Intelligent Agents

For each possible percept history, select an action that is expected to maximize its performance measure, given the evidence by the percept history and whatever built-in knowledge the agent has.

Performance Measure, Environment, Actuators, Sensors

sensors detect all relevant properties of the world for the current action

only one agent, no cooperation and no competition between agents

next state determined by the current state and the performed action

agent’s experience is divided into “atomic” parts (independent from each other, which means current actions don’t affect future action)

In episodic environments, the choice of action only depends on the current episode

the world does not change during the reasoning time of the agent

static, but the performance score decreases with deliberation time

world properties have discrete values, e.g. time, number of possible states,...

(Doesn’t go on infinitely)

state of knowledge about the “laws of physics” of the environment

1. simple reflex agents

2. reflex agents with state (model-based reflex agent)

3. goal-based agents

4. utility-based agents

• no memory

• No sequences of percepts

• Looping possible

• memory -> maintain/ update world state

• reason about unobservable parts

• deal with uncertainty, implicit goals

• model the world, goals, and actions & their effects explicitly

• more flexible, better maintainable

• Search & Planning for single/sequence of action/s to achieve a goal

• assess goals with a utility function

• resolve conflicting goals

• use expected utility for decision