Foundations of Machine Learning
What Is Machine Learning?
4 min read
A fraud problem that rules can't keep up with
ABC Inc., an e-commerce company, is struggling with a surge in fraudulent transactions on its website. Its manual review process can't keep pace — orders are delayed, legitimate customers get frustrated waiting for approval, and the fraud keeps slipping through anyway.
The obvious fix might seem to be: write more rules. Flag any order over $500. Flag any order from a new account. Flag any order shipped to a different address than the billing address. But fraud patterns shift constantly, and every new rule either blocks real customers or misses a new trick scammers have already moved on to.
ABC Inc. decides to try something different: instead of programming explicit rules, it will feed its system thousands of past transactions — some fraudulent, some legitimate — and let the system learn the difference on its own. That system will also learn each customer's purchase history to improve product recommendations along the way.
That "let the system learn on its own" idea is machine learning, and it's what this course is about.
Defining machine learning
Machine learning (ML) is a subset of artificial intelligence that enables systems to learn and improve automatically from experience, without being explicitly programmed for every scenario they'll encounter.
Two ideas in that definition matter more than the rest:
- Subset of AI — machine learning is one way to build AI, not a separate field. Every machine learning system is an AI system; not every AI system uses machine learning.
- Learns from experience — instead of a human anticipating every rule in advance, the system is shown examples and works out the pattern connecting them.
The term isn't new. Arthur Samuel coined "machine learning" in 1959, describing it as giving computers "the ability to learn without being explicitly programmed" — while building a checkers-playing program that improved by playing games against itself. Decades before smartphones or the cloud, the core idea was already in place: a program that gets better through experience rather than through a programmer rewriting its rules.
Traditional programming vs. the machine learning approach
The clearest way to see what's different about machine learning is to put it next to the way software has traditionally been built.
| Traditional programming | Machine learning |
|---|---|
| Uses predefined rules and logic explicitly written by human developers | Learns patterns from data to make predictions or decisions |
| Relies on rules a person has to think of in advance | Uses statistical techniques to discover patterns on its own |
| Requires manual feature engineering — a human decides which details in the data actually matter | Automatically learns which features matter directly from raw data |
| Struggles with complex, unstructured data (images, text, audio) without heavy preprocessing | Handles images, text, and audio directly, with far less manual preprocessing |
| Accuracy is capped by how complete and correct the human-written rules are | Accuracy tends to improve as more data and training iterations are added |
Put simply: a traditional program is a fixed recipe a human writes in full. A machine learning model is a recipe the system writes for itself, by studying enough finished dishes to infer what the recipe must have been.
Why this distinction matters
Fraud detection is a perfect example of why the traditional approach breaks down. Rules are static; fraud is not. A rules engine can only catch patterns a human has already seen and written a rule for. A machine learning model, trained on a large and continually updated set of past transactions, can pick up on subtler and shiftier patterns — combinations of order size, timing, location, and account age that no one explicitly programmed it to look for.
That's the trade a company makes when it adopts machine learning: less manual rule-writing, more dependence on having enough good-quality data to learn from. The rest of this course is about understanding that trade in more depth — how it works, what forms it takes, and what tools are used to build it.
Key takeaway
Machine learning is a subset of AI in which systems learn patterns directly from data instead of following rules a human has explicitly programmed. Coined by Arthur Samuel in 1959, the idea has become the dominant way modern software handles problems — like fraud detection — where the rules are too complex, too numerous, or too fast-changing for a person to write by hand.
What's next?
Machine learning existed conceptually since 1959, but it's only in the last couple of decades that it has become a default part of how software gets built. Next, we'll look at what changed — the practical factors that drove machine learning's adoption.