Intelligent Agents: Rationality and PEAS

• Intelligent agents interact with their environment using sensors to perceive and actuators to act.
• The agent function mathematically describes an agent's behavior, mapping percept sequences to actions.
• The agent program is the physical implementation of the agent function running on an architecture.

Rationality

• Rationality is determined by performance measures defining success criteria given the agent's knowledge, actions, and percept history.
• A rational agent chooses actions maximizing expected performance based on perceived sequences.
• Rationality differs from omniscience; agents may lack complete information and make actions with uncertain outcomes.
• Rationality involves exploration, learning and acting autonomously.

PEAS

• Designing a rational agent requires specifying the PEAS: Performance measure, Environment, Actuators, and Sensors.
• Tasking an automated taxi includes a PEAS of:
  • Performance: safety, destination, profits, legality, and comfort.
  • Environment: US streets, traffic, pedestrians, and weather conditions.
  • Actuators: steering wheels, accelerators, brakes, horns, and speaker.
  • Sensors: video cameras, accelerometers, gauges, engine sensors, keyboard, and GPS.
• Designing an Internet shopping agent includes:
  • Performance: price, appropriateness, efficiency, and quality.
  • Environment: current/future websites, vendors, and shippers.
  • Actuators: display to user, URL following and form completion.
  • Sensors: HTML text, graphics and scripts.

Environment Types

• Fully observable environment: the agent can access the complete state of the environment.
• Deterministic environment: the next state is completely determined by the current state and the agent's action.
• Episodic environment: the agent's experience is divided into atomic episodes; the choice of action in each episode depends only on the episode itself.
• Static environment: the environment does not change while the agent is deliberating.
• Discrete environment: a limited number of distinct, clearly defined percepts and actions exist.
• Single-agent environment: only one agent operates in the environment.
• The real world is considered a partially observable, stochastic, sequential, dynamic, continuous, multi-agent environment.
• Environment properties influence the agent design.

Agent Types in Order of Increasing Generality:

• Simple reflex agents:
  • Act based solely on the current percept, using condition-action rules.
  • Are limited in partially observable environments.
  • May loop if a location sensor is missing.
• Reflex agents with state
  • Maintain a state to compensate for partial observability.
  • Incorporate information about the past to choose current actions.
• Goal-based agents
  • Use goals to guide actions, considering future states resulting from actions.
• Utility-based agents
  • Optimize selecting actions by considering preferences to which internal state to be in.
  • Represent performance metrics and trade offs which is distinct from goal-based agents.
• All of these agent types can be turned into learning agents.

Explore intelligent agents, their interactions, and rationality. Learn about performance measures, agent functions, and the PEAS framework (Performance, Environment, Actuators, Sensors) for designing rational agents. Understand how agents perceive, act, and make decisions in their environment.