2.1.2. Types of Agents

● Simple Reflex Agents ○ Act only on the current percept. ○ Use condition-action rules (if-then statements). ○ Example: A thermostat that turns on the heater if the temperature is below a certain threshold.

Simple Reflex Agents are the most basic type of agents. They operate based solely on their current perception without considering the past. They follow straightforward rules: if a certain condition is detected (the perception), then they perform a specific action. For instance, a thermostat is a practical example of this agent type; it recognizes when the temperature drops below a set point and activates the heater without any memory of past temperatures.

Imagine a light switch that turns on when it detects motion in the room. Just like the simple reflex agent, the light switch doesn’t remember that someone was in the room before; it simply reacts to the current situation.

● Model-Based Reflex Agents ○ Maintain some internal state to handle partially observable environments. ○ Use models of how the world works. ○ Example: A robot vacuum cleaner that remembers areas it has already cleaned.

Model-Based Reflex Agents build on the foundation of Simple Reflex Agents by incorporating some internal state. This internal state helps them understand the environment better, especially when it is not fully observable. For instance, a robot vacuum cleaner uses its internal memory to keep track of areas it has already cleaned, allowing it to plan its next moves effectively rather than repeatedly cleaning the same spot.

Consider a person navigating through a mall. If they remember which stores they’ve already visited, they can plan their route better instead of wandering aimlessly. Similarly, a robot vacuum uses its internal memory to avoid cleaning the same area twice.

● Goal-Based Agents ○ Act to achieve specified goals. ○ Perform search and planning. ○ Example: A chess-playing AI trying to checkmate its opponent.

Goal-Based Agents are designed to achieve specific objectives. Unlike their predecessors, they do not just react to stimuli; they think strategically about their actions to reach a particular goal. This type of agent often performs searches and plans ahead. For example, a chess-playing AI evaluates various moves to find the best sequence that ultimately leads to checkmate.

Think of a student preparing for an exam. They set a goal of achieving a certain score. To reach this goal, they plan their study schedule, review the relevant materials, and practice with past tests, much like a goal-based agent plans and strategizes its moves in chess.

● Utility-Based Agents ○ Aim to maximize a given utility function (a measure of "happiness" or performance). ○ Handle trade-offs between competing goals. ○ Example: A self-driving car optimizing for speed, safety, and fuel efficiency.

Utility-Based Agents are designed to maximize their overall effectiveness or ‘utility’. They can evaluate different options and make decisions that balance competing goals. For instance, a self-driving car must optimize for speed, safety, and fuel efficiency all at once. This means when making decisions, it assesses each option to find the best overall outcome, not just focusing on one aspect.

Imagine a person choosing a restaurant. They want to enjoy good food (utility), dine at a reasonable price (trade-off), and experience a pleasant atmosphere (complementing goal). They weigh these factors to select a place that provides the best overall experience.

● Learning Agents ○ Improve performance through experience. ○ Have components for learning and performance. ○ Example: Recommendation systems that adapt based on user behavior.

Learning Agents are capable of adapting and improving their performance by learning from their experiences. They possess components designed for learning, which allows them to adjust their actions based on feedback and past interactions. A common example is a recommendation system, like those used by streaming services, which analyzes user behavior to provide personalized suggestions over time.

Think of a teacher who learns from student feedback and exam results. Each year, they modify their teaching techniques and materials based on what worked and what didn’t, just as a learning agent adjusts its performance to be more effective in the future.

--