Artificial Intelligence - Intelligent Agents Chapter 2

Artificial Intelligence - CSB2104

• Course taught by Prof. Abdel-Rahman Hedar

Intelligent Agents (Chapter 2)

• Intelligent Agents are anything that acts on behalf of a user, using sensors to perceive the environment and effectors to act on it.
• A key component is their intelligence, ranging from simple rules to complex learning algorithms.
• Agents adapt to environmental changes, not only reacting but often proactively.
• They can cooperate with other agents and even migrate to different systems.
• They can also communicate with the user, system, and other agents.

What is an (Intelligent) Agent?

• The term "Intelligent Agent" is sometimes overused.
• An agent perceives its environment through sensors.
• It acts on the environment using effectors.
• Its goal is to maximize progress.
• The concept of an agent as a tool to analyze systems.
• Agents are not a definitive division of the world; rather, they're like object-oriented vs. imperative approaches
• The notion of an Agent encompasses Percepts, Actions, Goals and the Environment.
• Task-specific and specialized agents possess well-defined goals and structured environments.

Example: Windshield Wiper Agent

• Goals: Keep windshields clean and maintain visibility.
• Percepts: Raining, dirty.
• Sensors: Camera (moist sensor).
• Effectors: Wipers (left, right, back).
• Actions: Off, Slow, Medium, Fast.
• Environment: US inner cities, freeways, highways, and weather conditions.

Example: Autonomous Vehicles

• Collision Avoidance Agent (CAA): Avoids obstacles, perceives distance, velocity and trajectory.

• Sensors: Vision, proximity sensing.

• Effectors: Steering wheel, accelerator, brakes, horn, headlights.

• Actions: Steer, speed up, brake, signal (headlights).

• Environment: Freeways.

• Lane Keeping Agent (LKA): Stays in lane, perceives lane center and boundaries.

• Sensors: Vision.

• Effectors: Steering wheel, accelerator, brakes.

• Actions: Steer, speed up, brake.

• Environment: Freeways.

Agent PEA Description

• This section details the performance measures, environments, actuators, and sensors for various agent types.
• Examples include Medical diagnosis systems, satellite image analysis systems, part-picking robots, refinery controllers, and interactive English tutors.

Conflict Resolution by Action Selection

• Agents can override or arbitrate actions.
• Agents can compromise actions to satisfy multiple goals.
• Challenges include deciding which action is best for the specific situation.

Behavior and Performance of IAs

• Perception, followed by action mapping, is fundamental.
• Ideal mapping: Specific actions for a given state.
• Performance measure: Subjective assessment (e.g., speed, power use, error rate, cost).
• Degree of autonomy: Agent's ability to make decisions independently.

The Right Thing = The Rational Action

• Rational action is the best possible action given the agent's knowledge and abilities.
• Rational involves recognizing constraints and limitations.
• Rationality is not omniscience, clairvoyance, or guaranteed success.

How is an Agent Different from Other Software?

• Agents are autonomous.
• Agents have some intelligence (from simple rules to learning algorithms), allowing adaptation.
• They are not limited to reactive actions, but can also act proactively.
• Agents have some social ability, interacting with users, systems, or other agents.

Environment Types

• Categorizes environments according to characteristics.
• Deterministic vs. non-deterministic: Predictability of future states.
• Static vs. dynamic: Stability of the environment over time.
• Episodic vs. non-episodic: Sequence of events.
• Discrete vs. continuous: Nature of the environmental states.
• Single vs. multi-agent: Number of agents involved.

IA Structure

• An agent is composed of an architecture and a program.
• The agent program defines the perception-action mapping.
• The architecture is the device that executes the agent program (e.g., a computer).

Using a Look-up Table

• Example of how to use a look-up table for collision avoidance in various scenarios.
• The size of the table is determined by the number of possible percepts and actions.

Table-Driven Agent

• A table-driven agent is useful for agents with limited complexity.
• For a given sequence of percepts, use a pre-defined table to determine the correct actions.

Conclusion

• Intelligent agents perceive and act on their environments to achieve goals.
• Rational actions maximize the expected value of performance measures, given current data.