Artificial Intelligence Fundamentals: Concepts, Agents, and Algorithms

Posted on Jun 25, 2025 in Other university degrees and diplomas

What is Artificial Intelligence?

Artificial Intelligence (AI) is the field of computer science dedicated to creating intelligent agents, which are systems that can reason, learn, and act autonomously. Essentially, it is about making computers think and behave like humans.

Key Features of Artificial Intelligence

• Learning: The ability to acquire information and rules for using the information. This includes memorizing specific instances and discovering new rules through experience.
• Reasoning: The capacity to draw inferences, solve problems, and make decisions based on the available information.
• Problem Solving: The capability to formulate problems, find solutions, and achieve desired goals.
• Perception: The ability to sense and interpret the environment through sensors (like cameras or microphones), similar to human senses.
• Language Understanding: The skill to process, interpret, and generate human language, enabling natural communication.
• Planning: The capacity to set goals and develop sequences of actions to achieve those goals.
• Adaptability: The ability to adjust strategies and behaviors in response to changing environments and new information.

Fundamental Concepts of Artificial Intelligence

Several fundamental concepts underpin the field of AI:

• Agents: As mentioned, these are entities that perceive their environment and act upon it.
• Environment: The surroundings in which an agent operates.
• Rationality: The quality of an agent’s behavior – an agent is rational if it does the “right thing” based on its knowledge and goals.
• Learning: Mechanisms that allow agents to improve their performance over time.
• Search: Techniques for exploring possible solutions to a problem.
• Knowledge Representation: Methods for encoding information in a way that an AI system can understand and use.
• Inference: The process of deriving new conclusions from existing knowledge.
• Machine Learning: A subfield of AI that focuses on enabling systems to learn from data without being explicitly programmed.
• Deep Learning: A subfield of machine learning that uses artificial neural networks with multiple layers to extract complex patterns from data.

Applications of Artificial Intelligence

AI has permeated numerous aspects of our lives. Here are some prominent applications:

• Healthcare: Diagnosis, drug discovery, personalized medicine, robotic surgery.
• Finance: Fraud detection, algorithmic trading, risk assessment, personalized financial advice.
• Education: Intelligent tutoring systems, personalized learning platforms, automated grading.
• Transportation: Autonomous vehicles, traffic management, route optimization.
• Entertainment: Recommendation systems (Netflix, Spotify), game AI, content generation.
• Manufacturing: Robotics, quality control, predictive maintenance, supply chain optimization.
• Customer Service: Chatbots, virtual assistants, automated support.
• Security: Surveillance systems, threat detection, biometric identification.
• Smart Homes: Voice assistants, automated lighting and temperature control, smart appliances.
• Agriculture: Precision farming, crop monitoring, automated harvesting.

AI Agents: Definition and Types

In AI, an agent is any entity that can perceive its environment through sensors and act upon that environment through actuators. A human agent has eyes, ears, and other sensory organs as sensors, and hands, legs, mouth, and other body parts as actuators. A robotic agent might have cameras, infrared range finders, and touch sensors as sensors, and various motors and effectors as actuators. A software agent receives percepts (inputs) from its environment (which might be the internet, a user interface, etc.) and acts on that environment by displaying output on a screen, writing files, sending network packets, etc.

Types of AI Agents

There are four basic types of agents in AI, categorized based on their architecture and complexity:

• Simple Reflex Agents: These agents react directly to the current percept, ignoring the history of percepts. They have condition-action rules: “If condition, then action.” They are simple but have limited intelligence.
• Model-Based Reflex Agents: These agents maintain an internal state, which is a representation of the world based on the percept history. They use a “model” of the world to determine the best action, especially when the environment is partially observable.
• Goal-Based Agents: These agents have a goal in mind. Their actions are chosen to reach that goal. They consider not only the current state but also the future states that might result from their actions. Search and planning are key components of goal-based agents.
• Utility-Based Agents: These agents go beyond just having a goal; they also consider how “happy” they will be in the resulting state. Utility is a measure of success or happiness. These agents try to maximize their utility, which allows them to make more sophisticated decisions when multiple goals are achievable or when there are trade-offs.

Simple Reflex Agents in AI

A simple reflex agent operates based on a set of predefined rules that directly map percepts to actions. It does not have any memory of past states or future consequences. Its decision-making process is solely based on the current sensory input.

How Simple Reflex Agents Work

• It perceives the current state of the environment through its sensors.
• It uses a set of condition-action rules (also known as reflex rules).
• If a rule’s condition matches the current percept, the corresponding action is executed.

Example: A thermostat is a simple reflex agent. If the temperature is below a certain threshold (condition), it turns on the heater (action). It does not remember past temperatures or predict future ones.

Limitations of Simple Reflex Agents

• Only works well if the environment is fully observable and the correct action can be determined solely based on the current percept.
• Cannot handle partially observable environments where past information is needed.
• Lacks the ability to learn or adapt to new situations not covered by its rules.

Model-Based Reflex Agents

A model-based reflex agent overcomes some limitations of simple reflex agents by maintaining an internal “model” of the environment. This model represents the agent’s understanding of how the world works, how it evolves over time, and how its actions affect the environment.

How Model-Based Reflex Agents Work

• It perceives the current state through its sensors.
• It updates its internal model based on the percept and its knowledge of how the world changes.
• It uses rules that consider the current percept and the internal model to decide on an action.

Example: A vacuum cleaner robot that has a map of the house (its model). When it encounters dirt (percept), it uses its map to decide the most efficient way to clean it, potentially remembering areas it has already cleaned.

Advantages Over Simple Reflex Agents

• Can handle partially observable environments by keeping track of the unobserved aspects.
• Can make more informed decisions based on its understanding of the world.

Limitations of Model-Based Reflex Agents

• The accuracy of the agent’s performance depends on the accuracy of its model.
• Designing and updating the model can be challenging.

Goal-Based Agents in AI

Goal-based agents are more sophisticated than reflex agents. They not only consider the current state and a model of the world but also have explicit goals they are trying to achieve. Their actions are chosen to reach these goals.

How Goal-Based Agents Work

• It perceives the current state.
• It maintains a model of the world.
• It has a defined goal or set of goals.
• It uses search and planning algorithms to find a sequence of actions that will lead to the goal state.

Example: A navigation system in a car. Its goal is to get you to a specific destination. It uses a map (model) and considers the current location (percept) to plan a route (sequence of actions) to reach the destination.

Advantages Over Model-Based Reflex Agents

• Can make decisions based on future desired outcomes, not just immediate situations.
• More flexible and can handle complex tasks that require planning.

Limitations of Goal-Based Agents

• Finding a sequence of actions to achieve a goal can be computationally expensive, especially in large and complex environments.

Utility-Based Agents in AI

Utility-based agents are an extension of goal-based agents. While goal-based agents simply aim to reach a goal, utility-based agents try to maximize their own “happiness” or “utility.” Utility is a measure of the desirability of a state.

How Utility-Based Agents Work

• It perceives the current state.
• It maintains a model of the world.
• It has a utility function that assigns a numerical value (utility) to different states.
• It chooses the action that will lead to the state with the highest expected utility.

Example: A self-driving car might have a goal of reaching a destination, but a utility-based agent would also consider factors like travel time, safety, comfort, and fuel efficiency, choosing the route that maximizes overall utility.

Advantages Over Goal-Based Agents

• Can handle situations where there are multiple ways to achieve a goal, allowing it to choose the “best” way.
• Can make rational decisions even when there are conflicting goals or no clear path to a single goal.

Limitations of Utility-Based Agents

• Defining a precise and accurate utility function can be challenging.

AI Environments: Definition and Types

In AI, the environment is the surroundings in which an agent operates. It provides the agent with percepts and is acted upon by the agent’s actions. The nature of the environment significantly influences the design of the agent.

Types of AI Environments

Environments can be categorized based on several properties:

• Fully Observable vs. Partially Observable:
  • Fully Observable: The agent’s sensors give it access to the complete state of the environment at each point in time. The agent knows everything relevant to make an optimal decision.
  • Partially Observable: The agent can only perceive a part of the environment’s state. It needs to maintain its own internal state to keep track of the unobserved aspects.
• Deterministic vs. Stochastic (Non-Deterministic):
  • Deterministic: The next state of the environment is completely determined by the current state and the agent’s action. There is only one possible outcome for each action.
  • Stochastic (Non-Deterministic): The next state of the environment is not fully determined by the current state and the agent’s action. There are multiple possible outcomes, and the agent might not know exactly which one will occur.
• Static vs. Dynamic:
  • Static: The environment remains unchanged while the agent is deliberating or acting.
  • Dynamic: The environment can change while the agent is thinking or taking action. This requires the agent to be more reactive.
• Discrete vs. Continuous:
  • Discrete: Finite or countably infinite number of states and actions.
  • Continuous: States and actions can take on a continuous range of values.
• Single Agent vs. Multi-Agent:
• Episodic vs. Sequential:

Fully Observable vs. Partially Observable Environments

Deterministic vs. Stochastic Environments

Static vs. Dynamic Environments

Discrete vs. Continuous Environments

Problem-Solving Agents in AI

A problem-solving agent is a type of goal-based agent that aims to find a sequence of actions that will lead it from an initial state to a desired goal state. It formulates a problem, searches for a solution (a sequence of actions), and then executes the actions.

Key Characteristics of Problem-Solving Agents

• Focuses on achieving a specific goal.
• Uses search algorithms to explore the space of possible actions.
• Does not necessarily interact with a changing external environment during the problem-solving process (it might operate in a static or discrete environment).

Example: A route-finding agent that needs to find the shortest path between two cities. It formulates the problem (initial city, destination city, possible roads), searches for a path using algorithms like A*, and then outputs the sequence of cities to travel through.

Types of Problems in Artificial Intelligence

Problems in AI can be categorized in several ways:

• Well-defined: The initial state, goal state, and the set of possible actions are clearly specified. Most problems studied in AI are well-defined.
• Ill-defined: The goal state or the set of actions is not clearly specified, making it harder to determine when a solution has been reached or what actions are permissible.
• Single-state vs. Multiple-state:
  • Single-state: The agent knows the initial state and can predict the outcome of its actions.
  • Multiple-state: The agent might not know the exact initial state or the outcome of its actions might be uncertain. This often occurs in partially observable or stochastic environments.
• Deterministic vs. Non-deterministic: (This relates to the environment, but also characterizes the problem-solving process)
  • Deterministic: Each action leads to a unique next state.
  • Non-deterministic: Actions can have multiple possible outcomes.
• Discrete vs. Continuous:
  • Discrete: The state space and action space are discrete.
  • Continuous: The state space and action space are continuous.
• Search vs. Planning:
  • Search Problems: Finding a sequence of actions to reach a goal without explicit consideration of time or resources.
  • Planning Problems: Similar to search, but often involve more complex goals, resource and time constraints, and potentially multiple agents.

Steps for AI Problem Solving

The general steps involved in problem solving using AI are:

1. Problem Formulation: Define the problem precisely, including the initial state, goal state, the set of possible actions, and the cost of each action (if relevant).
2. Search: Explore the space of possible action sequences starting from the initial state to find a path to the goal state. This involves using various search algorithms.
3. Solution: Once a path to the goal state is found, the sequence of actions along this path constitutes the solution.
4. Execution: The agent performs the actions in the found sequence to reach the goal in the real environment.

Components of AI Problem Formulation

Problem formulation in AI involves precisely defining the elements needed to solve a problem. These components are:

• Initial State: A description of the starting situation or configuration of the problem. This is where the agent begins its problem-solving journey.
• Goal State: A description of the desired situation or configuration that represents the solution to the problem. There can be one or more goal states.
• Actions: A set of operators or possible moves that the agent can take to transition from one state to another. Each action has a precondition (the state in which it can be applied) and an effect (the resulting state after the action is performed).
• State Space: The set of all possible states that can be reached from the initial state through any sequence of actions. The search for a solution takes place within this space.
• Path Cost (Optional but often important): A function that assigns a numerical cost to each path (sequence of actions) from the initial state to another state. This is crucial when the agent needs to find the optimal solution (e.g., the shortest path, the least expensive sequence of actions).

AI Search Algorithms: Purpose and Types

The primary use of search algorithms in Artificial Intelligence is to find a sequence of actions (a path) from an initial state to a goal state within a problem’s state space. When a problem is formulated, the solution is often hidden within the vast number of possible states and transitions. Search algorithms systematically explore this state space to locate a path that satisfies the goal condition.

Types of Search Algorithms

Here are the different types of search algorithms, broadly categorized:

Uninformed Search (Blind Search)

These algorithms do not have any information about the location of the goal state other than the problem definition itself. They explore the state space in a systematic way. Examples include:

• Breadth-First Search (BFS)
• Depth-First Search (DFS)
• Depth-Limited Search (DLS)
• Iterative Deepening Depth-First Search (IDDFS)
• Uniform-Cost Search (UCS)

Informed Search (Heuristic Search)

These algorithms use domain-specific knowledge, often in the form of heuristic functions, to guide the search towards the goal. Heuristic functions estimate the cost from the current state to the goal state. Examples include:

• Greedy Best-First Search
• A* Search
• Memory-Bounded Heuristic Search (e.g., IDA*, MA*)

Local Search Algorithms

These algorithms operate by starting from an initial state and iteratively trying to improve the current state until a goal state is reached or a satisfactory solution is found. They do not systematically explore paths. Examples include:

• Hill Climbing
• Simulated Annealing
• Genetic Algorithms
• Local Beam Search

Uninformed Search vs. Informed Search

The key distinction between uninformed and informed search lies in whether they use any knowledge beyond the problem definition itself. Uninformed search, also known as blind search, operates without any clue about how close a state is to the goal. These algorithms systematically explore the state space, trying all possibilities in a predefined order until a solution is found. Think of it like wandering through a maze without a map or any idea of where the exit might be. Examples include Breadth-First Search, Depth-First Search, and Uniform-Cost Search. While some uninformed methods like BFS and UCS can guarantee finding the shortest or least-cost path, they can be very inefficient, especially in large state spaces, as they might explore many irrelevant paths.

In contrast, informed search, also called heuristic search, leverages domain-specific knowledge to guide the search process. These algorithms use heuristic functions, which estimate the “distance” or cost from the current state to the goal state. This extra information allows the search to prioritize exploring paths that seem more promising, potentially leading to a solution much faster. Imagine having a map of the maze that gives you an idea of the direction of the exit. A* search and Greedy Best-First Search are prominent examples of informed search. The effectiveness of informed search heavily depends on the quality of the heuristic function; a good heuristic can significantly speed up the search, while a poor one might lead the search astray or offer no advantage over uninformed methods. If the heuristic is “admissible” (never overestimates the true cost to the goal) and consistent, A* can guarantee finding the optimal solution more efficiently than uninformed approaches.

Breadth-First Search (BFS) vs. Depth-First Search (DFS)

Breadth-First Search (BFS) and Depth-First Search (DFS) are two fundamental uninformed search algorithms that explore the state space in different ways.

BFS explores the state space level by level. It starts at the initial state, then examines all its immediate neighbors, then all the neighbors of those neighbors, and so on. It is like exploring a tree by looking at all the siblings before moving down to their children. BFS uses a queue to keep track of the nodes to visit. This level-by-level exploration guarantees that if a solution exists at a finite depth, BFS will find the shallowest solution (the one with the fewest steps). However, BFS can be memory-intensive because it needs to store all the nodes at the current level.

DFS, on the other hand, explores as far as possible along each branch before backtracking. It starts at the initial state and goes down one path until it reaches a dead end or a goal state. Then, it backtracks to the last unexpanded node and explores another branch. Think of it as going down one hallway in a maze until you hit a wall, then turning back and trying another hallway. DFS typically uses a stack (implicitly through recursion) to manage the nodes to visit. DFS can be more space-efficient than BFS if the search depth is large, as it only needs to store the current path. However, DFS is not guaranteed to find the shortest solution, and it can get stuck in infinite loops if the state space contains cycles or is infinite.

In summary, BFS prioritizes exploring broadly, level by level, ensuring the shortest path is found. DFS prioritizes exploring deeply along a single path, potentially finding a solution quickly but without any guarantee of optimality or even completeness in certain scenarios.