Last month a friend asked me: "So you're learning AI, does that mean you're building a robot?" That question says a lot. Most people picture artificial intelligence either as a humanoid robot out of science fiction, or as some kind of magic locked inside a black box. Both are wrong.
Over this 30-part series, we'll rebuild what you know about AI from the ground up. Not by drowning you in jargon, but not staying shallow either. The first few chapters will be intuitive, we'll build a feel for the concepts. Then we'll gradually work into the mechanics. By the end of the series you'll be able to use tools like ChatGPT deliberately, like an informed user, and also answer "why does this model behave that way?" like an engineer.
But first, let's start with the truly fundamental question.
What Does AI Actually Do?
Think of it this way. Write a rulebook: "If the email contains 'you've won free money', mark it as spam." That's a rule. A human writes it, the computer applies it. The computer doesn't learn anything, it just follows the instruction.
Now try a different approach. Give the computer 10,000 emails. Half are spam, half aren't. Say "figure out which ones are spam yourself." The computer finds patterns in that data, and over time it can look at a new email and say "this is probably spam." You didn't write a rule. You made it learn.
That second approach is the essence of machine learning. And machine learning is the most powerful branch of AI today.
The concept of AI itself is broader: it can be defined as "machines performing tasks that normally require human intelligence." But that definition is a bit slippery, because what counts as "a task that requires intelligence" keeps changing. In the 1990s, playing chess required intelligence. Today a phone app does it. Nobody calls that AI anymore.
In practical terms, when people say "AI" today they usually mean one of these three things:
- Machine Learning (ML): systems that learn patterns from data
- Deep Learning: the powerful branch of machine learning that uses neural networks
- Large Language Models (LLMs): massive text-based models like ChatGPT, Claude, Gemini
Here's how the three relate: deep learning is a type of machine learning. LLMs are a specific application of deep learning. Think of them as nested circles.
You're Already Using It, Without Noticing
This isn't your first encounter with AI. You've probably been living inside it for years.
Spotify and Netflix recommendations: the playlist you open on your morning commute runs on "people who like this artist also like that one" logic. That's a machine learning model. It analyzes your listening habits, compares you to similar profiles, and produces a personalized prediction. Recommendation systems are complex sets of algorithms companies use to predict what you might want to buy, watch, hear, or read.
Face recognition: your phone's lock screen. Passport control at the airport. Your photo app automatically tagging people. All the same technology, models that turn the patterns in an image into numerical data and compare them.
Gmail's spam filter: back to the example above. Gmail catches spam among billions of emails every day. Not with a list of rules, with machine learning.
Google Translate: when it translates a sentence from Turkish to English, it doesn't look words up in a dictionary one by one. It uses language patterns learned from billions of translation examples.
Voice assistants: when you say "Hey Siri" or "OK Google," your voice first passes through a speech recognition model (audio to text), then a natural language understanding model (text to intent), then a response generation model (intent to speech). Three separate AI models running in sequence.
What all of these examples have in common: they all learned from data. Nobody wrote a rule telling these systems "this is what spam looks like" or "this is what English means." The systems extracted their own patterns from a large number of examples.
"Intelligent," or "Statistical"?
Here's an important question worth asking: are these systems actually intelligent?
The honest answer: no, at least not compared to human intelligence. AI models are, in practice, incredibly complex statistical pattern-matching machines. They don't "understand," they just find patterns.
When ChatGPT tells you "Paris is the capital of France," it doesn't "know" that, it learned that these two concepts consistently appear together in its training data. In most cases this distinction doesn't matter, the answer comes out correct. But in some cases it's critical: a model can state something it saw often but that happens to be wrong, with exactly the same confidence. We'll come back to this in later chapters under the name "hallucination."
For now, hold onto this: AI is not a machine that thinks like a human. AI is a system that learns patterns from data and makes predictions using those patterns.
If you plant that distinction in your head, you'll find it much easier to understand both "why this works" and "why it sometimes fails spectacularly" in the chapters ahead.
Where Are We in 2024-2025?
AI research started decades ago, but the last 3-4 years have genuinely been transformative. Business adoption accelerated significantly in 2024; the number of newly founded generative AI startups nearly tripled. According to Stanford's 2025 AI Index report, job postings in the US seeking AI skills grew by 20 percent between 2023 and 2024.
Models like GPT-4o, Claude, and Gemini no longer just write text, they understand images, generate code, and talk out loud. Generative AI, with its ability to produce text, images, and video, has worked its way into everyday life. And it's no longer just a tech-industry story; healthcare, law, education, finance, every field is being reshaped.
That pace can feel frightening. But know this: the foundations of this technology go back to the 1950s. And someone who understands those foundations can put even today's most advanced system into a reasonable perspective. That's exactly the goal of this series.
What's Ahead in This Series?
Think of the 30-chapter roadmap like this:
Chapters 1-3 (you are here): what AI is, how it learns, the intuition behind neural networks
Chapters 4-7: inside language models, tokens, attention, transformers; prompt engineering; RAG and agents
Chapters 8-10: infrastructure, limitations, integrating this into real projects
Each chapter builds on the last. If you read them in order without skipping, by the end of the series you'll be able to use AI tools deliberately as a user, and explain why these systems behave the way they do as an engineer.
Next chapter, we tackle the first big question: how does a machine actually "learn"? We'll cover data, models, and training, without writing a single line of code, but in enough depth to really understand it.
This series is built for people with zero prior AI background, but it refuses to stay shallow. Each chapter builds on the last. Reading from the start gives you the best experience.
Abdulkadir Akyurt is a QA Engineer and Test Automation Specialist.
Follow me on Medium for more articles · Connect on LinkedIn · Instagram · X / Twitter · Visit abdulkadirakyurt.com