Understanding AI

Artificial Intelligence (AI) is often seen as something futuristic, but it is actually the result of decades of evolution in how humans collect, process, and understand data. This document explains key AI concepts, such as the historical eras of computing, data structures, and machine learning types, using accessible examples from the real world.

The eras of data: tabulation, coding and AI

The era of tabulation (Before 1950s)

This was the mechanical era, when data processing relied on manual or mechanical systems.

• Example: Census data was processed using punched cards and tabulating machines invented by Herman Hollerith in the 1890s (which later led to IBM).
• Computers didn’t “think”; they just counted and sorted data mechanically.

The era of coding (1950s - 2000s)

This was the programming era. Humans wrote explicit instructions (code) that computers followed step-by-step.

• Example: If you wanted a computer to sort names alphabetically, you wrote a sorting algorithm (like Bubble Sort or QuickSort).
• The computer only did what it was told, nothing more, nothing less.

The era of Artificial Intelligence (2000s - Today)

In this modern era, computers learn from data instead of being explicitly programmed.

• Example: Instead of coding the rules for identifying cats in pictures, we feed the computer thousands of labeled cat and non-cat images, and it learns the patterns by itself.
• The “code” is not written by humans but discovered by the machine through learning algorithms.

Types of Data: From chaos to structure

AI depends heavily on the type and quality of data it learns from. Data can be divided into three main categories.

Structured data

Data that follows a defined format, usually stored in tables or databases.

• Example: An Excel spreadsheet of customers with columns like “Name,” “Age,” and “Purchase History.”
• Traditional programming and data analysis tools handle structured data efficiently.

Non-structured (unstructured) data

Data that doesn’t follow a specific format or is difficult to analyze directly.

• Examples: Text in emails, social media posts, audio recordings, images, and videos.
• AI models like GPT (for text) or image recognition systems (for photos) are designed to interpret this kind of data.

Obscure data

Data that exists but is not accessible or not digitized.

• Example: Handwritten notes in a scientist’s notebook or conversations that were never recorded.
• AI cannot use this data unless it’s made digital.

Coding vs. Automatic Learning

Aspect	Coding	Automatic (Machine) Learning
Definition	Writing explicit instructions that tell the computer what to do.	Letting the computer find patterns or rules from examples (data).
Human Role	The human defines the rules.	The human provides the data; the machine defines the rules.
Example	You code: “If temperature > 30°C, then turn on air conditioning.”	The AI learns from many days of data (temperature vs. comfort level) to decide when to turn it on automatically.
Flexibility	Limited to what’s explicitly programmed.	Adapts to new data and situations.

Automatic learning allows computers to improve with experience, just like humans do when they gain knowledge over time.

Types of AI learning

There are three main approaches to machine learning: supervised, unsupervised, and reinforcement learning.

Supervised Learning

The AI learns from labeled data, that is, examples with known answers.

• Example: A spam filter is trained on thousands of emails labeled “spam” or “not spam.”
• The system learns the characteristics of each and predicts labels for new emails.
• Analogy: Like a student learning with a teacher who provides the correct answers.

Unsupervised Learning

The AI learns from unlabeled data, discovering patterns or groups by itself.

• Example: A marketing system that groups customers into segments based on buying habits without being told what those segments should be.
• Analogy: Like a researcher who explores a new planet and groups creatures by similarity, without knowing what species they are.

Reinforcement Learning

The AI learns by trial and error, receiving rewards or penalties for its actions.

• Example: A robot learns to walk by getting a reward every time it moves forward and a penalty when it falls.
• Analogy: Like training a dog with treats, it learns which behaviors lead to rewards.
• This method is used in self-driving cars, game-playing AIs (like AlphaGo), and robotics.