COMP5001 Artificial Intelligence
Week 1 — Introduction
Dr Sakib Anwar
Week 1: Introduction. Plan: a 45-minute lecture, then the seminar, which runs from the Week 1 seminar sheet (Seminars/week01-seminar.pdf; answers in week01-seminar-tutor.pdf): AI or not AI, four puzzles, AI around us. There is no practical this week. The module slides (about me, the module, the AI policy, reading) take about 15 minutes, so keep the icebreaker short.
About Me
Dr Sakib Anwar, Senior Lecturer in Economics and Finance
BSc Economics (Nottingham), MSc Economics (Surrey)
MRes Economics (Essex), PhD Economics (Lancaster)
Research: game theory, behavioural and experimental economics
More at sakibanwar.com/research
About 2 minutes. Game theory is the study of strategic decisions; it links directly to adversarial search and minimax in Week 2.
How About You?
Introduce yourself to the room. Round the room
Your name
Your programme
One thing you would love AI to do for you
One thing AI should never do
Icebreaker, about 5 minutes. Go round the room at 20 seconds each, or take a sample if the group is large. Note the answers: they make good examples for the Discuss slides later in the term.
Learning Outcomes
Main areas of AI: symbolic and statistical
Search, planning and reasoning
Designing AI for a scenario
AI in businesses and organisations
By the end of the module students can: (a) develop an understanding of the main areas of AI, including symbolic and statistical approaches; (b) understand different methods for search, planning and reasoning; (c) design and discuss applications of AI to a particular scenario; (d) demonstrate awareness of the uses and implications of AI technologies within businesses and organisations. Point out that (c) and (d) are about using AI in organisations, not only the algorithms; the seminars cover them every week.
Assessment
One portfolio: 100% of the module
AI system implementation: 30%
Discussion and topic research: 55% (1,500 words)
Reflection: 15% (500 words)
Deadline: 3.30pm, Wednesday 6 January 2027
The implementation counts as 1,000 words equivalent, so the portfolio is 3,000 words in total. More on the portfolio at the end of today's seminar. Students do not need to start yet, but short weekly notes will help with the reflection.
AI Policy: Not Permitted
No generative AI in any part of your portfolio Chatbots: ChatGPT, Claude, Gemini, Copilot, DeepSeek, Perplexity
Coding tools: GitHub Copilot, Cursor, Claude Code, Codex, Windsurf
AI features in other software, and AI translation or paraphrasing tools
One exception: Grammarly-style fixes to sentences you wrote yourself
Ask the room to list the AI tools they use before showing the list; many miss that code completion in their editor counts. Full policy: students must not use generative AI tools at any stage of the portfolio. This includes chatbots and assistants (ChatGPT, Claude, Gemini, Microsoft Copilot, DeepSeek, Perplexity); agentic tools that write, edit or run code (Claude Code, Google Antigravity, Cursor, OpenAI Codex, Windsurf); AI features built into other software (GitHub Copilot, AI code completion in VS Code, Gemini in Google Colab, Copilot in Word); and AI translation and paraphrasing tools such as DeepL or QuillBot used to produce or rewrite text. The ban covers every part of the work: ideas, understanding the readings, writing or fixing code, choosing methods, and drafting or rewriting text. The one exception: Grammarly or similar to correct spelling, grammar and punctuation, and to improve the clarity of individual sentences students wrote themselves, but not to generate content or rewrite whole paragraphs.
AI Policy: What You Can Use
Module slides, seminar sheets and textbooks
Documentation: Python, scikit-learn, TensorFlow, NLTK
Ordinary web searches, but not AI-generated summaries
Help from the module team, and ideas from classmates
In class: AI tools when an activity asks you to use them
Ask first: why ban AI tools in a module about AI? The portfolio tests whether students can build, explain and judge AI systems themselves; studying AI is not the same as having AI do the work. In practicals and seminars students may use AI tools when an activity asks for it (for example, Week 9 compares an LLM chatbot with a Naive Bayes classifier), but nothing produced by an AI tool may go into the portfolio. Classmates can discuss ideas, but code and writing must be their own.
AI Policy: How It Is Checked
You may be asked to explain your code and choices in a short conversation
Keep your version history, for example Git commits
Suspected AI use is handled as academic misconduct
Sign the declaration on your portfolio’s title page
Being unable to explain your own work may be treated as suspected academic misconduct. Version history protects honest students. The declaration: "I confirm that I have not used generative AI tools at any stage in producing this portfolio or its code, other than to correct spelling, grammar and punctuation or to improve the clarity of individual sentences I wrote myself." The full policy is in the assessment brief.
10 Weeks
Introduction
Search
Heuristics and Logic
Knowledge and Planning
Uncertainty
Optimization
Learning
Neural Networks
Language
AI in Practice and Revision
Weeks 2 to 9 each teach one topic. Search is big enough for two weeks, so Week 3 starts with heuristics (the end of search) and then moves on to logic. Week 4 is knowledge and planning. From Week 4 the seminar includes an "AI in practice" activity, which is practice for the portfolio. Week 10 brings everything together and is also revision.
Each Week
Lecture: one AI topicSeminar: AI in the real world, discussion and activitiesPractical: Python code you run and changeAll three feed into your portfolio
Each week is one 3-hour block: about an hour of lecture, a short seminar, a break, then a practical. Week 1 is different: a short lecture and a longer seminar, and no practical. Level 4 Python is enough: the practicals start from working code that students run and change.
Essential Reading
Russell, S. and Norvig, P. (2020) Artificial Intelligence: A Modern Approach . 4th edn. Pearson
Poole, D.L. and Mackworth, A.K. (2023) Artificial Intelligence: Foundations of Computational Agents . 3rd edn. Cambridge University Press
Poole and Mackworth is free online at artint.info
The chapters for each week are on the module page
Russell and Norvig is the standard AI textbook. Poole and Mackworth covers the same ground and is free online. Chapter 1 of either is a good start this week.
Office Hours
Module leader: Dr Sakib Anwar
Email: sakib.anwar@winchester.ac.uk
Office: UWBS 005
Wednesdays 4 to 5pm, 30-minute slots
Book at calendly.com/sakib-anwar
Point to office hours and the module page, which has the slides, seminar sheets, reading and the assessment brief.
Section 1
What is AI?
About 10 minutes for this section. Warm-up question: what is the most impressive AI you have used, and the most disappointing?
artificial intelligence
Getting computers to do tasks that would need intelligence if people did them.
Ask the class for their own definition first, then click to show this one. There is no single agreed definition; this one is simple and good enough for the module. The target moves: once computers do something reliably (chess, route finding), people stop calling it AI.
Four Goals for AI
Think like people
Act like people
Think logically
Act to get the best result
From Russell and Norvig, chapter 1. The Turing test (coming up) is about acting like people. Most modern AI, and this module, aims at the last one: an agent that acts to get the best result it can. Ask: which goal does ChatGPT aim at?
agent
Entity that perceives its environment and acts upon that environment.
CS50's definition, used all term. An agent perceives (inputs) and acts (outputs). Examples: a sat-nav, a chess program, a robot vacuum, a spam filter. Ask what each one perceives and how it acts. The Week 1 notes (Sections 4 to 6) go further: sensors and actuators, PEAS, environment types and five kinds of agent.
Artificial Intelligence
Search
Logic
Knowledge and Planning
Uncertainty
Optimization
Learning
Neural Networks
Language
The eight topics of the module, one per week from Week 2 to Week 9 (Week 3 also finishes search, with heuristics). Each week ends with real-world examples of the topic and a discussion.
symbolic AI
AI that works with facts, rules and symbols that people write down.
Search, Logic, Knowledge, Planning and Optimization (Weeks 2, 3, 4 and 6). Easy to explain and check, but people have to write the knowledge down.
statistical AI
AI that learns patterns from data and works with probabilities.
Uncertainty, Learning, Neural Networks and Language (Weeks 5, 7, 8 and 9). Needs lots of data and is harder to explain. Many real systems combine both families: a sat-nav finds routes with search (symbolic) but predicts traffic from data (statistical).
Section 2
A Short History of AI
About 20 minutes. Each milestone links to a week of the module; mention the week as you go. The pattern to notice: big promises, then disappointment.
1950
Alan Turing asks “Can machines think?” and proposes the imitation game, now called the Turing test
A judge chats by text with a hidden person and a hidden machine. If the judge cannot tell which is which, the machine passes. This is the "act like people" goal. Ask: would a chatbot pass today? Would that make it intelligent?
1956
A summer workshop at Dartmouth College gives the field its name: artificial intelligence
John McCarthy coined the name in the proposal for the workshop, written with Marvin Minsky, Nathaniel Rochester and Claude Shannon. They hoped a small group could make big progress in one summer. Over-optimism like this is a theme of the whole history.
1958
Frank Rosenblatt introduces the perceptron, an early neural network that learns from examples
Newspapers reported that it would one day walk, talk and be aware of itself. In 1969 Minsky and Papert showed the limits of a single perceptron, and interest in neural networks faded for years. Perceptrons return in Week 7 (Learning) and Week 8 (Neural Networks).
1966
ELIZA, a chatbot that follows simple pattern rules, convinces some users that it understands them
Written by Joseph Weizenbaum at MIT. It turned the user's words back into questions, like a therapist. Weizenbaum was worried by how readily people confided in it. People reading understanding into simple programs is still called the ELIZA effect. Week 9 (Language).
1966–1972
Shakey, the first mobile robot that could plan its own actions
Built at SRI (Stanford Research Institute). It was slow, but it was the first robot to combine seeing, planning and moving. Work on Shakey produced A* search (1968) and the STRIPS planner (1971). A* appears in Week 3 and STRIPS in Week 4.
1973
The Lighthill Report tells the UK government that AI has not delivered on its promises
Sir James Lighthill, a mathematician, reviewed AI for the Science Research Council. UK funding for AI was cut, and interest fell in the US too. This is the first "AI winter" of the mid-1970s.
1980s
Expert systems put human experts’ rules into software, and businesses start to use AI
Example: XCON at Digital Equipment Corporation used thousands of rules to configure computer orders and saved the company millions a year. Week 4 (Knowledge). Expert systems were costly to keep up to date, and the market collapsed in the late 1980s: the second AI winter.
1997
IBM’s Deep Blue beats the world chess champion, Garry Kasparov
Deep Blue used minimax search with alpha-beta pruning on fast custom hardware. Kasparov had beaten an earlier version in 1996. Week 2 (Search).
2012
A deep neural network called AlexNet wins an image recognition contest by a wide margin
The ImageNet challenge. AlexNet's error rate was about 15%, against about 26% for the next best entry. Large datasets and graphics cards (GPUs) made deep learning practical. Week 8 (Neural Networks).
2016
DeepMind’s AlphaGo beats Lee Sedol, one of the world’s best Go players
AlphaGo combined search with neural networks. Go has more possible positions than there are atoms in the observable universe, far too many for Deep Blue-style search on its own. Weeks 2, 7 and 8.
2017–2022
Transformers are invented, leading to large language models such as ChatGPT
Google researchers introduced the transformer in 2017; it is also behind modern translation tools. OpenAI released ChatGPT in November 2022 and it became one of the fastest-growing apps ever. Week 9 (Language).
2024
The EU AI Act, the first comprehensive law on AI, comes into force
It sorts AI systems into risk levels, with the strictest rules for high-risk uses such as hiring. Its rules apply in stages from 2025. It matters to UK organisations that sell into the EU. The Week 7 seminar uses it (with UK GDPR), and so does Part 2d of the portfolio.
A Short History of AI
winter
winter
1950
1960
1970
1980
1990
2000
2010
2020
1950
Turing test
1956
Dartmouth: the name AI
1958
Perceptron
1966
ELIZA
1966–72
Shakey the robot
1973
Lighthill Report
1980s
Expert systems
1997
Deep Blue
2012
AlexNet
2016
AlphaGo
2017–22
Transformers, ChatGPT
2024
EU AI Act
Big promises, then disappointment and funding cuts: the pattern has happened twice. Ask: are we in a boom now? What might cause the next winter?
Hype and Winters
Big promises attract money and attention
Results fall short, so funding is cut: an “AI winter”
Winters in the mid-1970s and the late 1980s
Ask of any AI claim: what can it actually do, and how do we know?
Links to LO (d). Organisations buying AI today face the same question. Encourage healthy scepticism, not cynicism.
Your Portfolio
Part 1: build a small AI system (30%)
Part 2: advise an organisation on AI (55%)
Part 3: reflect on your learning (15%)
The weekly practicals and seminars prepare you for each part
Point students to the assessment brief. Weekly practicals give code to adapt for Part 1. The seminar activities from Week 4 are practice for Part 2. Short weekly notes help with Part 3.
Before Week 2
End of the lecture (about 45 minutes). Hand out the Week 1 seminar sheet; its last section is the same setup checklist, so students can finish the setup at the end of the seminar or at home.
Thank you
Seminar: the Week 1 seminar sheet (no computers needed)
Recap: AI is a family of methods, not one thing; it has a history of hype and winters; next week the methods start with search. Before next week: the setup on the previous slide. Optional reading: Russell & Norvig, chapter 1.
COMP5001 Artificial Intelligence — Week 1: Introduction
