Agents and Environments

• An agent generally interacts with its environment through sensors for perception and effectors or actuators for actions.

Agent Examples

• Human agents use eyes, ears, skin, and taste buds for sensors, and hands, fingers, legs, and the mouth for actuators, which are powered by muscles.
• Robots use cameras, infrared, and bumpers for sensors, and grippers, wheels, lights, and speakers for actuators, often powered by motors.
• Software agents function as sensors using information provided as input in the form of encoded bit strings or symbols and function as actuators using results to deliver the output.

Agents and Environments

• An agent perceives its environment through sensors and can change it through actuators.
• A complete set of inputs at a given time is called a percept.
• The current percept or a sequence of percepts can influence an agent's actions.
• An operation involving an actuator is called an action, and actions can be grouped into action sequences.

Agents and Their Actions

• A rational agent does "the right thing," which leads to the best outcome.
• An agent function maps percept sequences to actions, described abstractly.
• An agent program is a concrete implementation of the agent, running on a specific agent architecture or platform.
• Problems may arise with AI agents such as, what ‘the right thing’ is and how to measure the best outcome

Performance of Agents

• Performance is measured by the outcome and expenses of the agent.
• Performance criteria should be objective and are task-dependent, time is also important.

Performance Evaluation Examples

• For a vacuum agent, performance can be measured by the number of tiles cleaned.
• The number of tiles cleaned can be determined by the agent's report or validated by an objective authority.
• Performance evaluations often do not consider the expenses of the agent or side effects like energy, noise, loss of useful objects, damaged furniture, or scratched floors.
• A vacuum agent might re-clean clean tiles, cover only part of the room, or drop dirt to have more tiles to clean, which leads to unwanted activities.

Rational Agent

• These selects the action that are expected to maximize its performance based on a performance measure, taking into account percept sequence, background knowledge, and feasible actions.

Rational Agent Considerations

• Considers the performance measure for successful completion of a task, complete perceptual history (percept sequence), background knowledge (especially about the environment, including dimensions, structure, and basic "laws"), and feasible actions, like the capabilities of the agent.

Omniscience

• A rational agent is not omniscient and doesn't know the actual outcome of its actions or certain aspects of its environment.

Limited Rationality

• Takes into account the limitations of the agent, including percept sequence, background knowledge, and feasible actions.
• It deals with the expected outcome of actions.

PEAS Description

• PEAS stands for Performance, Environment, Actuator and Sensors
• PEAS is used to specify the composition of designed AI agents.

PEAS Metrics: Autonomous Taxi Driver

• Performance: A safe, fast, legal, comfortable trip to maximize profits.
• Environment: Roads, other traffic, pedestrians, customers.
• Actuator: Steering wheel, accelerator, brake, signal, horn.
• Sensors: Cameras, sonar, speedometer, GPS, odometer, engine sensors, keyboard.

Environment Types

• Fully Observable vs. Partially Observable: A fully observable environment's sensors give complete access to its state at any point in time, and partially observable is the reverse.
• Deterministic vs. Stochastic: A deterministic environment’s next state is completely determined by the current state and the action executed by the agent, whereas a stochastic environment is the opposite.
• Episodic vs. Sequential: In an episodic environment, the agent's experience is divided into atomic "episodes," where the choice of action depends only on the episode itself; sequential is the opposite.

Enviroment Classifications

• Static vs. Dynamic: A static environment is unchanged while the agent is deliberating; the opposite is dynamic.
• Discrete vs. Continuous: A discrete environment has a limited number of distinct, clearly defined percepts and actions, the opposite is continuous.
• Single Agent vs. Multiagent: A single agent environment has the agent operating by itself, the opposite has multiple agents in the environment.
• Known vs. Unknown: This distinction refers not to the environment itself but to the agent's or designer's state of knowledge about the "laws of physics" of the environment.

Environment Type Examples

• Chess with a Clock: Fully observable, deterministic, not episodic, semi-static, discrete, single agent.
• Chess without a Clock: Fully observable, deterministic, not episodic, static, discrete, single agent.
• Taxi Driving: Not fully observable, not deterministic, not episodic, not static, not discrete, single agent.
• The environment type largely determines the agent design.
• The real world is (of course) partially observable, stochastic, sequential, dynamic, continuous, and multi-agent.

Agent Program Types

• Agent Programs achieve mapping from percepts to actions with different levels of complexity.
• These programs can be:
  • Simple reflex agents.
  • Model-based agents, which keep track of the world.
  • Goal-based agents, which work towards a goal.
  • Utility-based agents.
  • Learning agents.