AI, machine learning, deep learning and generative AI: the map
AI, ML, deep learning, generative AI and LLMs are not five names for the same thing. Learn how they fit inside each other, with everyday Indian examples, so you can explain the difference clearly in class, at work and in interviews.
People use "AI" for everything from a phone's face unlock to ChatGPT. That makes the field sound more confusing than it is. The terms you hear most often fit inside each other like nested boxes, or like the places in an address: India contains Karnataka, Karnataka contains Bengaluru, Bengaluru contains Indiranagar.
Once you know which box a tool belongs to, you know roughly how it was built, what it is good at and how it can fail.
The nested boxes
From the biggest box to the smallest
| Term | Plain meaning | Everyday example |
|---|---|---|
| Artificial intelligence (AI) | Any computer system doing a task we would call intelligent if a person did it | A chess program, a route planner, ChatGPT |
| Machine learning (ML) | AI that learns patterns from examples instead of following hand-written rules | A bank's model that flags unusual UPI transactions |
| Deep learning (DL) | ML using large neural networks with many layers | Face unlock, Google Lens, speech-to-text |
| Generative AI (GenAI) | Deep learning models that create new content: text, images, audio, code, video | ChatGPT writing an email, an app making a poster |
| Large language model (LLM) | A generative model trained on huge amounts of text to predict and produce language | The models behind ChatGPT, Gemini and Claude |
Read the table top to bottom: every LLM is generative AI, all of today's generative AI is deep learning, all deep learning is machine learning, and all machine learning is AI. The reverse is not true. A spam filter is machine learning but not generative AI. A chess engine from the 1990s is AI but not machine learning.
Rules versus learning
The biggest dividing line is between AI that follows rules people wrote and AI that learns from data.
Imagine you want to sort Brew & Bean's customer reviews into "happy" and "unhappy". A rule-based approach: a programmer writes rules such as "if the review contains 'cold' or 'late', mark it unhappy". It is easy to understand but breaks quickly: "the cold brew was perfect" gets marked unhappy.
A machine learning approach: you collect a few thousand reviews already labelled happy or unhappy, and an algorithm finds the patterns itself. Nobody writes the rules. The model is only as good as the examples it learned from.
Three main ways machines learn
| Type | What the data looks like | Example |
|---|---|---|
| Supervised learning | Examples with the right answer attached | Emails labelled spam / not spam; house details with prices |
| Unsupervised learning | Examples without answers; the model finds groups | Grouping shoppers by buying habits |
| Reinforcement learning | Actions that earn rewards or penalties | A game-playing agent; fine-tuning chatbots with human ratings |
Want the maths later?
This track stays non-technical. If you want to train models yourself, the Data Science & Machine Learning track covers this properly, starting with *How machine learning learns: supervised vs unsupervised* and *Neural networks explained*.
Where deep learning and generative AI fit
A neural network is a large mathematical function with millions or billions of adjustable numbers called parameters (or weights). Training nudges those numbers again and again until the network's outputs match the examples. "Deep" just means many layers of these calculations stacked together.
Deep learning took off in the 2010s, when large datasets and powerful graphics chips (GPUs) made big networks practical. It is why speech recognition, translation and photo search suddenly got good.
Generative AI is deep learning pointed at creating rather than classifying. A classifier answers "is this photo a cat?". A generative model answers "draw me a cat wearing a cricket helmet". LLMs are the text branch of generative AI; image, music and video generators are other branches.
Predictive AI vs generative AI
| Question | Predictive (classic ML) | Generative AI |
|---|---|---|
| Output | A label, number or score | New text, image, audio, code |
| Example | "This transaction is 92% likely to be fraud" | "Here is a polite reply to the customer" |
| Typical risk | Wrong label for unusual cases | Fluent content that is false (hallucination) |
| Who builds it | Often trained by the company for one task | Usually a large general model, used through an app or API |
Common misconceptions
Five things people get wrong
1. "AI means ChatGPT." Chatbots are one small, recent corner of AI. Recommendation systems, fraud detection and maps ETAs are AI too.
2. "Machine learning always means neural networks." Many ML systems in companies use simpler methods such as decision trees or linear regression, because they are cheaper and easier to explain.
3. "The AI understands like a person." Today's models find statistical patterns. They can be very useful without having beliefs, feelings or intentions.
4. "AI learns from me while we chat." A model's parameters are fixed after training. Some apps save your chats and may use them to train future versions, which is a privacy question, not live learning.
5. "More AI is always better." If a simple rule solves the problem reliably ("block orders above ₹50,000 without OTP"), a rule is cheaper, faster and easier to audit.
How this comes up in interviews
"What is the difference between AI, ML and deep learning?" is a common warm-up question for freshers, including in non-AI roles. A strong answer is short, uses the nested-boxes idea and gives one example per level. Avoid reciting textbook definitions you can't explain.
A follow-up you may get: "Where would you not use machine learning?" Good answers mention small or no data, a need for exact and explainable decisions, or a problem simple rules already solve.
I am preparing for a [fresher / internship / role name] interview. Ask me: "What is the difference between AI, machine learning, deep learning and generative AI?"
Wait for my answer. Then:
1. Tell me what was correct and what was missing or wrong.
2. Rate it out of 10 for clarity.
3. Show a model answer under 120 words with one example per term.
Then ask me one follow-up question an interviewer might ask.What to expect: Type your own answer first; reading the model answer before you try teaches you much less. Check that the model answer keeps the nesting correct (LLM inside GenAI inside DL inside ML inside AI).
Try it: Sort these into the boxes
For each item, write the smallest box it belongs to (AI, ML, DL, GenAI or LLM) and one line explaining why.
1. A Sudoku solver that tries every possibility using fixed rules. 2. Gmail moving a message into the spam folder. 3. Your phone recognising your face to unlock. 4. Gemini writing a leave application. 5. An app that turns your selfie into a cartoon. 6. A food-delivery app estimating your delivery time from past orders.
Key takeaways
- AI contains ML, ML contains deep learning, deep learning contains generative AI, and LLMs are the text branch of generative AI.
- Rule-based AI follows instructions people wrote; machine learning finds patterns in examples.
- Predictive AI outputs labels and numbers; generative AI creates new content and can produce fluent falsehoods.
- Most AI in daily life (fraud checks, recommendations, maps) is not a chatbot.
- In interviews, explain the nesting simply with one example per level.
Check yourself
Finishing the quiz marks the lesson as done.
1. Which statement is true?
2. A bank's system scores each UPI payment from 0 to 1 for fraud risk, based on millions of past transactions. Which description fits best?
3. What does "deep" in deep learning refer to?
4. You need to reject any order over ₹50,000 that has no OTP. What is the most sensible approach?
5. Which risk is most typical of generative AI rather than classic predictive ML?
Saved on this device. Create a free account to keep it on every device.
Practise this with Sam
A real mock interview, out loud, with an honest scorecard. Free to start.