Skip to content

Mind Magic Spells, early sketch

Mind Magic Spells, early sketch by mimmi
Mind Magic Spells, early sketch, a photo by mimmi on Flickr.

I found this when packing – we are moving offices. It is an early sketch of what spells players can use depending on what mood their characters are in.

Looking at this I remember that the game play I was considering then did not include that when a strong emotion was felt, a manifestation of the it  would be spawned. (Sorrow for example, would give a “Sail of Sorrow”. It then keeps casting the spell Wet Net of Tears on passers-by). The characters’ emotion, sorrow, diminishes when the manifestation is spawned, and the mood is not affected so extremely as it was when the game play looked as it does in this sketch.

As the game play is now, there is great focus on mood-balancing, but in this sketch it was even more focussed on it. There was no emotional ‘release’ in forms of manifestations diminishing the the emotion values, and also there are a few things here I had completely forgotten, but I remember them now, looking at this picture.

Combinations minimum the inner mood value with maximum or minimum outer mood values would result in death of the player characters, but in spectacular ways.

Kamikaze (lower-left) would lead to death of self and everyone in proximity, both friend and foe. This would happen if the character is completely depressed and at the same time utterly furious (IF innerMood < minVal AND outerMood < minVal)

Self Sacrifice would cause the death of self, but give the party full energy and resistance (health and mana). It would happen if a character was deeply depressed and jubilant at the same time. (Not very likely) (IF innerMood < minVal AND outerMood >maxVal)

Combinations of maximum harmony (inner Mood) with extremes on the outer mood scale wouldn’t cause death, but powerful effects – good things for party on the positive scale, and things bad for opponents on the other end.


Four case studies of AI based game design

In this article we present four case studies of AI-based game design. Each of these games and AI systems were designed by one of the authors, and the projects are in varying stages of maturity. The Pataphysic Institute and Mind Music are completed games, although further iterations may be done in the future. Rathenn has a completed computational prototype, and is currently undergoing a second iteration of both the game and AI system with plans to release the game at the end of 2011. Prom Week has a fully playable prototype and is nearing a beta release. Mismanor is in the early design phase with a computational prototype scheduled for late Spring 2011.

A. The Mind Module, the  Pataphysic Institute, and  Mind Music

The Mind Module (MM) is an agent architecture that give characters personality traits, emotions and moods. Characters have sentiments, which is their individual likes and dislikes for objects and object-types in a world, such as fear of spiders or love for another character.

The MM has been used in several experimental prototypes, necessary for seeing to what extent the MM adds to the playing experience. Each prototype in which the MM has been used has given pointers towards what can be explored and improved for the next iteration. Early in the process of the work with the MM (Eladhari 2003) the author was curious to establish what effect the MM could have, if added as an extra feature to a ‘typical’ MMO (as described by Bartle 2003). As the research developed it seemed more meaningful to create prototypes where the game mechanics were increasingly based on the MM. Having started out with the aim to find general solutions to questions regarding story construction and characterisation for typical MMOs with the use of psychology-inspired AI-applications this work developed towards more and more specific solutions. Here, two of the prototypes are described briefly, The Pataphysic Institute and the Mind Music application.

The Mind Music application (Eladhari et al 2006) was an experiment in expression, where the aim was to find a more intuitive way to represent dynamic emotional states. Instead of representing mood and emotion of the own avatar by numericial values or with body language (which in real-life is used to infer other peoples emotional states), we wanted to the player to hear the soundtrack of the own avatar’s mind. In a small arcade style game we used time signature (groove) and harmony to express moods, and leit-motifs (personal themes) to express emotions.

The Patazphysic Institute (PI) (Eladhari 2010) is the most extensive prototype, where the game mechanics are most intertwined with the MM. It is a game based on player cooperation, where all actions possible to perform depend on the avatars own personality, emotional state and memories of previous interactions. PI is a multi-player world in 3D that is played via a web browser. Players need to defeat physical manifestations of negative mental states. In order to do so, they can cast spells on them, but the spells available are constrained by avatars’ personalities and current moods. Players can chance each others moods by using ‘affective actions’. The system supports players to act ‘in chararacter’ with their personality and current moods – happiness, harmony, anger and depression is the basis for the core game play.

Players co-create the content of the world. If an avatar is, for example, burdened by guilt they need to ‘externalise’ the emotion to get rid of it. The player can author a manifestation of guilt (for example a martyr grandmother, complete with short dialogue and martyr behaviour), which becomes a boss. To free the player from the guilt-burden a team of players need to figure out how to put the martyr grandmother ‘to rest’ using their special personality abilities.

B. Rathenn and Launchpad

Rathenn is an under-development 2D platformer game that tightly integrates procedural content generation (PCG) into the game’s mechanics and aesthetics. Mechanics-wise, Rathenn is consciously positioned within the tradition of the platformer. The basic actions available to the player are those typical of existing platformers, including jumping over gaps, killing enemies, avoiding stompers, and climbing ladders. The game consists of a series of level segments connected by ladders; level segments are generated at play-time at the end of a climbed ladder. A key aesthetic in Rathenn is that of discovery: the player is encouraged to explore both the physical space of the level and the generative space of the procedural level generator through climbing ladders. Each ladder is color-coded, and each color influences the generator in different directions. For example, climbing a red ladder will increase the frequency of enemies appearing in the level, or climbing a purple ladder increases the avatar’s movement speed and jump velocity. The first iteration of the game is available online.

Rathenn uses the Launchpad AI system (Smith et al 2011), which is a rhythm-based level generator for 2D platformers that are characterized by dexterity challenges. The rhythm-based approach is based on observations (Nicollet 2004, Bleszinski 2000, Kremers 2009 pp. 263 – 267) that rhythm and pacing are deliberately designed and crucial to challenge in this kind of game. Launchpad allows a designer to refine its generative space by manipulating parameters that have a clear and intuitive impact on generated levels. These parameters dictate a general path that the level should take, the types and frequencies of geometry components, and how collectible items are distributed throughout the level. Adjusting these parameters can drastically alter the generated level.

Rathenn is currently in its second iteration on both the game and AI system: there is now a concept of tiered level generation, which monitors the player’s progress along a path and provides more challenging level elements as the player improves. For example, the player will first encounter spikes, then moving enemies, and finally flying enemies as they progress down the “enemies” path in the generative space. This addition requires changes to both the game design and the AI system. There is also now a story to the game, determined from the affordances of procedural level generation. The game is set in a dreamscape, and the player is striving to overcome different fears characterized by the various types of challenges the game can provide.

C. Prom Week and Comme il Faut

Prom Week is a social simulation game about the dramatic week leading up to a high school prom. Players of Prom Week indirectly sculpt the social landscape by having characters engage in social exchanges with each other. There are a large number of varied results of these social exchanges—ranging from mild changes in sentiment to characters professing their eternal love. The possible exchanges and results are informed by over 3500 sociocultural considerations managed by the AI system Comme il Faut (CiF) (McCoy et al 2010). Through shifting the interpersonal relationships and learning the personal intricacies of the characters, the player can solve a series of “social puzzles”, such as making the class nerd the prom king, or bringing peace between feuding jocks and preppies.

CiF is an AI system that enables an interactive, authorable model of social interaction for autonomous agents. Social interactions are multi-character actions whose function is to modify the social state existing within and across the participants. Dramaturgical analysis (Goffman, 1959) is the basis for extending the power of social interactions from their initial form in Façade (Mateas & Stern, 2005)—where their variation in performance was implicitly encoded in behaviors and they were not reusable between characters—to their current, reusable form that supports performance variation explicitly. Through the use of social interactions along with additional encoded social context, CiF models the social aspects of virtual worlds.

Prom Week has undergone several iterations in both the game and AI design. First, a computational prototype of CiF was developed and was paired with a paper prototyped game system. As this computationally-assisted paper prototype was tested, the game play provided a critical perspective on the domain of the AI system: that early prototype of CiF was more about satisfying the psychological needs of the characters in the game world rather than providing a play experience about the relationships between characters. In the new, broadened requirements of emphasising the more social aspects of game play, CiF was re-engineered to include affordances like subjective social feelings and relationships between characters.

D. Mismanor and GrailGM

Mismanor is a role-playing game in which the core mechanic is social interaction instead of combat, with a focus on supporting player-driven story. The story emerges and changes based on the choices the player makes, who they interact with, and how they choose to interact with the other characters. With such a strong focus on social interaction, the game setting was chosen to be a manor party set in 1930s England in which the player character begins to learn and unravel the mystery of the history of the family that owns the manor through interacting and helping the other characters in the game. Mismanor uses two AI systems, CiF (as described above) which handles the game level social interactions, and GrailGM which manages the story structure and quest system.

It was necessary to make modifications to the CiF system to work within the role-playing game (RPG) genre. For instance, we needed to add support for items and quest giving, as well as adding the concept and structure for social statistics (stats), similar to typical RPG stats but for a social domain. The player and NPCs each have social stats such as confidence, perception, and persuasion, associated with them, as well as traits, statuses, and relationship networks. This framework is used to help govern what types of interactions are available for each character and the player. We chose to make giving a quest one of the social actions available to the NPCs, allowing CiF to govern the social logic involved with when to assign a quest, while the specifics of what quest to give are left to GrailGM. Finally, Interacting with items uses the same type of logic as interacting with NPCs; depending on the history, player state and world state, different ways of examining, taking, and using items will be available to the player.

The GrailGM layer supports dynamic quest selection as well as story-level author goals. Which quest is chosen to offer to the player is based on story-level authorial goals, the type of quest, past events, traits and statuses of the character the player is talking to, and the relationship between the character and that NPC. Each quest has pre-conditions specifying what type of NPC can give that particular quest, along with what game state is necessary for the quest. A quest has a success state, and a failure state; achieving either state will allow the story to progress in different ways, and have different effects on the state of the world and how the characters view the player. By defining the completion as a state instead of a series of actions that must be performed, the player is able to choose how they complete a quest.

Mirjam P. Eladhari,
Gillian Smith
Josh McCoy,
Anne Sullivan,

April 2011, Santa Cruz, CA.


Bartle, R. (2003). Designing Virtual Worlds. New Riders.

Bleszinski, C. (2000). The art and science of level design. [Online]. Available:

Eladhari, M. P. (2010, September). Characterising Action Potential in Virtual Game Worlds applied with the Mind Module. Retrieved from

Eladhari, M., & Lindley, C. A. (2003). Player Character Design Facilitating Emotional Depth in MMORPGs.

Eladhari, M., Nieuwdorp, R., & Fridenfalk, M. (2006). The Soundtrack of Your Mind: Mind Music – Adaptive Audio for Game Characters. Proceedings of the 2006 ACM SIGCHI international conference on Advances in computer entertainment technology (p. 54–es). ACM. doi: 10.1145/1178823.1178887.

Goffman, E. (1959). The Presentation of Self in Everyday Life (1st ed.). Anchor. Retrieved from;path=ASIN/0385094027.

Kremers, R. (2009). Level Design: Concept, Theory, and Practice. Peters/CRC Press.

Mateas, M, & Stern, A. (2005). Structuring Content in the Façade Interactive Drama Architecture (Vol. 3).

McCoy, J., Treanor, M., Samuel, B., Tearse, B., Mateas, Michael, & Wardrip-Fruin, N. (2010). Authoring Game-based Interactive Narrative using Social Games and Comme il Faut.

Nicollet, V. (2004). Difficulty in dexterity-based platform games. [Online]. Available:

Smith, G., Whitehead, J., Mateas, Michael, & Treanor, M. (2011). Launchpad: A Rhythm-Based Level Generator for 2D Platformers. IEEE Transactions on Computational Intelligence and AI in Games, 3(1), 1-16.

Whiteboard: wicked AIGD

Whiteboard: wicked AIGD by mimmi
Whiteboard: wicked AIGD, a photo by mimmi on Flickr.

We had a pretty heated discussion – each of us had a pen,  several favorite arguments, we all had some favorite AI philosopher’s quote to blast. We (Anne, Gillian, Josh and I) managed to confuse ourselves and each other to an amazingly high degree! Then, we focussed on applying all this (quite knowledgeable!) confusion on our own games and AI systems… and were able to diminish our confusion a little. To an at least tolerable level.

Some of these discussions continued in a chaotic google doc. Last word isn’t said yet, and if we’re lucky, it never will be. At least not as long as we build things – new issues will most likely surface again and again. Wicked problem space indeed.

Whiteboard: AI based game design

After heated discussions we (Josh, Anne, Gillian and I) ended up with these scribbles on the whiteboard. Anne later made a more orderly diagram of this. It illustrates the process of AI-based game design. The AI system affords certain mechanics and aesthetics for the game design, while the game design provides the context in which the AI operates. Domain information, such as theories.

Unthinking machines?

MIT Technology Review summarised an interesting panel discussion, Unthinking Machines, where the lack of progress of AI during the last decades was discussed.

In the panel, several reasons for this were pointed out, but what resonated most strongly for me was that the panelists brought forth what I use to call the “tool-box approach”. It is not so much that researchers would focus on the tool rather than on solving a problem – no one wants that. What I think is a problem is that communities and conferences tend to form around methods and technologies – and this leads to a situation where the focus is on the method, since that is what the community has in common. The problems that one attempts to solve become those that the technology and method can solve, and then it can be difficult to break out of that in order to address fundamental problems. But that’s what we need to do. When designing AI for games – innovative games –  what else can we do?

…naturally I don’t agree with all said in the article, especially not with that ‘intelligence’ needs to be modelled from human intelligence. (For start, it could be a good thing to know what that is, and then, there are many other sorts of intelligences or intentionalities to model and invent.)

/Mirjam, Tuesday morning, 10th of May, downtown Santa Cruz.

AI based Game Design – we will explore!

This is our first take on what type of resources we would like to have here. While we work it out, will be at (obs login required)