Learning Outcomes
SLO | After this lecture you will be able to... | CLO | Bloom |
|---|---|---|---|
SLO 2.1 | Describe the foundations of AI and the main milestones of its history, including the AI winters | CLO1 | C1 Remember |
SLO 2.2 | Explain applications of AI in different sectors, with local examples | CLO1 | C2 Understand |
SLO 2.3 | Differentiate between AI and robotics, and list advantages of AI | CLO1 | C2 Understand |
SLO 2.4 | Classify AI systems by capability (narrow, general, super) and by functionality | CLO1 | C2 Understand |
SLO 2.5 | Explain the risks and ethical issues of AI and the principles of responsible AI | CLO1 | C2 Understand |
CLO1: Explain the applications of AI and agent-based approach to AI. (SLO 2.5 also supports PLO9 Ethics.)
1. Foundations and History of AI
§1.1 Subjects that built AI
Slide L02-S06
AI is not only a part of computer science. Ideas from many subjects (disciplines) came together to build it:
Subject | What it gave to AI | Example |
|---|---|---|
Philosophy | Logic and reasoning; can a mind follow rules? | Aristotle's syllogisms (Lecture 1) |
Mathematics | Formal logic, probability, what can be computed | Bayes' rule (Lecture 20) |
Economics | How to choose the best decision (utility = value or usefulness) | Choosing the cheapest bus + train route |
Neuroscience | How brain cells (neurons) work | Artificial neural networks (Lecture 26) |
Psychology | How people think, learn and remember | Cognitive modelling (Lecture 1) |
Computer engineering | Fast hardware to run AI | GPUs (graphics cards) |
Linguistics | Grammar and meaning of language | Natural language processing (Lecture 30) |
Control theory | Machines that correct themselves using feedback | An AC thermostat switching on and off |
§1.2 The birth of AI (1943 - 1966)
Slide L02-S07
Year | Milestone | Why it matters |
|---|---|---|
1943 | McCulloch and Pitts publish the first mathematical model of a brain neuron | Seed of neural networks |
1950 | Alan Turing's paper "Computing Machinery and Intelligence" asks "Can machines think?" | Proposes the Turing Test (Lecture 3) |
1956 | Dartmouth workshop; John McCarthy coins the term "Artificial Intelligence" | The birth of AI as a field |
1958 | McCarthy creates the LISP language; Rosenblatt builds the Perceptron | First AI language; first learning machine |
1966 | Weizenbaum's ELIZA chatbot imitates a counsellor (therapist) with simple patterns | First famous chatbot |
The mood was great hope (optimism). Some researchers said a machine as clever as a human would arrive within about 20 years. This promise was far too big.
§1.3 AI winters and expert systems (1969 - 1993)
Slide L02-S08
Definition: An AI winter is a period when people lost trust in AI and the money for AI research (funding) was cut, because AI had promised much more than it delivered. |
Period | What happened |
|---|---|
1969 | Minsky and Papert's book "Perceptrons" showed that a single-layer perceptron cannot learn simple problems such as XOR. Neural network research slowed down. |
1974 - 1980 | First AI winter. The UK Lighthill report (1973) said AI had failed its promises, and funding was cut in the UK and USA. |
1980s | Expert systems became a business. An expert system is a program of IF-THEN rules that copies a human expert. MYCIN diagnosed blood infections; XCON configured computer orders for the company DEC. |
1987 - 1993 | Second AI winter. Expert systems were expensive, hard to update and could not learn, so companies stopped buying them. |
1990s | Quiet return: probability and machine learning grew slowly in the background. |
§1.4 The modern AI boom (1997 - today)
Slide L02-S09
Year | Milestone |
|---|---|
1997 | IBM's Deep Blue beats world chess champion Garry Kasparov. |
2011 | IBM Watson wins the quiz show Jeopardy!; Apple launches Siri, the first popular voice assistant. |
2012 | AlexNet, a deep neural network trained on GPUs, wins the ImageNet image contest by a big margin. The deep learning era begins. |
2016 | DeepMind's AlphaGo beats Go champion Lee Sedol. |
2017 | Google researchers publish the Transformer, a new neural network design. It is the base of ChatGPT and similar tools. |
Nov 2022 | OpenAI releases ChatGPT. It reaches about 100 million users in two months. |
2023 - today | Generative AI spreads: GPT-4, Gemini, Claude, Meta AI, and AI agents that can use tools. |
§1.5 Why AI failed before and succeeds now
Slide L02-S10
- Big data: billions of digital photos, videos, posts and web pages to learn from. In 1970 there was almost none.
- Computing power: GPUs (graphics cards first made for games) do thousands of calculations at the same time.
- Better algorithms: deep learning (2012) and Transformers (2017) learn far better than older methods.
Analogy: a fast bowler needs fitness, a good ball and good technique. Missing any one, he fails. Early AI had good ideas but no data and no computing power. Lesson: over-promising led to AI winters, so we should judge AI claims carefully even today.
2. Applications of AI
§2.1 Applications by sector
Slide L02-S12
Sector (field) | Application | Local / everyday example |
|---|---|---|
Healthcare | Reading medical images; predicting disease outbreaks | AI marks possible TB signs on chest X-rays; dengue outbreak prediction from weather and case data |
Agriculture | Crop disease detection; irrigation advice | A phone app identifies cotton or wheat disease from a leaf photo |
Banking | Fraud detection; customer chatbots | Fraud alerts and chat support in banks and wallets such as JazzCash and Easypaisa |
Transport | Driver matching, fares, traffic prediction | Careem / inDrive matching; Google Maps travel time |
Government | Identity checks; traffic control | NADRA biometric (fingerprint and face) verification; Safe City cameras reading number plates for e-challans |
Education | Tutoring, grading, plagiarism checks | AI tutors such as ChatGPT and Gemini; plagiarism software |
§2.2 Three applications up close
Slide L02-S13
Every application follows the agent pattern from Lecture 1: INPUT (percepts) → AI → OUTPUT (action).
Application | Input (percepts) | What the AI does | Output (action) |
|---|---|---|---|
Chatbot | Typed or spoken message | Understands the intent (what the user wants) using NLP and finds an answer | A reply, or a task done (e.g. balance shown) |
Self-driving car | Cameras, radar, LiDAR (a laser distance sensor) | Detects lanes, cars and people; plans a safe path | Steer, brake, speed up - with little or no human input |
Face recognition | Camera photo | Finds the face, measures features (distance between eyes, nose shape) and compares with stored faces | Match / no match (e.g. phone unlocks) |
§2.3 AI vs robotics
Slide L02-S14
Feature | Artificial Intelligence | Robotics |
|---|---|---|
What it is | Software that thinks and decides | Machines that act in the physical world |
Works in | Mostly a computer (digital world) | The real, physical world |
Input | Data: text, images, numbers | Sensors: camera, touch, sound |
Needs | A general-purpose computer | Special hardware: sensors and effectors (motors, arms, wheels) |
Example of only one | ChatGPT, a spam filter | A factory welding arm repeating one fixed movement |
They meet in intelligent robots: a robot vacuum, a self-driving car, or a drone that sprays only the sick plants in a field. AI is the brain; robotics is the body. Not every robot is intelligent.
§2.4 Advantages of AI
Slide L02-S15
Advantage | Example |
|---|---|
Fewer human errors | Weather forecasts made from large data are more accurate |
Doing risky jobs | Robots defuse bombs, work in mines and explore the deep sea or Mars |
Available 24 × 7 | A bank chatbot answers at 2 a.m. without getting tired |
Faster decisions | Fraud checks in milliseconds; a chess engine finds the best move in seconds |
Repetitive work | Reading thousands of forms, sorting e-mails |
Expert help where experts are few | A farmer far from a crop expert gets disease advice on his phone |
3. Types of AI
§3.1 By capability (how much it can do)
Slide L02-S16
Type | Also called | Meaning | Example | Exists? |
|---|---|---|---|---|
Narrow AI | Weak AI, ANI | Does one task or a small set of tasks well; cannot go beyond them | Face unlock, Google Maps, chess engine, spam filter, ChatGPT | Yes - all AI today |
General AI | Strong AI, AGI | Could learn and do any intellectual task a human can | None yet | No - research goal |
Super AI | ASI | Better than the best humans in every field | Only in theory and films | No - an idea only |
Important: ChatGPT is very broad but it is still Narrow AI. It makes mistakes humans would not make, and it cannot learn any new task by itself the way a person can. |
§3.2 By functionality (how it uses memory and understanding)
Slide L02-S17
# | Type | Meaning | Example | Status |
|---|---|---|---|---|
1 | Reactive machines | React to the present situation only; no memory of the past | IBM Deep Blue (1997) | Exists |
2 | Limited memory | Use recent past data for a short time | Self-driving car tracking nearby cars; a chatbot remembering this conversation | Most AI today |
3 | Theory of mind | Would understand people's emotions, beliefs and intentions | - | Research only |
4 | Self-aware | Would have its own consciousness (awareness of itself) | - | Imagined only |
Types 1 and 2 are Narrow AI. Types 3 and 4 would need General or Super AI.
§3.3 Worked example: classifying systems
Slide L02-S18
System | Capability | Functionality | Reason |
|---|---|---|---|
Tic-tac-toe program that looks only at the current board | Narrow | Reactive | One task; no memory |
Careem arrival time from live traffic | Narrow | Limited memory | Uses recent traffic data |
Robot teacher that truly understands a sad student's feelings | General (does not exist) | Theory of mind | Needs real understanding of emotions |
YouTube suggestions from this week's history | Narrow | Limited memory | Uses recent watch history |
4. Risks, Ethics and Responsible AI
§4.1 Disadvantages and controversies
Slide L02-S20
Problem | Meaning / example |
|---|---|
High cost | Large AI models need huge computers, electricity and skilled people. |
Job changes | Routine jobs (data entry, call centres) shrink; new jobs (AI engineers, data labellers) appear. |
Over-dependence | Copying AI answers stops students from thinking and learning. |
No real emotions | AI has no feelings and limited true creativity; it mixes patterns it has seen. |
Hallucination | Chatbots can state false facts confidently. Always check the source. |
Energy use | Data centres use a lot of electricity and water for cooling. |
§4.2 Ethical issues
Slide L02-S21
Ethics means the principles of right and wrong. Computing graduates must consider them (PLO9).
Issue | Meaning | Example |
|---|---|---|
Bias | AI repeats unfair patterns hidden in its training data | A hiring AI trained mostly on men's CVs prefers men (Amazon dropped such a tool in 2018) |
Privacy | Personal data collected or used without permission | Loan apps reading phone contacts and messaging them to shame borrowers |
Deepfakes | AI-made fake video, photo or voice that looks real | A cloned voice note asking parents for urgent money; fake political videos |
Safety and accountability | Who is responsible when AI causes harm? | A self-driving car accident: owner, company or programmer? |
Transparency | The AI cannot explain its decision (a 'black box') | A bank's AI refuses a loan but gives no reason |
§4.3 Responsible AI
Slide L02-S22
Definition: Responsible AI means building and using AI in a way that is fair, safe and respectful of people. |
Principle | Meaning |
|---|---|
Fairness | Treat all groups equally; test for bias |
Transparency | Explain how and why the AI decides |
Privacy | Collect only the data needed, with permission |
Safety | Test well and stop harm before release |
Accountability | A human or company is responsible for the AI |
Human control | People can check and override (cancel) the AI's decision |
Governments are making rules. The European Union's AI Act (2024) bans some uses and puts strict checks on high-risk AI. Pakistan's federal cabinet approved a National AI Policy in 2025 to build AI skills, use AI in public services and set ethical guidelines. As a student: do not submit AI text as your own work, check AI facts, and never upload other people's private data to AI tools.
5. Common Mistakes
Slide L02-S24
Wrong idea | Correct idea |
|---|---|
AI started with ChatGPT in 2022. | AI started in 1956 (Dartmouth). ChatGPT is a recent result of 70 years of work. |
ChatGPT is General (strong) AI. | ChatGPT is Narrow AI. General AI does not exist yet. |
Every robot is AI. | A robot repeating one fixed movement is not AI. |
AI winter means AI used in winter. | AI winter = a period when trust and funding for AI were cut. |
AI is neutral, so it cannot be unfair. | AI learns from human data and can copy human bias. |
Word Bank
Word | Simple meaning | Example |
|---|---|---|
Milestone | An important event in a history | 1956 Dartmouth workshop |
Funding | Money given for research or a project | Funding was cut in the AI winter |
Optimism | Strong hope that things will go well | "Human-level AI in 20 years" |
Expert system | A program of IF-THEN rules that copies a human expert | MYCIN for blood infections |
GPU | Graphics processing unit: a chip that does many calculations at once | Used to train deep learning |
Sector | A field of work or business | Healthcare, banking |
Effector | A part of a robot that acts on the world | Motor, wheel, arm |
Consciousness | Being aware of yourself and your thoughts | Needed for self-aware AI |
Bias | Unfair preference for or against a group | Hiring AI preferring men |
Deepfake | AI-made fake video, photo or voice | Cloned voice note |
Hallucination (AI) | A confident but false answer from an AI | A chatbot inventing a book that does not exist |
Accountability | Being responsible for results | Company answers for its AI's harm |
Summary
- AI grew from philosophy, mathematics, economics, neuroscience, psychology, engineering, linguistics and control theory.
- Key milestones: 1943 neuron model, 1950 Turing, 1956 Dartmouth, two AI winters (1974-80, 1987-93), 1997 Deep Blue, 2012 deep learning, 2016 AlphaGo, 2022 ChatGPT.
- Today's boom = big data + GPUs + better algorithms.
- AI is used in healthcare, agriculture, banking, transport, government and education. AI is the brain; robotics is the body.
- By capability: narrow (all AI today), general, super. By functionality: reactive, limited memory, theory of mind, self-aware.
- Main ethical issues: bias, privacy, deepfakes, safety/accountability, transparency. Responsible AI addresses them.
Self-Check Questions
- Name four subjects that contributed to AI and what each gave. (see §1.1)
- Why is 1956 called the birth year of AI? (see §1.2)
- What is an AI winter? Give the two periods and one cause of each. (see §1.3)
- What is an expert system? Name one example. (see §1.3)
- Give three reasons for the recent growth of AI. (see §1.5)
- Describe one AI application in Pakistani agriculture or healthcare as input → AI → output. (see §2.1-2.2)
- Differentiate between AI and robotics with an example of each. (see §2.3)
- Is ChatGPT narrow or general AI? Justify. (see §3.1)
- Differentiate reactive machines from limited memory AI with examples. (see §3.2)
- Explain bias and deepfakes with one example each. (see §4.2)
- List the six principles of responsible AI. (see §4.3)
Reading: Kaplan (2016) Chapters 1 and 7; Govers (2018) Chapter 1; Russell & Norvig (4e) Sections 1.2-1.3.