Foundation of AI › Unit 1

Unit 1 of Foundation of AI covers six things in eight hours: what intelligence is and how natural intelligence differs from artificial intelligence, a working definition of AI, the history and evolution of AI from early computers to today, the three types of AI (narrow, general and super), where AI already appears in everyday life, and the three components every AI system needs, which are data, algorithms and computing power.

This is the vocabulary unit. Every question in Units 2, 3 and 4 assumes you can use these terms precisely, and most students lose marks later for something they got slightly wrong here. So it is worth doing properly rather than quickly.

1. What is intelligence? Natural versus artificial

Before defining artificial intelligence, the syllabus asks a harder question: what is intelligence in the first place? The reason is that AI is defined by reference to human intelligence, so a vague idea of the second gives you a vague idea of the first.

Intelligence is the ability to acquire knowledge, understand and reason with it, adapt to new situations, and use what has been learned to solve problems and make decisions.

Note what is packed into that sentence. Intelligence is not one skill. It is at least five: learning, reasoning, problem solving, perception and language. Any system that does one of these well, and nothing else, is doing something intelligent without being generally intelligent. Hold on to that idea, because it is exactly what separates narrow AI from general AI later in this unit.

Basis Natural intelligence Artificial intelligence
Origin Biological. Develops in the brain through growth and experience. Engineered. Built by people using data, algorithms and hardware.
Learning Learns from very few examples. A child needs to see two or three dogs. Usually needs thousands or millions of examples for the same reliability.
Speed Slow at calculation, fast at judgement in unfamiliar situations. Extremely fast at calculation, poor at situations unlike its training data.
Energy The human brain runs on roughly the power of a dim light bulb. Training a large model consumes many orders of magnitude more energy.
Creativity and common sense Present. Can reason about a situation never encountered before. Limited. Recombines patterns in its data; has no common sense of its own.
Consistency Varies with fatigue, mood and bias. Perfectly consistent, which also means it repeats its mistakes perfectly.
Emotions and intent Has them. They shape decisions, for better and worse. Has none. It can recognise and imitate emotion without experiencing it.

Exam tip. If a question says “differentiate”, the answer is a table, not a paragraph. Six rows of difference with a clear basis in the first column scores far better than half a page of prose that makes the same points.

2. A simple definition of artificial intelligence

Artificial intelligence is the branch of computer science concerned with building systems that perform tasks which would normally require human intelligence, such as understanding language, recognising images, making decisions, and learning from experience. The system derives its behaviour from data rather than from instructions written out in advance for every case.

That last sentence is the part students leave out, and it is the part that actually distinguishes AI from any other software. A calculator performs a task a human mind can do, and does it faster, but nobody calls a calculator intelligent. The difference is where the behaviour comes from.

AI versus ordinary software

Ordinary software Artificial intelligence
A programmer writes every rule by hand. The system finds the rules itself, from examples.
Given the same input, always gives the same output. Output improves as it is given more data.
Fails completely on an input nobody anticipated. Produces its best guess on an input it has not seen.
A payroll program calculating tax from fixed slabs. A system that flags which salary claims look fraudulent.

3. History and evolution of AI

The syllabus phrases this as “from early computers to modern AI”. Examiners usually ask for a timeline with four or five milestones and one line on each. Click any point below to read it.

Interactive AI TimelineClick any milestone to read what happened and why it mattered

1950Turing asks the questionAlan Turing publishes Computing Machinery and Intelligence, opening with the question “Can machines think?” and proposing the imitation game now known as the Turing Test. No machine existed that could attempt it.
1956The field gets its nameAt the Dartmouth Conference, John McCarthy coins the term “artificial intelligence”. A small group proposes that every aspect of learning and intelligence can in principle be described precisely enough for a machine to simulate it. This is the founding moment of AI as a discipline.
1966The first chatbotJoseph Weizenbaum builds ELIZA, which imitates a psychotherapist by reflecting the user’s sentences back as questions. It understands nothing, yet users confide in it. Weizenbaum is disturbed by this, and the effect is still visible in how people talk to chatbots today.
1974The first AI winterProgress falls far short of the promises made. Funding in Britain and the United States is cut sharply. The lesson: the field had solved toy problems and assumed real ones were only larger.
1980Expert systems arriveRule-based expert systems encode a specialist’s knowledge as if-then rules and are sold commercially to companies. This is the high point of symbolic AI, and it is the architecture you study first in Unit 2.
1987The second AI winterExpert systems prove brittle and expensive to maintain, because every new situation needs a new rule written by hand. The market collapses. This failure is what pushes the field towards learning from data.
1997Deep Blue beats KasparovIBM’s Deep Blue defeats world chess champion Garry Kasparov. It wins by searching millions of positions per second rather than by understanding chess, which makes it a perfect illustration of narrow AI.
2012Deep learning breaks throughA deep neural network called AlexNet wins the ImageNet image recognition competition by a wide margin, using graphics processors for training. Vision research changes direction almost overnight and the modern era begins.
2016AlphaGo winsDeepMind’s AlphaGo defeats Lee Sedol at Go, a game long considered too intuitive for machines. It plays moves human masters call creative, while still being a narrow system that can do nothing else.
2017The TransformerA research paper introduces the Transformer architecture for handling language. It is the design underneath almost every modern language model, including the one you have probably used this week.
2022Generative AI reaches everyoneChatGPT puts a large language model in front of the general public. AI stops being an industry topic and becomes a household one, which is the reason courses like this one now exist.
2026AI as infrastructureAI is embedded in payments, healthcare, governance and education, and policy attention shifts from capability to accountability. India runs one of the world’s largest public AI capacity-building efforts, and AI enters the school curriculum.

The two gaps in that timeline have a name: the AI winters. Funding collapsed twice, in the mid 1970s and again around 1987, because the field had promised results it could not deliver. That pattern is worth remembering when you read today’s predictions, and it is a good closing line for a long answer on the history of AI.

The four phases, if you need a shorter answer

  1. Foundations, 1943 to 1956. The idea that a machine could think is formalised. Turing proposes a test for it; the term “artificial intelligence” is coined at Dartmouth.
  2. Symbolic AI and expert systems, 1956 to 1990. Intelligence is treated as rules and logic. Systems encode a specialist’s knowledge as if-then rules. Two funding winters interrupt the work.
  3. Machine learning, 1990 to 2011. The approach flips. Instead of writing rules, feed the machine data and let it find the patterns. Cheap storage and the internet make this possible.
  4. Deep learning and generative AI, 2012 to now. Neural networks with many layers, trained on enormous datasets, beat every previous method at vision and language, and eventually produce text and images of their own.

4. Types of AI: narrow, general and super

This is the single most frequently examined topic in Unit 1, and the one most often answered wrongly. Try the classifier before you read the table.

Which type of AI is it?Pick a real system and see how it classifies, and why

Choose any system above to see its classification.

Type Also called What it can do Status today Examples
Narrow AI Weak AI, ANI One task or a narrow set of related tasks, often better than a human The only type that exists Google Maps routing, UPI fraud detection, Alexa, ChatGPT, face unlock
General AI Strong AI, AGI Any intellectual task a human can do, transferring learning across domains Does not exist. Actively researched. Hypothetical only
Super AI ASI Surpasses the best human minds in every field including creativity and social skill Theoretical. A subject of debate, not engineering. Hypothetical only

The mistake that costs marks. ChatGPT is narrow AI, not general AI. It handles language tasks across many subjects, which makes it look general, but it cannot transfer that competence to an unrelated domain it was never trained on, it has no goals of its own, and it does not learn from your conversation once training is finished. Breadth of subject matter is not the same as generality of intelligence.

5. AI in everyday life

The syllabus names four examples. Here is each one with what the AI is actually doing underneath, because “it uses AI” scores nothing.

Google Maps

Maps predicts your arrival time by combining historical traffic patterns for that road at that hour with live location data from phones currently on it, then searching for the fastest route through a graph of roads. That search is the same category of problem as the maze example in Unit 2. In Indian cities it also learns local patterns that no global rule would capture, such as which flyovers jam at which times.

E-commerce recommendations

When Amazon or Flipkart shows “customers who bought this also bought”, the system is not reading your mind. It compares your behaviour with millions of other users and finds people whose past purchases resemble yours, then suggests what they bought next. This is called collaborative filtering, and it is why a single unusual purchase can distort your recommendations for weeks.

Chatbots

A bank or telecom chatbot converts your typed sentence into a machine-readable form, identifies your intent, pulls the relevant answer, and responds in language. Indian deployments increasingly do this across Hindi, Marathi, Tamil and English in the same conversation, which is a much harder problem than it appears.

Social media feeds

Instagram and YouTube rank what you see by predicting which posts you are most likely to engage with, using your watch time, likes, pauses and shares. This is the everyday example with the sharpest ethical edge, because a system optimising for engagement will learn that outrage holds attention. Unit 4 returns to this.

6. Key components of AI: data, algorithms and computing power

Every AI system rests on three things. Click any one to switch it off and see what breaks.

The three pillars, and what happens without eachClick a pillar to remove it
DataThe examples the system learns from
→
AlgorithmsThe method that finds patterns in the data
→
Computing powerThe hardware that runs it at scale

All three present. The system can learn, improve and run at scale. Click any pillar to remove it.

The three pillars also explain the timing of the current AI boom, which is a strong point to make in a long answer. The algorithms behind neural networks were largely understood by the 1980s. What arrived later was the data, from the internet and smartphones, and the computing power, from graphics processors. AI did not become possible because someone had a new idea. It became possible because two of its three ingredients finally became cheap.

Model exam answers

Define AI and explain its main characteristics (5 marks)

Artificial intelligence is the branch of computer science concerned with building systems that perform tasks normally requiring human intelligence, such as understanding language, recognising patterns, making decisions and learning from experience. Its distinguishing feature is that its behaviour is derived from data rather than written out rule by rule in advance.

Its main characteristics are: learning from data and improving with more of it; reasoning, in drawing conclusions from available facts; problem solving, in searching systematically for a solution; perception, in interpreting images, speech and text; and adaptability, in handling inputs it has not seen before. For example, Google Maps learns traffic patterns from millions of devices and predicts arrival times that no fixed rule could produce.

Explain the types of AI with examples (7 marks)

AI is classified by capability into three types.

Narrow AI, also called weak AI, is designed for one task or a narrow set of related tasks, and may exceed human performance at it while being incapable of anything else. All AI in existence today is narrow AI. Examples: UPI fraud detection systems, Google Maps route prediction, voice assistants, and large language models such as ChatGPT.

General AI, or AGI, would perform any intellectual task a human can, transferring what it learns in one domain to an unrelated one. It does not exist; it is an active research goal.

Super AI, or ASI, would surpass the best human minds across every field, including creativity and social intelligence. It is theoretical and currently a subject of philosophical and policy debate rather than engineering. A common error is to classify ChatGPT as general AI; it is narrow, because breadth of subject matter is not the same as generality of intelligence.

Practice questions

Unit 1 MCQ practice10 questions with explanations. Answers appear as you choose.

Common mistakes in Unit 1

  • Calling ChatGPT general AI. It is narrow. This single error appears in a large share of answer scripts.
  • Defining AI as “machines that think like humans”. Too loose, and it invites the follow-up question you cannot answer. Define it by task and by where the behaviour comes from.
  • Confusing AI, machine learning and deep learning. They are nested, not parallel. Every deep learning system is machine learning; every machine learning system is AI; the reverse is not true.
  • Listing history without dates. Four dated milestones beat a page of narrative.
  • Giving only foreign examples. An Indian example shows you understood the concept rather than copied it.

Frequently asked questions

Who is called the father of artificial intelligence?

John McCarthy, who coined the term “artificial intelligence” for the 1956 Dartmouth Conference and organised the meeting where the field was founded. Alan Turing is often described as the father of computer science and of the idea of machine intelligence, because of his 1950 paper proposing what became the Turing Test, but the term itself is McCarthy’s.

What is the Turing Test?

A test proposed by Alan Turing in 1950. A human judge holds a text conversation with a human and a machine without knowing which is which. If the judge cannot reliably tell them apart, the machine is said to have passed. It measures imitation of human conversation, not understanding, which is why modern researchers treat it as a historical landmark rather than a real benchmark.

Is narrow AI the same as weak AI?

Yes. Narrow AI, weak AI and artificial narrow intelligence (ANI) are three names for the same category: an AI system built for one task or a narrow set of related tasks. Exam papers use all three terms, sometimes in the same question.

What are the three components of an AI system?

Data, which the system learns from; algorithms, which are the methods that find patterns in that data; and computing power, which is the hardware that makes training and running the system practical. Remove any one and the system cannot work: without data there is nothing to learn from, without algorithms the data stays raw, and without computing power the method cannot run at a useful scale.

← Course hub: all four units
Unit 2: AI Structures, Architectures and Techniques →

Useful alongside this unit: the 120-term AI glossary for any word you meet here for the first time, and the AI literacy test once you have finished all four units.

Written by Piyush Wairale, M.Tech, IIT Madras, founder of PiyushAI Edtech. Last reviewed 22 September 2026.