Introduction to
Artificial Intelligence
From Turing's dream to the ChatGPT era — understand what AI actually is, how it differs from ML & Deep Learning, and where it's reshaping the world.
↓ scroll to begin
From Turing's dream to the ChatGPT era — understand what AI actually is, how it differs from ML & Deep Learning, and where it's reshaping the world.
A clear, grounded definition — stripped of Hollywood myths and marketing buzzwords.
Before any algorithm names, hold one idea: intelligence in nature is not a single switch—it is a bundle of capacities (perceiving, remembering, reasoning, planning, communicating). Artificial intelligence is the attempt to realise some of those capacities with machines. That is why “AI” appears in philosophy (what is thinking?), mathematics (what can be computed?), cognitive science (how do humans do it?), and engineering (what can we ship this quarter?). Your course sits in the engineering corner, but the vocabulary comes from all of them.
What “learning” means in this course
In everyday speech, learning implies understanding. In AI, learning often means: adjusting internal parameters from data so measured behaviour improves. A spam filter “learns” in that narrow sense even if it has no self-awareness. Later modules split supervised (examples with answers), unsupervised (find hidden structure), and reinforcement (trial and error with rewards)—three different mathematical meanings of “learn.”
Figure 1 — AI is an interdisciplinary target: algorithms, data, hardware, and (sometimes) biological inspiration overlap. No single department owns the whole circle.
Artificial Intelligence (AI) is a branch of computer science focused on building systems that can perform tasks which normally require human-level intelligence — things like understanding language, recognising images, making decisions, and solving problems.
The word "artificial" just means made by humans (not natural), and "intelligence" means the ability to learn, reason, and adapt. Together: machines that can think and learn.
Think of AI like a very smart intern. At first, you show them examples of how to do the job. They observe patterns, ask questions, and gradually get better at tasks — even handling situations they haven't seen before.
Just like that intern, an AI system:
Understanding and generating human language — text, speech, translation, summarisation. Powers ChatGPT, Google Translate, Siri.
Basic Example: Spam FilterInterpreting images and video — object detection, face recognition, medical imaging. A camera's "intelligence".
Basic Example: Face UnlockEvaluating options and choosing the best action given goals and constraints. Used in chess engines, recommendation systems, autopilot.
Example: Netflix RecommendationsBreaking complex challenges into steps and finding optimal solutions — scheduling, logistics, drug discovery.
Example: Google Maps RoutesFinding patterns in data to forecast future outcomes — stock prices, weather, disease diagnosis, customer churn.
Advanced: AlphaFold Protein FoldingSensing and responding to physical environments — robotics, self-driving cars, smart assistants.
Advanced: Tesla AutopilotAI isn't new — it has a 70-year story of booms, busts, and breakthroughs.
Historically, AI research alternates between optimism (“true thinking machines are near!”) and sober reassessment (“we need more data, compute, and better theory”). Periods of reduced funding are often called AI winters. They are not failures of science—they are corrections when promises outpace evidence. Understanding this cycle helps you read news headlines critically: breakthroughs are real, but deployment at scale usually lags demos by years.
Figure 2 — Stylised eras (not to scale): symbolic methods dominated early decades; statistical learning rose with data; depth + compute produced today’s applications.
Alan Turing published "Computing Machinery and Intelligence" asking "Can machines think?" He proposed the Turing Test: if a machine can converse indistinguishably from a human, it can be called intelligent. This was the philosophical spark that ignited the field.
John McCarthy, Marvin Minsky, and others held the first AI workshop at Dartmouth College. The term "Artificial Intelligence" was coined here. Optimism was sky-high — they believed human-level AI was just 20 years away.
First AI programs could solve algebra, prove theorems, and speak English. ELIZA (1966) was the first chatbot. Robots like Shakey (1969) could navigate rooms. Huge government funding poured in.
Progress stalled. Computers lacked the power to deliver on bold promises. Funding dried up. This pattern of "boom and winter" would repeat — a key lesson in managing AI expectations.
Rule-based "Expert Systems" encoded human knowledge as if-then rules. MYCIN (medical diagnosis) and DENDRAL (chemistry) showed real-world value. Japan's Fifth Generation Computer project invested billions. Then... another winter in the late 80s.
IBM's Deep Blue defeated world chess champion Garry Kasparov. A milestone moment that showed AI could surpass human performance in specific, well-defined tasks — even if it didn't "understand" chess like a human does.
A neural network called AlexNet dramatically outperformed all competitors in the ImageNet image recognition challenge. This sparked the modern Deep Learning era. Suddenly, AI could recognise cats, dogs, and faces better than humans in some tests.
DeepMind's AlphaGo beat world Go champion Lee Sedol (2016) — considered impossible due to Go's complexity. OpenAI released GPT models (2018–2020), revolutionising language AI. Self-driving cars entered public roads.
ChatGPT reached 100 million users in just 2 months (fastest product ever). Generative AI can now write code, create art, compose music, and generate video. Models like GPT-4, Claude, and Gemini are transforming every industry. We are living through the most rapid AI adoption in history.
These three terms are often confused — here's the definitive breakdown.
Every DL is ML. Every ML is AI. But not all AI is ML.
Definition: The broad field of computer science dedicated to creating systems capable of performing tasks that typically require human intelligence — reasoning, planning, learning, perception, and language.
Analogy: AI is the whole toolbox 🧰. Machine Learning and Deep Learning are specific tools inside it. Not every AI system learns — some just follow strict rules written by programmers.
Key distinction: AI includes rule-based systems (no learning) AND learning-based systems (ML/DL). A chess engine that uses hand-coded rules is AI. ChatGPT that learned from billions of texts is also AI — but a very different kind.
Definition: A subset of AI where systems learn patterns from data and improve their performance without being explicitly programmed for every rule. Instead of writing instructions, you show examples.
Analogy: Teaching a child to identify dogs 🐕. You don't give them a manual listing every dog feature. You show them 1,000 photos — "this is a dog, this is not" — and they learn the pattern themselves. ML works the same way.
How it works: Feed data → algorithm finds patterns → model makes predictions. The model improves as it sees more data. Examples include spam filters, Netflix recommendations, fraud detection, and weather forecasting.
Definition: A subset of ML using multi-layered artificial neural networks inspired by the human brain. "Deep" refers to the many layers of processing. DL excels at unstructured data — images, audio, text, video.
Analogy: Traditional ML is like a student learning from labelled flashcards (structured data). Deep Learning is like a student learning by immersion — watching thousands of hours of French TV and gradually figuring out grammar, vocabulary, and context on their own.
Key difference from ML: In traditional ML, humans must manually engineer features (tell the algorithm what to look for). In DL, the network automatically discovers the relevant features from raw data — it's self-supervised feature extraction.
| Attribute | AI (Broad) | Machine Learning | Deep Learning |
|---|---|---|---|
| Definition | Machines simulating human intelligence | Systems learning from data | ML via multi-layer neural networks |
| Relationship | Parent | Subset of AI | Subset of ML |
| Data Needed | Varies — can be rule-based (no data) | Moderate (hundreds to thousands) | Large (millions of data points) |
| Feature Engineering | Manual (for rule-based) | Manual by humans | Automatic (learned by network) |
| Interpretability | High (rule-based) | Moderate | Low ("black box") |
| Best For | Reasoning, planning, rules | Structured tabular data | Images, text, audio, video |
| Compute Required | Low–Medium | Medium | High (GPUs/TPUs needed) |
| Real Example | Chess Engine, Expert System | Spam Filter, Churn Prediction | ChatGPT, FaceID, AlphaFold |
AI is classified in two main ways: by capability level, and by how it learns.
Capability asks: how wide is the competence? A chess program is hyper-focused (narrow). A hypothetical system that could learn any intellectual task the way a human does would be general. Superintelligence is a speculative stage beyond human capability—useful in safety thought experiments, not in product roadmaps today.
Figure 3 — Narrow systems live inside the solid box; AGI remains an outer, still-hypothetical layer. “Strong AI” is a synonym in some texts—always ask what the author means.
Designed for ONE specific task. Cannot generalise beyond its training. This is ALL current AI — it's extremely good at its job but completely helpless outside it.
📌 Status: Exists today
Hypothetical AI with human-level intelligence across all domains — reasoning, creativity, emotional understanding, learning new tasks from scratch. The holy grail.
📌 Status: Does NOT exist yet
Hypothetical AI that surpasses human intelligence in every domain — science, creativity, social skills, and more. A concept studied in AI safety research.
📌 Status: Theoretical / Future
Trains on labelled data — each example has the correct answer. The model learns the mapping from input to output.
Trains on unlabelled data. The model discovers hidden patterns and structures on its own — no answers provided.
Agent learns by interacting with an environment, receiving rewards for good actions and penalties for bad ones. Trial and error at scale.
First, a schematic of where AI sits in industrial value chains; then expand each sector for concrete deployments.
Figure — Theory recap: AI components consume information and emit actions or recommendations; industry cards spell out domain specifics.
10 questions covering all Module 1 topics. Instant feedback on every answer.
Everything from Module 1 distilled into 8 core ideas you should carry forward.