Intelligent Agents and Their Environments

Intelligent Agents

• Intelligent agents are entities that perceive and act within an environment.
• Agents can encompass animals, humans, robots, software agents (softbots), internet agents (crawlers, webbots, email agents, search agents, chatbots), and more. Examples include Cortana, Siri, Google Assistant, Watson, Alexa, and others.
• Agents can be single or multi-agent (distributed AI).
• Agents operate through sensors that perceive aspects of the environment, and actuators that affect the environment.
• Perception involves interpreting sensor responses to gain information about the world where the agent exists.
• Actuators include hands, legs, vocal tracts, and mechanized equivalents such as accelerator, steering, and braking mechanisms.

Agents and Environments

• Agents perceive their surroundings through sensors.
• Sensors provide input in various forms – vision, hearing, touch, radio, infrared, GPS, wireless signals, and active sensing (sending out signals that reflect and are perceived to determine an object's characteristics).
• Agents act on their environment utilizing actuators.
• An example is a robot vacuum cleaner, where percepts relate to the position and state (e.g., clean or dirty) of each portion of the floor. Actions may be "Left", "Right", "Suck" or "NoOp."

Sensors and Actuators

• Sensors measure environmental features.
• The IoT (Internet of Things) encompasses environmental data sources.
• Perception is the processing of sensor data to extract information.
• Actuators trigger actions within the environment, and are exemplified by hands, legs and vocal mechanisms, and their equivalents in automated systems (e.g., accelerators, steering, braking).

Example: Vacuum-cleaner world

• A robot vacuum, e.g., iRobot Roomba, is an agent that navigates a room and maintains it clean.
• The room's features are perceived by the robot.
• Actions include moving, and cleaning (suctioning).

Agent Functions

• Agent functions specify the actions an agent takes based on its perceptions.
• An agent's behavior is entirely defined by its function, which maps past perceptions to actions.
• For any given environment, ideal agents provide optimal performance.
• Implementing the ideal function concisely remains a challenge. Rational agents act to maximize the expected value of the performance measure given the perceived sequence up to that time.

Example: A vacuum-cleaner agent

• An agent analyzes perception to choose an action based on a table of states and actions.
• Table-driven agent receives a percept, adds it to a sequence, and looks up the resulting action in a table.

Agent Programs

• Agent programs execute on a physical architecture.
• An agent's program and architecture constitute the agent function.
• The program takes input from sensors, keeps an internal state (often in memory), and produces output for actuators.
• A skeleton agent is a basic agent function template, which includes steps for updating memory and choosing the best action according to memory.

Other forms of algorithms

• Pseudocode is a simplified, informal way to represent algorithms.
• It uses programming-like syntax and natural language.
• It helps in designing algorithms before implementing them in a specific programming language.

Algorithm

• An algorithm is a detailed set of instructions for solving a problem.
• Algorithms can be applied to many types of problems.
• Time and space complexity analyze how algorithms perform.

The pseudocode

• The pseudocode is a concise way to represent algorithms.
• Variables have italicized lowercase names.
• Functions are named using capitalized letters.
• Pseudocode may use mathematical formulas or English to describe elements.
• Conditional statements (e.g., "if...then...else") and loops (e.g., "for each" and "while") have significant indentation.
• It supports assignment (e.g., "x ← value"), de-structuring assignment (e.g., "x, y ← pair"), loops (e.g., "for each x in c do"), and generators that produce a sequence of values.

The code

• Algorithms in pseudocode can be implemented in a variety of programming languages, such as Python, Java, C/C++, Lisp, and Prolog.
• These implementations will typically involve specific code for agents, algorithms, environments, and domains.

A Python coding

• Python is a common language for implementing agent programs.
• The code structure usually consists of separate directories for agents, algorithms, environments, domains.

Example: A vacuum-cleaner agent (Python)

• A table-driven approach can implement the agent, using a dictionary style table to map percept sequences to actions.

Agent Environment in AI

• An environment is everything surrounding an agent, but excluding the agent itself.
• Environments provide agents stimuli for sensing and reacting.
• Environments can be characterized as static or dynamic, and observable or non-observable.

Features of Environment

• Environments can have diverse characteristics, from fully observable to partially observable, static to dynamic, discrete to continuous, deterministic to stochastic, single agent or multi-agent, episodic or sequential, known and or unknown, and accessible or inaccessible.

Rationality

• Rational agents aim for optimal performance.
• Goals dictate desirable outcomes, which are measurable in terms of a numerical performance measure, and adaptable to changing circumstances through exploration, learning, and autonomy.

Example: Vacuum-cleaner rational agent.

• Performance measures for vacuum robots could include points for cleaned squares, penalties per move, or for remaining dirty squares.

PEAS (Performance/Environment/Actuators/Sensors)

• To design a rational agent, PEAS specifies its performance measure, environment, actuators, and sensors (e.g., for an automated taxi: performance metrics include legality, safety, profitability, with the environment consisting of roads, vehicles and pedestrians, having actuators like steering. brake, accelerator, and sensors like GPS, cameras, etc.).

Example: Automated taxi agent

• Performance measures include safety, profit, schedule adherence.
• Environment factors include road conditions, traffic flow, pedestrians, and weather.
• Actuators include accelerator, braking mechanisms, steering, and communication devices.
• Sensors such as cameras, GPS trackers and other traffic and sensor systems will furnish the necessary information and data.

Example: Internet shopping agent

• Performance measures include efficiency, price, appropriateness.
• The environment comprises websites, vendors, and delivery services.
• Actuators include user interaction mechanisms and data processing.
• Sensors capture text, graphics, data and other data from websites to complete transactions.

Environments

• Features of different environments (e.g., Solitaire, Backgammon, Internet shopping, and Taxi) are detailed.

Agent Structures

• Different agent structures exist (simple reflex, model-based reflex, goal-based, and utility-based).
• These can be augmented by learning to become more effective.

Multi-agent systems (MAS)

• Multi-agent systems (MAS) are composed of multiple agents interacting amongst neighbours.
• Tasks are often divided into smaller tasks and assigned to specific agents.

Properties of Multi-Agent Systems (MAS)

• MAS properties include coordination, communication (agents interacting), fault detection and isolation (FDI), task allocation (distributing tasks), and localization (restricting an agent's view).

Agent Organization

• Methods for agent communication and structure within a MAS include flat (all equal), hierarchical, holons (grouped by characteristics), coalition (temporarily grouped by goals), team, matrix, and congregation (groups operating locally).

The agent projects

• Projects involve designing and implementing agents utilizing new knowledge (e.g., internet agents, intelligent robots).
• Programming languages and environments (such as those with which the students are already familiar) will be appropriate.
• Projects may involve single or multiple agents.

Explore the fascinating world of intelligent agents, spanning from humans and animals to software bots like Siri and Alexa. Discover how these agents perceive their surroundings using sensors and interact with their environments through actuators. This quiz covers the basics of intelligent agents, their types, and the mechanisms that allow them to operate effectively.