Introduction to Artificial Intelligence – My Structured Learning Notes

#ai #machinelearning #generativeai #learning

Introduction to Artificial Intelligence

Artificial Intelligence is everywhere — from your phone’s recommendations to chatbots on websites. But what really makes AI tick?

I recently started documenting my AI learning in a structured way to build strong fundamentals before jumping straight into tools. Here’s a concise summary of what I learned.

1. What is AI?

AI refers to systems designed to perform tasks that normally require human intelligence — like reasoning, learning, and decision-making.

Types of AI:

Weak AI: Task-specific systems like Siri or recommendation engines.
Strong AI: Human-level intelligence across domains (still theoretical).

Where AI shows up: Chatbots, expert systems, automation tools.

Key point: Weak AI can appear intelligent without true understanding.

2. Popular Uses of AI

Predictive AI: Learns from historical data to make predictions.

Example: Amazon product recommendations.
Generative AI: Learns from massive datasets to create new content — text, images, or code.

Examples: ChatGPT, DALL·E

More data usually improves predictions — but beware of bias.

3. Machine Learning (ML)

ML teaches machines to learn patterns from data instead of explicit instructions.

Artificial Neural Networks (ANNs): Inspired by the brain and useful for complex datasets.
ML allows systems to improve over time through experience.

ML Workflow:

Raw Data
│
▼
Data Preprocessing
│
▼
Feature Extraction
│
▼
ML Algorithm
│
▼
Predictions
│
▼
Feedback / Error
└─────────► Improve Model

4. Common AI Systems

Pattern recognition: Detects patterns humans can’t easily see (insurance, healthcare).
Robotics: Combines ML with sensors (self-driving cars).
Natural Language Processing (NLP): Machines process and generate language — context matters!
Internet of Things (IoT): Devices collect real-world data to feed AI (healthcare, behavior prediction).

5. Learning from Data

Supervised learning: Uses labeled data (e.g., spam detection).
Unsupervised learning: Finds structure in unlabeled data (e.g., customer segmentation).
Data models: Represent learned knowledge and improve over time.

Types of Learning:

Labeled Data -> Supervised
Unlabelled Data -> Unsupervised

6. Identifying Patterns

Classification: Predict categories (fraud detection).
Clustering: Group similar data (market segmentation).
Reinforcement learning: Learn via rewards/penalties (recommendation strategies).

7. ML Algorithms

Common algorithms include: KNN, K-Means, Regression, Naive Bayes

Algorithm choice depends on your data and problem.

8. Accuracy Matters

Bias: Systematic error from assumptions in the model.
Variance: Error from over-complexity.
Overfitting: Memorizing noise.
Underfitting: Missing patterns.

Bias-Variance Tradeoff:
High Bias ──► Underfitting Optimal
High Variance ──► Overfitting

9. Artificial Neural Networks

Structure: Input → Hidden → Output layers

Learning: Adjust weights, tune biases, use backpropagation

Simple ANN Diagram:

Input Layer Hidden Layer Output Layer
○ ○ ○
○ ─────► ○ ─────► ○
○ ○ ○

10. Improving Accuracy

Cost function: Measures error
Gradient descent: Minimizes error iteratively

11. Generative AI

Self-supervised learning: Uses pseudo-labels on unlabeled data
Foundation models: Multi-purpose AI systems
Large Language Models (LLMs): Predict words by probability (no true understanding)