AI vs machine learning vs deep learning vs GenAI

Arjun is in his final year at an engineering college in Nagpur. In one placement talk, the speaker says “AI” forty times, “ML” twelve times, “deep learning” twice and “GenAI” in every slide title. Arjun nods along. Later his friend Asha asks him, “So what is the difference?” He opens his mouth, and nothing comes out.

If that has happened to you, good news. These four words are not rivals. They fit inside each other, like an address. Once you see the nesting, you will never mix them up again.

When you post a letter to your nani, you write the address from small to big: house, area, city, state, country. Every house in Kothrud is in Pune. Every house in Pune is in Maharashtra. Every house in Maharashtra is in India. But not every house in India is in Kothrud.

The AI words work the same way. Artificial intelligence is the country. Machine learning is one state inside it. Deep learning is a city in that state. Generative AI is a busy neighbourhood in that city, the one everyone is talking about right now.

Now let us walk through each box, from the outside in, with one everyday example for each.

Remember the ludo app on your phone where you play against “the computer”? In many simple versions, nobody trained anything. A programmer wrote rules: if you can capture a piece, capture it; otherwise move the piece closest to home. It feels a little clever, and that is enough to count as AI.

For decades, a lot of AI was built this way. Chess programs searched through millions of possible moves. Route finders worked out the shortest path across a city map. Early “expert systems” held thousands of if-then rules written by specialists. All of these are AI, and none of them learned from examples.

The weakness of rules shows up fast. Try writing rules to recognise your friend Priya’s face in a photo. Eyes this far apart? Hair this long? Priya gets a haircut and your rules break. Some problems are too messy for a rulebook.

Think of a kirana shop owner in Indore, Mr. Jain. Nobody gave him a rule for how many packets of Maggi to stock. After years of watching sales, he knows: more before exams, more in the monsoon, fewer in wedding season. He learned from data, his own sales history.

Machine learning gives computers that same ability. You show the system lots of past examples, like sales numbers next to the week and the weather, and it finds the pattern by itself. Then it can forecast next week. Banks use this to estimate loan risk. Email apps use it to learn what spam looks like from emails people marked as spam.

Deep learning is a special, powerful kind of machine learning. It uses a neural network: layers of tiny number-crunching units loosely inspired by brain cells. Each layer passes its results to the next. In a photo app, early layers notice edges, middle layers notice shapes like eyes, and later layers recognise whole faces.

Deep learning shines on messy data like photos, speech and long text, the stuff rules could never handle. That is why your phone can recognise your face to open, type what you say in Hindi or Tamil, and translate a signboard through the camera. It needs a lot of data and a lot of computing power, which is why it took off only in the 2010s.

In lesson 1 you met the mehendi artist, the AI that creates instead of sorting. Almost every generative AI tool today is a deep learning model, usually a very large one trained on huge amounts of text, images or audio. The chat apps you use are built on large language models, deep learning models trained to write text. Image tools use other deep learning designs that you will meet in module 6.

Not all deep learning is generative. The face lock on your phone is deep learning, but it only answers “is this you, yes or no?”. It never draws a new face. So GenAI is a part of deep learning, not the whole of it.