6.6 Social Impact

The spectacular growth of the video game sector has major economic, technological, social, and psychological impacts. The commercial video game industry generates more revenue than the movie industry and the music industry combined. In 2022, the revenue from video games was estimated to be about $200 billion. In commercial video games, the non-player characters (NPCs) have to generate and execute a wide variety of believable behaviors responding to other agents’ behaviors and changes in the environment. The simplest form of planning is path planning in a spatial environment. A for path planning was widely adopted in video games. Autonomous NPCs have used a variety of representations of the set of possible behaviors. Early video games typically used finite state machines with explicitly programmed belief-state transition functions and command functions. Behavior trees specify complex behaviors as tree structures of tasks with interior nodes represent the agent’s higher-level tasks and leaf nodes corresponding to primitive actions, implementing a form of hierarchical control. Behavior trees allow for more expressive representations but they require pre-specifying all possible behavior explicitly.

Planners are now increasingly being used in video games instead of finite state machines or behavior trees. The advantage of planners is that they allow the agent to look further ahead in time, and they also allow much richer behavioral repertoires. F.E.A.R., published in 2005, was one of the first games to use planning successfully. Its planner used a simplified STRIPS approach, known as goal-oriented action planning, GOAP. The successful use of GOAP in F.E.A.R. inspired the use of STRIPS-like planners in many other games, such as Tomb Raider (2013) and Middle-Earth: Shadow of Mordor (2014).

Another successful AI planning approach is the hierarchical task network (HTN). The HTN model is a generalization of the STRIPS approach. An HTN provides a set of tasks, made up of primitive tasks (similar to the actions of STRIPS), compound tasks composed of simpler tasks, and goal tasks (generalizations of STRIPS goals). Killzone 2, published in 2009, was the first commercial video game built using the HTN approach to planning. Other games built subsequently using HTN include Transformers: Fall of Cybertron (2012, 2016) and Dying Light (2015).

The planners described here assume deterministic actions and complete knowledge of the world. Relaxing those restrictions requires the approaches to planning and acting with uncertainty described in Chapters 12, 13, and 14. Many of those techniques are used in modern video games.