Task environment
The task environment is the specific problem or scenario within which an intelligent agent operates. Characterizing task environments along several dimensi...
Classifying the World an Agent Lives In
The task environment is the specific problem or scenario within which an intelligent agent operates. Characterizing task environments along several dimensions is essential because the right agent design depends heavily on the nature of the environment. Russell and Norvig identify seven key dimensions for this characterization.
The first dimension is observability. A fully observable environment gives the agent complete access to the current state. A partially observable environment provides only limited sensor data, forcing the agent to maintain beliefs about unseen aspects.
The second dimension is the number of agents. Single-agent tasks involve only the designing agent. Multi-agent tasks involve other agents whose actions affect the environment. Chess is a two-agent competitive environment. Traffic involves many cooperating and competing agents simultaneously.
The third dimension is determinism. A deterministic environment means that the next state follows completely from the current state and the agent's action. If chance or other agents can unpredictably alter outcomes, the environment is stochastic.
Fourth is sequentiality. In episodic environments each decision is independent. In sequential environments, today's action affects tomorrow's situation — requiring long-term planning.
Fifth is dynamism. A static environment stays the same while the agent deliberates. A dynamic environment changes in real time, penalizing slow decisions.
A concrete example: a crossword puzzle is a fully observable, deterministic, sequential, static, single-agent, discrete task environment. By contrast, real-world stock trading is partially observable, stochastic, sequential, dynamic, multi-agent, and continuous.
The seventh dimension is discreteness versus continuity, which refers to whether states, time, and actions are countable or real-valued.
Exam tip: For any given scenario, practice classifying it along all seven dimensions. Examination questions frequently describe an environment and ask you to name its properties.