Game AI Eras
While the progression of game artificial intelligence is large a continuum, we have proposed that there are "eras" that can be identified much the way society breaks up music, film, etc. into eras (e.g. "80s music"). Note that this is different than breaking things up by genre (e.g. music's "classic rock"). The latter would be more a function of the types of underlying architecture. Instead, we are identifying longer-term trends in the way developers approached AI as a whole.
At present, we have identified the following as eras in game AI development.
Approximate Time Frame: early arcade, early home computer, 70's - mid-80's
The enemies are largely using specifically stored patterns to move the agents. Often, these patterns are repetitive, simple, and short-lived. Additionally, the patterns have little, if anything, to do with the actions of the player. Examples of this would be early arcade games such as Space Invaders and Donkey Kong.
Simple Hard-coded Rules
Approximate Time Frame: early arcade, early home computer, 70's - late-80's
The agents use hard-coded algorithmic rules to determine movement or actions. These may take into account the actions of the player, but usually only on a limited basis such as "move towards" or "move away". An example of this would be Pac Man where the AI ghosts each had their own ruleset for navigation through the maze.
Approximate Time Frame: early 80's - late 80's
The agents use hard-coded rules that are more complex that the basic ones above but are still largely oblivious to the presence or actions of the enemies. Examples include enemies in Zelda or Super Mario Brothers.
Approximate Time Frame: mid 80's - present
The agents include short-term, local reactions to local stimulus. Many fighting games such as Street Fighter, for example, would exhibit reactions such as blocking the players attacks and striking during openings in the players defenses. On a basic level, many of these simple behaviors continue to be used to this day.
Approximate Time Frame: mid 90's - present
The agents make move advanced decisions about their environment. This differs from Tactical Reaction in that the actions are generally longer-term (> 1 second) and may make use of multiple steps. Examples of this sort of behavior include intelligent pathfinding, use of cover, etc. Tactical Reasoning Decisions are still restricted to individual agents, however.
Approximate Time Frame: late 90's - present
As agents began to be more aware of each other, the advent of tactical group and strategic actions became more prevalent. Many of these games employed the use of "overseer" AI to coordinate multiple agents or affect local behavior patterns. Examples of games that leveraged this type of AI include the Halo series, Starcraft 2, and Left 4 Dead.