Difference between revisions of "Algorithmic Agents"

Revision as of 19:18, 13 July 2014

Algorithms created to provide what appears to be intentional behavior.

Many games worlds contains more entities such as animals, people, monsters, or robots that have behaviors that are not decided by players. To make this possible these entities are instead have more or less complex rules, algorithms, that decide which actions they should take. The simplest only contain a couple of rules taking not consideration to what the other players or entities have done while the most complex have processes for learning under which contexts one should perform what actions.

Examples

Already most of the earliest computer games, including Nimatron, OXO (a computerized version of Tic-Tac-Toe), Asteroids, Space Invaders, and the Bomberman series made use of algorithms to control opponents to the players. It is still used in many games, e.g. in the Assassin's Creed, God of War, Need for Speed, Doom, Quake, and Civilization series. In some of these cases (i.e. Tic-Tac-Toe, Need for Speed, Civilization, and, somewhat paradoxically, multiplayer versions of Doom, Quake), the Algorithmic Agents replaces human opponents while in other cases being part of the game world. The Left 4 Dead series uses Algorithmic Agents not only for the infected that attack the players' characters, but also to control other player characters if there are not four people available. In all these case the Algorithmic Agents are also used to provide opponents but in games such as Fable II, Fallout series, NetHack, and Torchlight the Algorithmic Agents also control animals that accompany the players' characters.

Algorithmic Agents are also used to provide a basis of behaviors to agents in a game which players can then modify, as for example in the Lemmings and Sims series, or indirectly control, of which the Black & White series is an example. Games such as RoboRally and Space Alert, where players have to choose several actions together before they are enacted, can also be seen as examples of games that use Algorithmic Agents even if players create the algorithms/instructions and the outcomes are fixed unless some external factor is changed (e.g. the position of the other robots in RoboRally. MUDs^[1] typically give experienced players privileges to modify the game environment, and for those supporting more generic coding such as LPMUDs^[2] this can be in the form of independent entities capable of initiative and adaptive responses. In contrast, programming games such as Crobots and P-Robots challenge players to create better algorithms than other players before gameplay begins.

Using the pattern

The use of Algorithmic Agents by necessity require game designer to consider what Agents should exist in the game and what algorithms control their behavior. Which algorithms to use is not only a question of creating them, but also considering who can create them. The most common case is that they are created in advance by game designers but another possibility is to allow players to create the algorithms. The simplest form of this is Action Programming (as for example in RoboRally) and can be seen more as command sets in that they do not take input from the environment while being executed. Examples of games supporting actual coding include both pure programming games, such as Crobots and P-Robots, as well as MUDs, which use it to provide user-created content.

The types of Agents available depends primarily on how they relate to players and their goals. Enemies have Preventing Goals compare to the players' goals and are typically Units or Boss Monsters but Algorithmic Agents can also be used to control Avatars or Characters, thereby creating AI Players. The latter can also be used to create Companions so Teams can be created with players (examples include various humans characters in the Fallout series and animals in NetHack and Fable II), either through having Mutual Goals or having individual Continuous and Supporting Goals in relation to the players' goals. Enemies and Companions may both be NPCs but these may also have their Own Agendas independent of player goals. Agents may of course also change what role they have during gameplay, either due to Narration Structures which may in turn cause Surprises or as effects of players' actions through use of the pattern Actions Have Diegetically Social Consequences. One example of this can be found in the Fallout series, where players' Companions may abandon or attack players who start behaving differently than when they started to cooperate. What constitutes relevant behavior for all these types of Algorithmic Agents depend on what ambitions for Thematic Consistency exist.

However, players may not perceived Enemies, Companions, or NPCs as Agents if their behavior is too predictable^[3] even if the entities diegetically convey that relation and behavior. Instead, players may reduce them to Converters, Containers, Obstacles, Self-Service Kiosks, or Traps. Although this may be impossible to hinder if players achieve Game Mastery or simply play long enough, there are many ways to make this take longer time. Initiative and Contextualized Conversational Responses can all make the Algorithmic Agents function more in relation to the current context in their Game Worlds, while Awareness of Surroundings and Actions Have Diegetically Social Consequences forces players to consider what the Algorithmic Agents can sense. Own Agenda, Sense of Self, and Goal-Driven Personal Development all show ways of how to prolong the time until the Algorithmic Agents are no longer seen as Agents (often this include an application of Enforced Agent Behavior in comparison to other similar Agents).

For games aiming at Replayability, Open Destiny can also be required since otherwise overarching patterns for the Algorithmic Agents may be found and their apparent agency might lessen. Many of these solutions require the development of AI systems and require extensive testing to ensure their functionality is what is intended. A complementary approach to this creating Unpredictable Behavior through the use of Randomness or significantly complex rules controlling the behavior. Ambiguous Responses can also be used to make the illusion of intentionality last longer since players have more possibilities to read intentionality into the actions (an approach made famously by the computer program ELIZA^[4]).

A less common use of Algorithmic Agents is to interweave it with a player's control of an Avatar or Character. Some games, e.g. the Super Mario series and Fable II, use them when players' have not provided input for extended periods of time to indicate diegetically that the Avatar or Character is becoming bored but return control to the players as soon as input is provided. Other games, e.g. the Lego Star Wars series and the Left 4 Dead series, have them as backup AI Players that can step in when Game Pauses are desired, and by doing so enable Drop-In/Drop-Out gameplay. Concurrent shared control between human players and Algorithmic Agents has been discussed in research^[5] for affecting the behavior Characters in general, and exists in FPS games through aimbots^[6]

Diegetic Aspects

Algorithmic Agents do not themselves exist in a Game World but since they are typically tied closely to specific Avatars, Companions, or Enemies, it is necessary to consider both how the Algorithmic Agents should be able to observe what happens in the Game World and how they can act in it.

Interface Aspects

In Meta Games focused upon creating Algorithmic Agents it is necessary to include interfaces for this, often including functionality to ensure that the algorithms are properly expressed and the possibility to test them. In games with Companions controlled by Algorithmic Agents it may be necessary to provide specific interface components so that players can modifying their behavior, something which exist for example in the Fallout series.

Narrative Aspects

Algorithmic Agents are necessary if players are to be able to interact with NPCs that exist in a Narration Structure, and in this case it is important that they maintain the game's Diegetic Consistency if one aims at evoking an Emotional Engrossment to the unfolding storyline. Providing Actions Have Diegetically Social Consequences to Algorithmic Agents allow them to show Emotional Attachment to what happens in the game world.

Consequences

Algorithmic Agents are powerful tools that can make sure a game includes Conflicts, since game designer can create them with Preventing Goals to the players' goals and be sure that these will be acted upon (which is not always the case when given to players). However, Algorithmic Agents can just as well used to support Cooperation through Companions. Regardless, the Algorithmic Agents can provide both Complex and Challenging Gameplay and this may be incorporated into Handicap Systems and Difficulty Levels. A side effect of this is that the Algorithmic Agents can be used to modulate how much Tactical Planning a game requires, and how difficult this is. Regardless, since most Agents in games interact with the game through actions in a Game World, Algorithmic Agents is a very common component when designing Game Worlds.

Since Algorithmic Agents can be used to create AI Players, it makes it possible to make Multiplayer Games into Single-Player Games and thereby providing players with a way to have a Smooth Learning Curve at the expense of Social Interaction. Another effect of this is to allow Game Pauses and Drop-In/Drop-Out gameplay styles in Multiplayer Games. Of course, Multiplayer Games can be turned into Zero-Player Games by the same approach, although this is typically only done for supporting No Direct Player Influence through letting players of a Meta Game create the Algorithmic Agents. When the use of Algorithmic Agents are temporary, i.e. they replace players for limited periods of time, this provides support for Interruptibility.

When made by the game designers or developers, Algorithmic Agents provides Enforced Agent Behavior and can offer ways of achieving Predictable Consequences on more general levels of a game. Their behaviors may however take time to understand and until this is achieved they may instead give Limited Planning Ability. In contrast, player-generated algorithms offers players some Creative Control and may be used to support Mules. This also typically encourages Stimulated Planning and Strategic Planning since creating the algorithms requires an understanding of the game system regardless of a specific game state. Even if the Algorithmic Agents typically are abstract construction, they might be considered in this case be viewed as Player Created Game Elements, especially when the creation is linked to Characters or Avatars that would otherwise not be present in the game. When the creation of the algorithms are part of the set-up phase, Algorithmic Agents can create Meta Games where the inner game has No Player Influence and might work against players having an Exaggerated Perception of Influence. This allows programming games such as Crobots and P-Robots to let players have Indirect Conflicts with each other in the sense that they try to create Algorithmic Agents that are in Conflicts with each other (due to Preventing Goals such as Elimination) but that these players themselves do not perform confrontational actions. When possible during gameplay, the creation of Algorithmic Agents by players is an example of Player Created Game Elements and player-controlled Game Element Insertion.

The combination of players and Algorithmic Agents influencing the actions of the same Characters or Avatars can lead to a conflict over the amount of agency players have, and for this case the pattern can work against Empowerment and Exaggerated Perception of Influence.

While Algorithmic Agents often are realized through code running on computers and thereby set a bound on Ubiquitous Gameplay to require computers, this is not always the case. Race for the Galaxy and Power Grid have expansion with Algorithmic Agents contained on Tiles.

Relations

Can Instantiate

Actions Have Diegetically Social Consequences, Agents, AI Players, Challenging Gameplay, Complex Gameplay, Creative Control, Enemies, Enforced Agent Behavior, Game Element Insertion, Interruptibility, Limited Planning Ability, Meta Games, Mules, No Player Influence, Player Created Game Elements, Predictable Consequences, Tactical Planning

with Actions Have Diegetically Social Consequences

Emotional Attachment

with Companions

Cooperation, Teams

with Creative Control

Stimulated Planning, Strategic Planning

with Creative Control and Meta Games

No Direct Player Influence

with Mutual Goals or Continuous Goals and Supporting Goals

Companions

with Preventing Goals

Conflicts

with Preventing Goals and Creative Control

Indirect Conflicts

Can Modulate

Avatars, Characters, Game Worlds, Narration Structure, NPCs, Stimulated Planning, Tactical Planning

Can Be Instantiated By

Action Programming, Rules

Can Be Modulated By

Actions Have Diegetically Social Consequences, Ambiguous Responses, Awareness of Surroundings, Contextualized Conversational Responses, Difficulty Levels, Enforced Agent Behavior, Goal-Driven Personal Development, Handicap Systems, Initiative, Open Destiny, Own Agenda, Randomness, Sense of Self, Thematic Consistency, Unpredictable Behavior

Possible Closure Effects

Stopping to consider Algorithmic Agents as Agents may cause the removal of Emotional Engrossment

Potentially Conflicting With

Empowerment, Exaggerated Perception of Influence, Ubiquitous Gameplay

with Multiplayer Games

Social Interaction

with Game Mastery or without Diegetic Consistency

Emotional Engrossment

without Open Destiny

Replayability

History

New pattern created in this wiki.

References

1. ↑ Wikipedia entry for MUDs.
2. ↑ Wikipedia entry to LPMUDs.
3. ↑ Lankoski, P. & Björk, S. (2007). Gameplay Design Patterns for Believable Non-Player Characters. DiGRA 2007 Conference.
4. ↑ Wikipedia entry for the ELIZA program.
5. ↑ Eladhari, M. & Lindley, C. A. Player Character Design Facilitating Emotional Depth in MMORPGs. Level Up Conference Proceedings. Utrecht: University of Utrecht, November, 2003.
6. ↑ Wikipedia entry for Aimbots.

Acknowledgments

Karl-Petter Åkesson