As an aside, it is interesting to view the AI from The Sims as an implementation of J. J. Gibson's ecological approach to vision. In this approach, Gibson states that animals view the world in terms of affordances rather than the traditionally assumed things like shapes, colours and textures. An affordance represents what the environment can offer the organism doing the viewing (Gibson, 1979). For example, when a human views a chair, it doesn't see a collection of cleverly arranged surfaces, it sees something that it can sit on and which will support its weight. When a fox views a rabbit it sees something that it can eat, rather than a small, fast moving, furry object.

The "smart terrain" (Cass, 2002) in The Sims is engineered in a very similar way to this notion of affordance-based vision. When agents (characters) in The Sims view an object, they don't see a pizza or a pool table, they see a set of value modifiers and the instructions to access these modifiers. For example, a pizza may offer to decrease an agent's hunger by a number of points and its instructions may include an animation for eating the pizza (McLean-Foreman, 2001). This is a fantastic approach to game AI for two reasons. First, because most of the behaviour instructions are contained (passively) within environmental objects, agent AI can (as fondue pointed out above) be very shallow (just determining which values it wants to satisfy, and choosing the best objects to do this) and therefore computationally cheap (so the AI doesn't mess with the precious framerate). Second, the original game is inherently extensible. When the developer wants to add new objects to the world of The Sims, all it has to do is create these objects insuring that they use the right set of instructions (modify existing values and provide a suitable animation) and they can drop them straight into the world. This ease is probably the reason for the cynically large number of expansion packs currently available for The Sims.

It is not known whether the developers at Maxis were aware of Gibson's work during the development of The Sims, but it is interesting to see the similarities between the two ideas. The AI in The Sims simplifies the issue a great deal (well it is a game after all), but it is clear to see that reasoning about acting in an environment is a lot easier when the low-level details (how to connect shapes into objects etc.) are abstracted away. In The Sims this abstraction is done by the object designer. Who knows what does it in humans and other animals?

Gibson, J. J. (1979). The Ecological Approach to Visual Perception. Houghton Mifflin.
Cass, S. (2002). Mind Games.
McLean-Foreman, (2001). An Interview with Will Wright.