How Machine Learning Works: Explained Simply (Part 2)
- Angie Okhupe
- Feb 14
- 4 min read
In my last post, we explored machine learning at a high level—how machines gather data, recognize patterns, build models, and refine them over time. We compared it to learning to drive—starting with practice, recognizing patterns on the road, and refining skills over time. But here’s the million-dollar question: How does a machine actually learn from that data? What’s happening under the hood?
It’s kind of like when someone tells you, “Just be yourself!” …Okay, but how? Do I dance? Do I make a joke? Do I recite fun facts about penguins? A little guidance would be nice! 😅
Well, machines learn in three distinct ways—just like students with different learning styles:
The Rule-Follower (Supervised Learning): Learns from labeled examples, like memorizing a driver’s manual.
The Explorer (Unsupervised Learning): Discovers hidden patterns in messy, unlabeled data.
The Trial-and-Error Pro (Reinforcement Learning): Masters skills through practice and feedback—like nailing parallel parking after a few failed attempts.
Each of these methods powers the technology we use every day, from Netflix recommendations to self-driving cars and everything in between. Some systems rely on just one type of learning, but many combine multiple approaches to improve accuracy and adaptability. By mixing these techniques, machines become smarter, more flexible, and better at adapting—just like humans do!
In this post, we’re cracking open the “how” of machine learning so you can see exactly what’s happening under the hood [see what i did there?]. Ready? Buckle up!
Learning Type 1: The Rule-Follower (Supervised Learning) 🏁
When you start learning to drive, you don’t just get in the car and go. Someone (a driving instructor, a parent, or YouTube) teaches you the basics and shows you how it’s done.
You watch and learn: You see how others drive, when to stop, how to use turn signals, and how to handle different roads.
You practice with feedback: You try driving yourself, and if you mess up, your instructor yells “BRAKE!” before you hit something.
You get better over time: Eventually, you start recognizing patterns — like paying attention when the light turns green so you can start driving before the person behind you honks at you or when the red light is coming on so you can stop on time.
💡Just like when you learn to drive with an instructor guiding you—telling you when to stop, when to turn, and how to parallel park—supervised learning works by training a machine with labeled data. The model learns from examples, just as a new driver learns from step-by-step instructions before confidently driving on their own.
Example: If an AI is learning to recognize apples, you give it thousands of images labeled “apple” and “not an apple” until it figures out the difference.
Learning Type 2: The Explorer (Unsupervised Learning) 🤔
Once you’ve got some driving experience, you’ll run into new situations no one specifically trained you for:
🚧 What if there’s road construction blocking your usual route?
🐢 What if someone in front of you is driving way too slow?
🗺️ What if Google Maps reroutes you through a sketchy neighborhood?
At this point, you figure things out based on patterns and past experiences—no one is giving you explicit instructions, but you adapt. Just like when you drive in a new place and adapt based on experience, ML algorithms do the same with new, unlabeled data.
In unsupervised learning—the machine isn’t given clear labels but finds patterns on its own.
Example: On your first day at a new school, you don’t know anyone, but you start noticing groups forming—some students always talk about sports, others are into music, and a few are really into tech. Without being told, you get a sense of which groups exist based on their shared interests. This is exactly how unsupervised learning works! A machine analyzes people’s behaviors (like social media interests or shopping habits) and clusters similar users together—for example, grouping online shoppers into "tech lovers," "fashion enthusiasts," or "budget-conscious buyers" without being explicitly told what category they belong to.
Learning Type 3: The Trial-and-Error Pro (Reinforcement Learning) 🏆
Let’s be real—when you first start driving, you make mistakes. Maybe you brake too hard, misjudge a turn, or (gasp) forget to signal. But over time, you learn from trial and error.
If you take a turn too fast, your heart races -->you slow down next time.
If you park badly and scratch the car --> you adjust and try again.
If you get honked at for waiting too long at a green light --> you learn not to daydream at intersections.
The same way you adjust your driving based on real-world consequences, the machine learns by exploring different actions, receiving rewards or penalties, and refining its decisions over time. This is reinforcement learning—where machines learn by trial and error, receiving rewards for good actions and penalties for mistakes.
Example: When you train a puppy, you don’t explain commands in detail—it learns through trial and error. If the puppy sits when told, it gets a treat (reward) 🦴. If it jumps on the couch, it gets scolded (penalty) 🚫. Over time, the puppy learns which behaviors lead to rewards and repeats them.
So, what’s the big takeaway? Machines don’t just “know” things—they learn, just like we do! Whether they’re following instructions (supervised learning), figuring things out on their own (unsupervised learning), or improving through trial and error (reinforcement learning)—they’re basically just overachieving students who never need coffee or sleep.
Next time Netflix perfectly predicts your next binge, your phone suggests the exact emoji you were thinking of, or a self-driving car smoothly merges into traffic, just remember—it’s machine learning working behind the scenes!
And just like us, these machines never stop learning, improving, and getting better. 🚗💡
Great job for making it to the end of this post. Now, before you go, let me know, if you were an AI, how do you think you would learn, Let me know in the poll below:
If you were an AI, how would you learn best?
The Rule-Follower: Just give me a manual, and I’ll memorize
The Explorer: Let me figure it out on my own
The Trial-and-Error Pro : Keep trying until I get it right
And finally, incase you missed part one of how ML works, you can read that here:
Comentarios