Core Concepts of Artificial Intelligence: Foundations & Techniques

Posted on Sep 5, 2025 in Other university degrees and diplomas

Understanding Artificial Intelligence Fundamentals

Artificial Intelligence (AI) is the simulation of human intelligence in machines that can learn, reason, and make decisions. AI enables systems to perform tasks such as problem-solving, speech recognition, learning from data, and decision-making without explicit human instructions. AI is classified into Weak AI (narrow AI, like Siri) and Strong AI (human-level AI, like AGI). AI systems use techniques like machine learning, deep learning, and neural networks to mimic human thought processes. Examples include chatbots, self-driving cars, and recommendation systems.

Categories of AI Definitions

AI definitions can be categorized based on capability and functionality:

  • Based on Capability:
    • Weak AI (Narrow AI): Performs specific tasks (e.g., Chatbots).
    • Strong AI (General AI): Mimics human intelligence.
    • Super AI: Hypothetical AI that surpasses human intelligence.
  • Based on Functionality:
    • Reactive AI: Responds to inputs (e.g., Chess AI).
    • Limited Memory AI: Uses past experiences (e.g., Self-driving cars).
    • Theory of Mind AI: Understands emotions (future research).
    • Self-Aware AI: Fully conscious AI (not yet developed).

AI Agents: Components & Practical Examples

An AI agent is an entity that perceives its environment and takes actions. Its key components include:

  • Sensors – Collects data from the environment (e.g., cameras, microphones).
  • Actuators – Performs actions (e.g., robotic arms, speakers).
  • Perception Module – Interprets sensory data.
  • Reasoning and Decision-Making – Uses AI techniques to make decisions.
  • Learning Module – Improves based on past experiences.

Self-Driving Car Agent Example

Consider a self-driving car as an AI agent:

  • Sensors: Cameras, radar, LiDAR.
  • Actuators: Steering wheel, brakes.
  • Decision-Making: AI predicts traffic behavior and navigates.

Types of AI Agents Explained

AI agents can be classified into several types based on their complexity and decision-making processes:

  • Simple Reflex Agents – Reacts based on current perception, without history (e.g., thermostat).
  • Model-Based Agents – Uses memory to improve decisions (e.g., chess AI storing past moves).
  • Goal-Based Agents – Makes decisions to achieve a specific goal (e.g., pathfinding AI in GPS).
  • Utility-Based Agents – Maximizes performance with optimization (e.g., self-driving car balancing speed & safety).
  • Learning Agents – Adapts and improves over time (e.g., recommendation systems like Netflix).

Knowledge Representation Methods in AI

Effective knowledge representation is crucial for AI systems. Common methods include:

  • Logical Representation: Propositional and Predicate logic.
  • Semantic Networks: Graph-based relationships.
  • Frames & Ontologies: Structured knowledge representation.
  • Production Rules: IF-THEN rules for decision-making.

Propositional vs. Predicate Logic: A Comparison

Propositional Logic deals with statements (propositions) that are either true or false. It uses logical connectives like AND, OR, NOT, implication (→), and biconditional (↔) to form complex expressions. An example of propositional logic is “It is raining” (P), which is a simple statement without variables. However, propositional logic lacks the ability to express relationships between objects.

Predicate Logic extends propositional logic by introducing predicates, variables, and quantifiers. It allows for more complex reasoning by defining properties of objects and relationships between them. Predicate logic uses universal (∀) and existential (∃) quantifiers to express general or specific truths. For example, the statement “All humans are mortal” can be represented as ∀x (Human(x) → Mortal(x)), meaning for every x, if x is a human, then x is mortal. Predicate Logic is more expressive and is widely used in artificial intelligence, databases, and knowledge representation, whereas Propositional Logic is primarily used in simpler reasoning and digital circuit design.

Skolemization in First-Order Logic (FOL)

Skolemization is a process used in First-Order Logic (FOL) to eliminate existential quantifiers (∃) by replacing them with Skolem functions or constants. This step is crucial when converting a logical formula to Conjunctive Normal Form (CNF) for automated reasoning and theorem proving.

A* Search Algorithm for Pathfinding

A* is an informed search algorithm used for pathfinding and graph traversal. It combines the advantages of Breadth-First Search (BFS) and Greedy Best-First Search, ensuring an optimal and efficient solution.

Conjunctive Normal Form (CNF) Explained

Conjunctive Normal Form (CNF) is a standard way of representing logical expressions as a conjunction (AND) of disjunctions (OR) of literals. In CNF, a logical formula is expressed as a set of clauses, where each clause is a disjunction (OR) of literals.

Converting First-Order Logic to CNF Steps

To convert a First-Order Logic (FOL) statement to CNF, follow these steps:

  1. Eliminate Implications & Biconditionals: Replace P→Q with ¬P∨Q; Replace P↔Q with (P∨¬Q)∧(¬P∨Q).
  2. Move Negation (NOT) Inward (Use De Morgan’s Laws): ¬(P∨Q) → ¬P∧¬Q; ¬(P∧Q) → ¬P∨¬Q.
  3. Standardize Variable Names (Avoid Conflicts): Ensure that variables are uniquely named across quantifiers.
  4. Eliminate Existential Quantifiers (∃) Using Skolemization: Replace existential variables with a function of universally quantified variables.
  5. Move Universal Quantifiers (∀) to the Left (Prenex Normal Form): Ensure that all universal quantifiers appear at the beginning of the formula.
  6. Distribute OR over AND (Convert to CNF Form): Use distributive law: P∨(Q∧R) → (P∨Q)∧(P∨R).
  7. Remove Universal Quantifiers: In CNF, universal quantifiers are implicit, so they can be dropped.

Hill Climbing Algorithm: Key Disadvantages

The Hill Climbing Algorithm, while simple, has several disadvantages:

  • Local maxima problem: Gets stuck at suboptimal solutions.
  • Plateau problem: Flat regions slow learning.
  • Ridge problem: Difficult to navigate narrow paths.
  • No backtracking: Cannot recover from poor choices.
  • Susceptibility to noise: Sensitive to small changes in input.

Forward and Backward Chaining: AI Reasoning

Forward Chaining and Backward Chaining are fundamental reasoning techniques used in AI for inference. Forward Chaining follows a data-driven approach, starting from known facts and applying rules to derive new conclusions until it reaches the goal. It is useful when all facts are available but the outcome is unknown, such as in medical diagnosis where symptoms lead to a disease prediction. An example is: “It is raining → Carry an umbrella.”

In contrast, Backward Chaining follows a goal-driven approach, starting with a hypothesis (goal) and working backward to verify if supporting facts exist. This method is efficient when the goal is known but the supporting facts need to be checked, such as in debugging systems, where the root cause of a failure is identified. An example is: “I need an umbrella → Is it raining?”

While Forward Chaining processes many facts, making it computationally expensive, Backward Chaining focuses only on relevant facts, making it more efficient.

Alpha-Beta Pruning for Game Trees

Alpha-beta pruning is an optimization technique used in the Minimax algorithm to reduce the number of nodes evaluated in a game tree. It eliminates branches that do not affect the final decision, improving efficiency.

Alpha (α):
The best (maximum) value that the maximizing player can guarantee so far.
Beta (β):
The best (minimum) value that the minimizing player can guarantee so far.

If α ≥ β at any node, further exploration of that branch is unnecessary (pruning occurs).