Revision notes

AI Fundamentals: Foundation

The key points, rules to remember, tables and diagrams from all 5 lessons, on one page for quick revision. To keep a copy, use your browser's Print and choose "Save as PDF".

Lesson 1

What is artificial intelligence? Meaning, narrow and general AI, and a short history

Rule-following program vs learning AI
PointOrdinary programAI system (learning type)
Who decides the rulesThe programmer writes every ruleThe system finds patterns in example data
New, unplanned situationsFails or gives an errorMakes a best guess from learned patterns
Gets better with more dataNoUsually yes, after retraining
Everyday exampleCalculator, ATM PIN checkSpam filter, face unlock, song suggestions
TypeAlso calledWhat it can doDoes it exist?
Narrow AIWeak AIOne task or a small set of related tasks, often very wellYes - the AI systems in use are classed as narrow AI
General AIStrong AI, AGI (Artificial General Intelligence)Any intellectual task a human can do, and switch between themNo - it remains a research goal

Narrow AI vs general AI

  • Narrow AI (weak AI)
    • Does one task or a small set of related tasks, often very well
    • Excellent inside its own job, helpless outside it
    • Exists: the AI systems in use are narrow AI
  • General AI (strong AI, AGI)
    • Would do any intellectual task a human can
    • Could switch between tasks
    • Does not exist: it remains a research goal

Three milestones

  1. 1950 - Turing Proposes the imitation game, now called the Turing test.
  2. 1956 - Dartmouth John McCarthy coins the term artificial intelligence.
  3. 1997 - Deep Blue Beats world chess champion Garry Kasparov.
SystemAI or not?Reason
A calculator working out 245 + 380 = 625Not AIFixed arithmetic rule, no judgement or learning
A traffic signal that changes every 60 seconds by timerNot AIFixed timer, ignores real traffic
A maps app predicting 6 pm travel time from Delhi to GurugramAIPredicts from patterns in past and live traffic data
A music app suggesting songs you have never playedAILearns your taste from listening patterns

AI = machines doing tasks that normally need human intelligence. Narrow (weak) AI does one job well; general (strong) AI would do any intellectual job a human can - and has not been achieved. 1950 Turing (the test) → 1956 McCarthy at Dartmouth (the name) → 1997 Deep Blue (chess).

Key takeaways: AI performs tasks that need intelligence, such as recognising, predicting and deciding. Ordinary programs follow fixed rules; learning AI finds patterns in data. Every AI in use is narrow AI; general AI (AGI) is still a goal. Remember 1950 (Turing), 1956 (McCarthy, Dartmouth) and 1997 (Deep Blue).

Read the full lesson

Lesson 2

AI versus machine learning: rules-based systems, learning systems and deep learning

TermPlain meaningExam-preparation comparison
Data / datasetThe examples the system learns from - numbers, text, images, soundsPrevious papers and practice questions
AlgorithmThe step-by-step method used to learn from the dataYour study method
ModelWhat the algorithm produces after learning: the stored patterns used to make predictionsThe knowledge in your head after studying
TrainingThe stage in which the model learns patterns from dataThe months of preparation
InferenceUsing the trained model on new input to produce an answerWriting the exam on new questions

How machine learning works

  1. Data + algorithm Examples, and the method used to learn from them.
  2. Training Needs much data and computing power.
  3. Model The stored patterns used to make predictions.
  4. Inference The model is used on new input to produce an answer.
Rules-based vs learning systems
FeatureRules-based systemLearning system
Source of knowledgeRules written by peoplePatterns learned from data
Needs large data?NoYes
Easy to explain its decision?Yes - you can read the ruleOften hard - patterns are spread across the model
Handles messy, varied inputPoorlyWell, if trained on similar examples
ExampleExpert system, a fixed rule 'block payments above a set limit'Face recognition, speech recognition, recommendations

Three ways machines learn

  • Supervised learning From examples labelled with the right answer.
    • Emails marked spam or not spam
  • Unsupervised learning From unlabelled data, finding groups on its own.
    • Grouping shoppers with similar buying habits
  • Reinforcement learning By trial and error, with rewards and penalties.
    • A program learning a game by playing it many times
TypeHow it learnsExample
Supervised learningFrom examples labelled with the right answerEmails marked spam or not spam
Unsupervised learningFrom unlabelled data, finding groups on its ownGrouping shoppers with similar buying habits
Reinforcement learningBy trial and error, with rewards and penaltiesA program learning a game by playing it many times

AI ⊃ machine learning ⊃ deep learning. Rules-based AI: people write the rules. Machine learning: data + algorithm → training → model; model + new input → inference → answer.

Key takeaways: AI is the whole field, ML is AI that learns from data, and DL is ML using many-layered neural networks. Expert systems are rules-based AI: easy to explain but brittle. Learning systems handle messy input but need data and are harder to explain. Training builds the model; inference uses it.

Read the full lesson

Lesson 3

Neural network basics: neurons, weights, layers and how training works

The three kinds of layer
LayerWhat it doesIn a network that recognises animals in photos
Input layerReceives the raw dataOne input for each pixel value of the photo
Hidden layersCombine the signals step by step to find patternsEarly layers pick out edges, later ones shapes, then parts such as ears and eyes
Output layerGives the answerA score for each label, such as 'cat' or 'dog'

Layers of a neural network

  1. Input layer Receives the raw data.
  2. Hidden layers Combine the signals step by step to find patterns; many make a network deep.
  3. Output layer Gives the answer.

How training adjusts the weights

  1. Show a labelled example Such as a photo marked 'cat'.
  2. Network gives its answer
  3. Measure the error How far the answer is from the correct label.
  4. Nudge the weights A little, to reduce the error (backpropagation).

Then back to step 1: Show a labelled example

Neuron: multiply each input by its weight, add up, compare with a threshold. Training: compare the answer with the label, nudge the weights to reduce the error, repeat. Deep = many hidden layers.

Key takeaways: a neuron multiplies inputs by weights, adds them and applies an activation rule. Networks have input, hidden and output layers; 'deep' means many hidden layers. Training adjusts the weights to reduce error. Large data, GPUs and better methods made deep learning work.

Read the full lesson

Lesson 4

Real-world applications of AI: vision, language, speech, recommendations, fraud, health and farming

Kinds of AI use

  • Seeing Computer vision
    • Face unlock
    • Number-plate reading
  • Language Natural language processing (NLP)
    • Chatbots
    • Translation
  • Speech Speech recognition and synthesis
    • Voice typing
    • Voice assistants
  • Recommendations Machine learning on past choices
    • 'You may also like'
  • Fraud detection Machine learning on usual patterns
    • Flagging an unusual payment
  • Health screening Mostly computer vision
    • Flagging X-rays or eye scans for a doctor
  • Farming Computer vision and machine learning
    • Spotting pests or disease in leaf photos
Kinds of AI use
Kind of useBranch of AIEveryday examples
SeeingComputer visionFace unlock, number-plate reading, reading printed text from photos (OCR), spotting faulty items on a factory line
LanguageNatural language processing (NLP)Chatbots answering routine questions, summarising long documents, grammar suggestions, translation
SpeechSpeech recognition and speech synthesisVoice typing, voice assistants, reading text aloud for people with low vision
RecommendationsMachine learning on past choices'You may also like' in shopping, music, video and news apps
Fraud detectionMachine learning on usual patternsA bank flagging a card or UPI payment unusual for the account
Health screeningMostly computer visionFlagging X-rays or eye scans that may show signs of disease, for a doctor to review
FarmingComputer vision and machine learningSpotting pests or disease in leaf photos, advice from weather and soil data

AI in health screening

  1. AI looks through images Many medical images.
  2. AI marks possible signs Images that may show signs of disease.
  3. Doctor looks at them first
  4. Doctor decides Examines the patient, checks the image and decides.
SituationKind of AIClue
A parking gate opens after reading a car's number plateComputer visionInformation taken from an image
A shopping app shows 'customers who bought this also bought'RecommendationPrediction from past choices
A bank sends an alert to confirm an unusually large night-time transferFraud detectionA break from the account's usual pattern
A help-desk chatbot answers a question typed in HindiNLPUnderstanding and producing language

Images → computer vision; language → NLP; voice → speech recognition and synthesis; 'you may also like' → recommendation; unusual payment → fraud detection. AI flags and suggests; a person confirms.

Key takeaways: most AI in daily life is one of a few kinds - vision, language, speech, recommendations and fraud detection - and real apps often combine several. In health and farming AI screens and suggests, while the doctor or the farmer with expert advice decides. Translation tools open services to speakers of every Indian language but need checking for important text.

Read the full lesson

Lesson 5

AI limitations and ethics: errors, hallucination, bias, privacy, deepfakes and human oversight

How AI goes wrong

  • False positive Says 'yes' when the truth is 'no'.
    • A genuine payment blocked as fraud
  • False negative Says 'no' when the truth is 'yes'.
    • A real fraud let through
  • Hallucination Fluent, confident, but false or made-up information.
    • An invented book title, quotation, date or law section
Risks and safeguards
RiskWhat it meansExampleSafeguard
BiasUnfair results because the training data were unbalanced or reflected past unfairnessA face system trained mostly on one group of faces works worse for othersBalanced, representative data; testing results across groups; human review
PrivacyPersonal data collected, stored or exposed without proper carePasting a customer's details into an online AI tool, which may store what is typedNever paste passwords, OTPs, Aadhaar numbers, bank details or confidential office files; remove names and numbers first
DeepfakesAI-made or altered video, audio or images showing people saying or doing what they never didA cloned voice of a relative urgently asking for moneyPause; verify through another channel, such as calling back on a known number; do not forward
Lack of explainabilityMany learning models are a 'black box': even their makers cannot fully say why an answer came outA loan rejected with no reason anyone can giveUse explainable methods for decisions about people; let people ask for a human review

Checking a suspicious video

  1. Do not forward it
  2. Check the official source The official website or notice of the body that conducts the exam.
  3. Not there? Unverified Treat it as unverified, however real it looks.

Use AI as an assistant, not an authority. Fluent is not the same as true: check every fact that matters against the official source, never paste personal or confidential data, and remember that a person stays responsible.

Key takeaways: AI makes errors - false positives, false negatives and hallucinations. Bias comes from data, privacy needs care with what you type, deepfakes need checking at the source, and black-box decisions about people need explanation. A person checks the output and takes responsibility for the decision.

Read the full lesson

Privacy · Terms · Refunds · Delivery