IAI | HSLU | Jana Köhler

Limited rationality is the idea of acting appropriately in a given situation under limited resources, recognizing that perfect rationality may not be attainable. Limited rationality acknowledges that rational agents often have constraints and incomplete information, and they make the best decisions within those limitations.

• Perceive the environment through sensors (Percepts) 
• Act upon the environment through actuators (Actions)

Complete history of what the agent has perceived to date

A function that maps any given percept sequence to a single action

• Mathematically abstract: any percept sequence must be mapped to an action, but we cannot store all percept sequences in limited memory
• allowing it to determine the appropriate action based on its past perceptions.

• – Specific implementation of the agent function
• – Takes only the current percept as input and returns an action to the actuators
•  – It may consider earlier percepts or actions depending on the agent function

Each action takes the world to another state.

The agent is considered to have performed well if the sequence of world states is desirable for an external observer.

The performance measure evaluates the STATE of the ENVIRONMENT independent of the AGENT. This independence is essential to prevent the agent from deluding itself into believing its performance is perfect.

-> You get what you reward.

A rational agent should select an action for each possible percept sequence that is expected to maximize its performance measure, based on the evidence provided by the percept sequence and the agent's built-in knowledge.

• The performance measure (goal) that the agent seeks to optimize.
• The percept sequence, which provides evidence about the agent's environment.
• The agent's knowledge of the environment, which includes its understanding of the world.
• The set of available actions that the agent can take.

A utility function is used by a rational agent to evaluate the desirability of a state of the world. It quantifies how desirable a particular state or outcome is in terms of the agent's goals.

A utility function maps a state (or a sequence of states) to an evaluation value, typically a real number. 

– Agent explores the effect of its planned action and determines the possible state of the world that results when this action is executed by the agent

1. – to select an action (or a sequence of actions) 

2. – to weigh the importance of competing goals

Answer 1: Rational agents are characterized by making decisions that maximize expected outcomes based on their goals and knowledge.

The three essential elements are the agent's percept sequence, agent function, and agent program, which together represent how an agent perceives and acts within its environment.

 The fundamental components of a basic agent architecture are the percept sequence, agent function, and agent program. These elements define how an agent perceives its environment and makes decisions and actions.

A creature without sensors may struggle to interact effectively with its environment since it lacks the ability to perceive and gather information. Sensory input is crucial for informed decision-making and actions.

The rational behavior of an agent depends on the agent's performance measure (goal), the percept sequence, knowledge of the environment, and the available actions. Rational agents make decisions that maximize expected outcomes based on these factors.

The performance measure is a metric used to assess how well an agent is performing in achieving its goals or objectives. It serves as a criterion for evaluating the quality of an agent's decisions and actions.

The utility function is used by an agent to evaluate the desirability of different states or outcomes. It assists the agent in making decisions that maximize its expected utility based on its goals and preferences.

1. Performance measure 
2. Enviroment
3. Actuators 
4. Sensors 

Properties of environments in agent-based systems include:

1. Completeness of knowledge (Known or Unknown)
2. Observability (Observable or Partially observable or unobservable)
3. Environment dynamics (Dynamic or Static)
4. Type of model (Discrete or Continuous)

Properties of agent actions in agent-based systems include:

1. Predictable action effects (Deterministic or Stochastic)
2. Dependency of action effects (Episodic or Sequential)
3. Number of agents (Single-Agent or Multi-Agent)

These properties help define the nature of the actions an agent can take and the characteristics of the environment in which the agent operates.

Properties that distinguish simple and difficult environments in AI applications include:

• Knowledge (Known or Unknown)
• Observability (Observable or Unobservable)
• Dynamics of Changes (Static or Dynamic)
• Detail of Models (Discrete or Continuous)
• Short-term Action Effects (Deterministic or Stochastic)
• Long-term Action Effects (Episodic or Sequential)
• Number of Agents (Single or Multi)

Understanding how these properties vary in environments is key to designing successful AI applications, as it influences the complexity of decision-making and problem-solving for agents.

 A PEAS description defines the Performance measure, Environment, Actuators, and Sensors, providing a structured representation of how an agent interacts with and perceives its environment.

The performance measure is included in a PEAS description because it assesses how well an agent is achieving its goals, providing a more direct measure of an agent's performance. The utility function, on the other hand, is used internally by the agent to make decisions based on its preferences.

Properties used to characterize agent environments include knowledge, observability, dynamics of changes, detail of models, short-term action effects, long-term action effects, and the number of agents. These properties influence the complexity and nature of agent behavior within those environments.

Yes, the properties of agent actions can change in different environments, impacting how agents make decisions and interact with their surroundings. For example, the predictability and impact of actions may vary based on environmental characteristics.

An agent with only episodic memory may struggle in a sequential environment. Episodic memory is typically short-term, while sequential environments require long-term memory to make decisions based on past actions with long-term effects. Success in such environments often depends on the agent's ability to remember and consider past actions and their consequences.

Agents differ in their capabilities:

• – Exploration: execute explorative actions for information gathering 
• – Learning: derive additional insights from percepts
• – Autonomy: improve partial or incorrect knowledge

5 Types of Agents
– Simple Reflex Agent
– Model-based Reflex Agent – Goal-based Agent
– Utility-based Agent
– Learning Agent

A Simple Reflex Agent senses the world and responds immediately by applying simple rules that interpret sensor input and link it to specific actions.

– No explicit world model, no "memory"