How Do Machines Learn?

A simple roadmap of all learning methods in Machine Learning (ML)

Think of a robot trying to become smarter.
There are only a few basic ways we can teach it:

1. Supervised Learning

You show the machine examples and give it the answers.

It learns by comparing its guesses to the correct answers and adjusting.

Example:
You show it 100 photos of cats and dogs labeled as “cat” or “dog”.
It learns to tell them apart.

We'll dive into this next!

2. Unsupervised Learning

You give it data, but no answers.

It tries to find patterns or group similar things.

Example:
You give it 1,000 customer reviews with no ratings.
It learns to group similar ones together (like “angry” vs. “happy”).

3. Reinforcement Learning (RL)

You give it a goal, and it learns by trial and error.

It gets rewarded or punished and learns what works.

Example:
A robot in a maze learns the right path by getting points for moving closer to the exit.

4. Self-Supervised Learning

It creates its own learning task from data.

Often used in modern AI like GPT.

Example:
It hides part of a sentence and tries to predict it:
“I love ____ cream.”
→ Learns to fill in “ice”.

5. Semi-Supervised Learning

A mix of supervised + unsupervised.

A small portion of data has answers; the rest doesn’t.

Example:
You label 10% of photos as “cat/dog,” and it uses those to guess the rest.

6. Evolutionary Algorithms / Genetic Learning

It evolves like nature does—through survival of the fittest.

Example:
Hundreds of bots try different walking styles. The best walkers "survive" and pass on traits.

7. Symbolic Learning / Rule-Based AI

You give it clear rules.

“If X, then do Y” — less learning, more logic.

Example:
“If it’s raining, take an umbrella.”
This is old-school AI.

Summary Roadmap

Learning Type	What it Learns From	Real-Life Analogy
Supervised	Data + Answers	Student with an answer key
Unsupervised	Only Data	Sorting laundry by color
Reinforcement	Rewards & Penalties	Training a puppy with treats
Self-Supervised	Patterns within data	Finishing your own sentences
Semi-Supervised	Some data is labeled	Getting hints on a test
Evolutionary	Trial + mutation + survival	Nature / evolution
Symbolic	Hand-written logic rules	Flowcharts or rulebooks