 2000 Spring Symposium on Artificial Intelligence and Interactive Entertainment, Technical Report SS-00-02, AAAI Press, pp 76-80.

Pisan - Building Characters: A Form of Knowledge Acquisition

This papers describes a knowledge acquisition approach to building computer characters, that uses model-based classification as the learning technique. The aim is to create characters that are able to learn from interactions with a user, and then adapt their behaviour. The first step is to create a model that contains all of the knowledge that the character will use. This kind of expertise is often difficult to encode. The character must decide on it’s next action based on it’s own observations and historical data.

In a deterministic world a character’s model converges to the world model. However, in a non-deterministic world the model of the world created by the character is the set of rules that most closely fits the available data. The experiment that the author explains in this paper is a simple predator-prey model. The predator must decide whether it will hunt or rest when it sees potential prey. The aim of the experiment was to determine how different levels of data availability change the quality and accuracy of the rules that the predator develops (variables include speed, hunger, terrain and prey type). As expected, the predator could determine the world better when it had more examples of data to work with. This article focuses on the behavioural aspect of characters, but uses an approach that can be applied to conversational agents. In that case, conversation could be used to change the attitude of another character.

Go to Pisan's "Building Characters: A Form of Knowledge Acquisition"

http://www.qrg.cs.northwestern.edu/aigames.org/2000/YPisan00.pdf

Reference details for Pisan's "Building Characters: A Form of Knowledge Acquisition"

Pisan,Y., 2000, Building Characters: A Form of Knowledge Acquisition in Papers from the AAAI 2000 Spring Symposium on Artificial Intelligence and Interactive Entertainment, Technical Report SS-00-02, AAAI Press, pp 66-68.

Funge & Shapiro - Cognitive Multi-character Systems for Interactive Entertainment

This article is a critical examination of some of the issues faced by researchers into Interactive Entertainment. The authors focus on cognitive modelling as an interesting new technique that can be used to improve game characters. Cognitive models go beyond the shallow emotional and behavioural models that have previously been used in game development. Cognitive modelling would allow for autonomous, semi-intelligent characters.

The authors are interested in what would happen if characters can attribute mental states to themselves, other characters or even the player? Representing mental states requires a language that is able to define behaviour in terms of knowledge, goals and intentions and can ensure that characters are able to access explicit knowledge of the environment. The authors give an example of a language that was able to convey the above properties. They also give an example of an attempt to implement mental states. One of the problems that arises from attempting to implement this sort of system is the real time performance. Some of the techniques that can be used to improve performance include: constraining the length of action plans that characters can search for and having a separate reactive behaviour system that can be used as a fallback. The authors finish by concluding that the idea of having separate reactive and cognitive systems can be taken further, for instance by making the reactive system a dynamic entity.

Go to Funge & Shapiro's "Cognitive Multi-character Systems for Interactive Entertainment"

http://www.qrg.cs.northwestern.edu/aigames.org/2000/JFunge00.pdf

Reference details for Funge & Shapiro's "Cognitive Multi-character Systems for Interactive Entertainment"

Funge, J., Shapiro, S., 2000, Cognitive Multi-character Systems for Interactive Entertainment in Papers from the AAAI 2000 Spring Symposium on Artificial Intelligence and Interactive Entertainment, Technical Report SS-00-02, AAAI Press, pp 27-29.

Nareyek - Intelligent Agents for Computer Games

This article gives on overview of the features of computer games that are of interest to AI researchers, and game - relevant agent architectures. The features of computer games that make them interesting and more complex than traditional games such as Go and chess include: the game is played in real time, the environment is dynamic, it is usual for a character to have incomplete knowledge of the world and restricted resources.

The author states that AI in computer games is mainly about characters. Agents are expected to display intelligent seeming behaviour, which involves having goals, sensing aspects of the environment and being able to execute actions. The different types of agents that the author discusses are reactive, triggering, deliberative, hybrid and anytime agents. Reactive agents respond to sensor information by performing a specific action, which is implemented by if-then rules. Triggering agents use finite state machines to include past information in rules and are able to perform sequences of actions. Deliberative agents rely on a planning system that consists of an initial world description, a partial world description and a set of actions that map from the current world description to another description. Hybrid agents combine an offline deliberative planner and a reactive component. Anytime agents have a continuous transition from reaction to planning and improve their current plan iteratively. The article concludes with a brief description of the agent developed in the EXCALIBUR project.

Go to Nareyek's "Intelligent Agents for Computer Games"

http://www.ai-center.com/references/nareyek-02-gameagents.html

Reference details for Nareyek's "Intelligent Agents for Computer Games"

Nareyek, A., 2002, Intelligent Agents for Computer Games, in Marsland, T. A., and Frank, I. (eds.), Computers and Games, Second International Conference, CG 2000, Springer LNCS 2063, 414-422.

Nicholson & Dutta - Intelligent Agents for an Interactive Multi-media Game

The aim of this paper was to develop intelligent agents in a game world that are able to interact with the player within predetermined scenarios. The agents are constrained in the actions that they can take – they do not develop plans of actions to achieve goals. The choice of an utterance or action made by an agent at any point in the game is determined by a number of factors. The factors are: the agent’s personality profile, the mood of the agent, relationships with other agents and the player and the nature of actions or utterances that have preceded the current time.

Each agent has a set of personality attributes determined by the Big Five Factor Structure. The five factors are introversion-extroversion, pleasantness, conscientiousness, emotional stability and intellect. Moods are determined by two scales: happy/sad and secure/anxious. Moods for each individual have upper and lower bounds which are determined by the personality of the individual. Mood swings are modelled using the laws of motion under gravity. Returning to the default mood is modelled as damped simple harmonic motion. The relationship between two agents is determined by two factors – the quality of recent interactions between them and the attitude of the second agent towards the first. So far, only a simple prototype of this model has been implemented. However, the agent model presented in this article is straightforward and effective.

Go to Nicholson & Dutta's "Intelligent Agents for an Interactive Multi-media Game"

http://www.csse.monash.edu.au/~annn/cv/pub.html

Reference details for Nicholson & Dutta's "Intelligent Agents for an Interactive Multi-media Game"

Nicholson, A.E. & Dutta, A., 1997, Intelligent Agents for an Interactive Multi-media Game, in Proc. of the International Conference on Computational Intelligence and Multimedia Applications (ICCIMA'97), pp 76 - 80, Gold Coast, Australia.

Mac Namee & Cunningham - A Proposal for an Agent Architecture for Proactive Persistent Non Player Characters

This report introduces the concept of proactive persistent non player characters (NPCs), to counter the current trend of NPCs in games that are shallow and uninteresting. These NPCs are always modelled to some extent even when the player is not in the vicinity of the character. Persistent NPCs are implemented using intelligent agent architectures. The motivation behind this concept is that agents that are capable of acting in an interesting way will add to the sense of immersion that a game-player feels. The agents need to maintain the illusion of believability to add to gameplay.

Each PPA has needs, desires and beliefs of its own. The authors discuss the current types of intelligent agents: reactive, deliberative, hybrid, broad/believable, emotional, social and anytime agents. The goals that the PPA architecture is designed to achieve are: autonomy, social ability, reactivity, pro-activeness, persistence, scalability, extendibility and configurability. The four major sub-systems of the architecture are: behaviour system, social system, goal based planner and the schedule. The behaviour and social systems are reactive, while the goal based planner is deliberative. In addition, the agent has a selection mechanism that acts as a mediator between the different sub-systems and also chooses a course of action. The agent also has an underlying knowledge base. This article does not present any details of implementation, as this architecture is currently only a proposal.

Go to Mac Namee & Cunningham's "A Proposal for an Agent Architecture for Proactive Persistent Non Player Characters"

http://citeseer.nj.nec.com/macnamee01proposal.html

Reference details for Mac Namee & Cunningham's "A Proposal for an Agent Architecture for Proactive Persistent Non Player Characters"

Mac Namee, B. & Cunningham, P., A Proposal for an Agent Architecture for Proactive Persistent Non Player Characters, Departmental Technical Report TCD-CS-2001-20, Trinity College Dublin, 2001.