The AI Landscape

Welcome to the course! Before writing any code, let us get a clear picture of the whole field. You will hear words like AI, ML, deep learning, NLP, GenAI and agents every day. In this first session, we will see what each one means, how they fit together, and where they are used in real companies.

01AI, Machine Learning and Deep Learning

Artificial Intelligence (AI) is the big umbrella: any computer system that does a task we normally think needs human intelligence, such as understanding speech, recognising faces or making decisions. The term was coined in 1956.

Machine Learning (ML) is a way of building AI where we do not write the rules ourselves. We show the computer many examples, and it learns the rules. Deep Learning (DL) is machine learning with large neural networks of many layers. Generative AI is deep learning that creates new content.

Here is the difference between writing rules and learning them, on a tiny fraud example. The rule a human wrote says “fraud if amount > ₹40,000 and new device”. The ML model learned from 12 past transactions instead:

from sklearn.linear_model import LogisticRegression

# [amount in ₹ thousands, hour of day, new device? (1 = yes)]  →  1 = fraud
X = [[2, 14, 0], [1, 11, 0], [45, 2, 1], [3, 19, 0], [60, 3, 1], [5, 10, 0],
     [38, 1, 1], [4, 16, 0], [52, 4, 0], [2, 9, 1], [48, 23, 1], [6, 13, 0]]
y = [0, 0, 1, 0, 1, 0, 1, 0, 1, 0, 1, 0]

# 1. Traditional programming: a human writes the rule
def rule(amount, hour, new_device):
    return int(amount > 40 and new_device == 1)

# 2. Machine learning: the computer learns the rule from examples
model = LogisticRegression().fit(X, y)

new = [[50, 3, 0], [3, 2, 1]]
for t in new:
    print(t, "rule:", rule(*t), "| ML:", model.predict([t])[0],
          f"(fraud probability {model.predict_proba([t])[0][1]:.2f})")

The ₹50,000 payment at 3 a.m. from a known device slips past the human rule, but the model flags it, because in the data large night-time payments were fraud even without a new device. That is the core idea of ML: patterns come from data. (Toy data: real fraud models train on millions of transactions.)

02Natural Language Processing (NLP)

NLP is the part of AI that works with human language: reading, understanding and writing text or speech. Language is hard for computers because the same word can mean different things (“bank” of a river vs a bank account), and people write with typos, slang and mixed languages like Hinglish.

03Generative AI

Most older AI predicts a label: fraud or not fraud, positive or negative. Generative AI creates something new: a paragraph, an image, a piece of code, a song. Large Language Models (LLMs) like GPT, Claude, Gemini and Llama are generative models for text.

At its heart, an LLM does one simple thing again and again: predict the next word (token). Repeat that thousands of times and you get essays, code and conversations. You will see exactly how in Modules 5–7.

04Agentic AI

A chatbot talks. An AI agent acts. An agent is an LLM placed in a loop where it can decide the next step, use tools (search, databases, APIs, a calculator), look at the result and keep going until the job is done.

05Real-world applications

Here is how NLP, generative AI and agentic AI show up in six industries. Pick an industry:

06Summary

• AI ⊃ ML ⊃ Deep Learning ⊃ Generative AI: each sits inside the one before.
• ML learns rules from data instead of a programmer writing them.
• NLP lets computers read and write human language.
• Generative AI creates new content; LLMs do it by predicting the next token.
• Agentic AI plans and takes actions with tools in a loop.

07Quick quiz

08References

• Stanford HAI: AI Index Report
• Google: Machine Learning Crash Course
• Anthropic: Building effective agents