Chapter 1 / 30, "AI from 0 to 100" series
A friend said to me last year: "Does this ChatGPT thing actually think, or is it just a really good search engine?" I paused before answering. Because that question actually sits in a genuinely accurate spot, neither completely wrong nor completely right. And I think the most common mistake people make about AI starts exactly there: either treating it as something mysterious and magical, or dismissing it as "just statistics."
This series isn't trying to preserve that mystique, nor reduce it to a dull math exercise. Over thirty chapters I'll explain, from scratch and honestly, how AI actually works, from tokens to transformer architecture, from GPU clusters to hallucination risk. This first chapter has no jargon, no math. Just an honest answer to one question: how does a machine behave intelligently?
AI Is Working Even While Your Brain Is Tired
You wake up in the morning and pick up your phone. It won't unlock without seeing your face, face recognition. You check your messages, a pile of promotional emails sit in the spam folder, email filtering. You open YouTube over breakfast, and it recommends a video on exactly the topic you were interested in last week, a recommendation algorithm. Once you're in the car, Google Maps has already calculated your route based on current traffic, traffic prediction.
And it's only 8:30 in the morning.
All of this is AI. But each one works on a completely different principle. Face recognition looks at pixel patterns in an image. The spam filter looks at word patterns. The recommendation algorithm looks at your behavioral history. Traffic prediction looks at sensor data and historical congestion patterns. Here's what they share: none of them were programmed to say "traffic's heavy, take the other road." They all learned it from data.
That distinction is critical. In classic software, a programmer writes every rule by hand: "if the email contains 'free gift', mark it as spam." But a system like that isn't realistic, spammers will use a different word tomorrow. AI works differently: you show it millions of emails, tell it which ones are spam, and it extracts its own patterns. You don't write the rule; the machine finds the rule in the data itself.
What Does "Learning" Mean? An Analogy
How does a human child learn to recognize a cat? Nobody hands them a list saying "a cat has four legs, pointed ears, and a tail." They see dozens, hundreds of different cats, big, small, yellow, black, long-haired, short-haired, and eventually a concept of "cat" forms in their brain. They can then recognize a cat they've never seen before.
Machine learning works on exactly this principle. You show the machine a large number of examples. For each example, you supply the correct answer. The machine extracts a "model" from those examples, meaning it tunes its internal parameters, its weights. Then it evaluates a completely new example using that model.
Here's the difference: we still don't fully know what's happening inside a human brain. Inside a machine's model, at least in principle, we can understand what's happening. Later in this series, we'll open that up and look inside.
AI, Machine Learning, and Deep Learning, in Three Sentences
People often use these three terms interchangeably, but there's a hierarchy between them:
Artificial Intelligence (AI): the broadest concept. The idea of "computers performing tasks specific to human intelligence." It emerged in the 1950s. The first chess-playing programs were also "AI," but they worked on a completely different principle than today's systems; those programs were rule-based, they didn't learn.
Machine Learning (ML): a subset of AI. The approach of "teach the machine with data instead of writing it rules." Most AI systems we use today fall into this category.
Deep Learning (DL): a subset of machine learning. The approach that uses artificial neural networks inspired by the human brain. ChatGPT, Stable Diffusion, face recognition, all of these are deep learning. Why "deep"? Because these neural networks are made up of many layers; we'll cover that "depth" in chapter three.
In everyday conversation, when people say "AI" they usually mean large, deep-learning-based models. But telling these three concepts apart, later in the series, will genuinely help you.
The Difference Between a Machine "Behaving Intelligently" and "Understanding"
Now let's face a slightly uncomfortable truth.
Netflix recommends a show that's a perfect fit. You loved it. Did Netflix "understand" that you'd love it? No. Netflix's system detected that thousands of users with viewing history similar to yours liked that show, and statistically assigned you to that group. There's no "understanding," there's pattern matching.
ChatGPT gives you a coherent, seemingly deep answer to a philosophical question. Did ChatGPT "understand" that question? The answer depends on how you define "understanding," and that's a philosophical question AI researchers are still debating. But I can say this for certain: ChatGPT learned statistical patterns from a massive amount of text data on the internet, and it generates its answer word by word based on those patterns.
I'm not saying this to dismiss the system as "just statistics." Quite the opposite: how powerful these statistical patterns turn out to be is genuinely striking. But it's important to keep the question of "does it understand or not" clear in your head, because the answer to that question also determines where AI is going to fail.
To see what this means in practice: if you ask ChatGPT "what is the population of Turkey?" you'll get a reliable answer. But if you ask "how many people died in the most recent earthquake in 2024?" and the model's training data doesn't reach that date, it can produce a wrong answer with extremely high confidence. This is called "hallucination," we'll look at it in detail in chapter 9.
What About "Artificial General Intelligence"?
You hear it all the time: "AGI is coming, AGI is dangerous." AGI, Artificial General Intelligence, describes a machine being equal to or better than a human at any intellectual task.
Today's systems are not AGI. GPT-4, Gemini, Claude, these are "narrow AI." There are things they're extremely good at, but their limits are just as clear. A language model can write you a poem, but it can't grasp a simple physical object, it "forgets" after learning (every conversation starts from zero), and it can't generate information outside its training, it can only rearrange it.
When will AGI arrive? Nobody knows. Some researchers say 5 years, others say 50. I avoid speculating on this, because there isn't even a solid definition of "general intelligence" yet.
What You Should Expect From This Series
By the end of this 30-chapter series, I want you to be able to do two things.
First, use AI tools deliberately as a "user." Know practically which tool is good for what, where it can mislead you, how to write a prompt, and how much to trust the output.
Second, understand the underlying mechanics at a basic "engineering" level. What a token is, why transformer architecture is so powerful, when fine-tuning is needed, how to integrate an API into a real project.
In the second chapter we get into "how does a machine learn": we'll cover data, models, and training intuitively. No mathematical formulas, but not shallow either, deep enough to actually understand the real mechanism.
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