Recap of Lecture 2 and Homework Check
LAST LECTURE · 5 MIN
History
Types
Ethics
1950 Turing, 1956 Dartmouth, two AI winters, 1997 Deep Blue, 2012 deep learning, 2022 ChatGPT.
Narrow (all AI today), general, super. Reactive, limited memory, theory of mind, self-aware.
Bias, privacy, deepfakes, accountability, transparency → responsible AI.
Homework check: read out your timeline. Did you mark BOTH AI winters? Did you start before 2022?
Today: the man behind 1950 - Alan Turing - gave AI two big ideas: a machine model of computing and a test for intelligence.
Today's Plan (90 minutes)
ROADMAP
0-5
Recap of Lecture 2 and homework check
S02
5-8
Plan and learning outcomes
S03-S04
8-45
Module 1: The Turing machine - parts, rules, two worked traces
S05-S11
45-52
Church-Turing thesis and the halting problem
S12-S13
52-68
Module 2: The Turing Test, its history and CAPTCHA
S14-S18
68-76
Module 3: Can machines think? The Chinese Room
S19-S21
76-85
Class activity: trace an addition machine
S22
85-90
Mistakes, recap quiz, homework
S23-S25
Learning Outcomes (SLOs)
BY THE END OF THIS LECTURE YOU WILL BE ABLE TO
SLO 3.1
Describe the parts of a Turing machine: tape, head, states and transition rules
→ CLO1 · C1 Remember
SLO 3.2
Trace a simple Turing machine on an input, step by step, and give its output
→ CLO1 · C2 Understand
SLO 3.3
Explain the Church-Turing thesis and the halting problem in simple words
→ CLO1 · C2 Understand
SLO 3.4
Explain the Turing Test, the total Turing Test and CAPTCHA, with their history
→ CLO1 · C2 Understand
SLO 3.5
Explain the Chinese Room argument and the main weaknesses of the Turing Test
→ CLO1 · C2 Understand
- Course Learning Outcome(s) served
- CLO1: Explain the applications of AI and agent-based approach to AI.
Slide 5
MODULE 1
The Turing Machine
1936: the simplest machine that can compute anything a computer can
Why Did Turing Invent This Machine?
MODULE 1 · TURING MACHINE
1936: 'computer' = a person
Turing's question
His answer
Clerks did calculations with paper and pencil by following fixed rules.
What exactly can be computed by following simple rules, one step at a time?
An imaginary machine: tape + head + states + rules. Paper: 'On Computable Numbers' (1936).
Big idea: a Turing machine is a MODEL (imaginary design) of computing. Your phone cannot compute anything a Turing machine cannot - it is only faster.
Why AI cares: if thinking is computing, then a Turing machine could, in principle, think. This links the two halves of today.
Four Parts of a Turing Machine
MODULE 1 · TURING MACHINE
␣
␣
␣
HEAD · state q0
1. Tape
2. Head
3. States
4. Rules
Very long strip of cells. Each cell holds one symbol: 0, 1, + or blank (␣).
Reads one cell, writes a symbol, moves Left (L), Right (R) or Stays (S).
The 'state of mind': q0 (start), q1, q2 ... and HALT (stop).
(state, read) → (write, move, next state). This is the program.
Analogy: a cashier with a long receipt roll, a pen, one finger on the current line, and a rule card.
Reading a Transition Rule
MODULE 1 · TURING MACHINE
Format: ( current state , symbol read ) → ( symbol to write , move , next state )
Machine 1: the BIT FLIPPER. It turns every 0 into 1 and every 1 into 0. Example input: 1011.
Try: before the next slide, predict what 1011 becomes and how many steps it takes.
Worked Trace 1: Bit Flipper on 1011
MODULE 1 · TURING MACHINE
Result
- Input 1011 → output 0100.
- The machine halted after reading the first blank.
Notice: the head never moves left here. Machine 2 (next) must move LEFT to carry a 1 - like doing addition by hand.
Machine 2: Add 1 to a Binary Number
MODULE 1 · TURING MACHINE
q0
Job: walk to the right end of the number.
q1
Job: add the carry while moving left.
By hand: 1011 + 1: 1+1 = 0 carry 1 → 1+1 = 0 carry 1 → 0+1 = 1 → answer 1100 (11 + 1 = 12).
Worked Trace 2: Increment 1011 to 1100
MODULE 1 · TURING MACHINE
Final tape (head on cell 1, HALT):
␣
␣
␣
HEAD · state HALT
Result: 1011 (eleven) → 1100 (twelve).
Exam tip: always show state, head position, symbol read and action for EVERY step.
Church-Turing Thesis and the Universal Machine
MODULE 1 · WHAT IT MEANS
Church-Turing thesis
Universal Turing machine
Your phone
Anything that can be computed by an algorithm (step-by-step method) can be computed by a Turing machine. A 'thesis' = a strong belief, not a proof.
One machine that reads another machine's rules from its tape and runs them = a stored-program computer.
One hardware, thousands of apps: a universal machine. Python and C++ are 'Turing complete'.
For AI: if the human mind follows an 'algorithm', then a computer could in principle do the same. This is the hope behind strong AI.
A Limit: The Halting Problem
MODULE 1 · WHAT IT MEANS
- Question: can a program look at ANY other program and say whether it will stop or run forever?
- Turing's answer (1936): NO. No program can do this correctly for every program.
- Idea of the proof: suppose a perfect checker H exists. Build a program that asks H about itself and then does the OPPOSITE. H is then wrong - a contradiction.
- Meaning: some problems are undecidable (no algorithm can ever solve them), however fast the computer.
Everyday example
Windows shows 'Not responding'. It cannot know whether the app will recover, so it asks YOU: wait or close?
Remember: computers - and AI - have limits that no extra speed can remove.
Slide 14
MODULE 2
The Turing Test
1950: 'Can machines think?' becomes 'Can a machine fool a judge?'
The Imitation Game (Turing, 1950)
MODULE 2 · TURING TEST
Judge (C)
Hidden human (B)
Hidden machine (A)
Types questions to both players. Must decide: which one is the machine?
Answers honestly and tries to help the judge.
Answers and tries to make the judge believe it is the human.
What a Machine Needs to Pass
MODULE 2 · TURING TEST
Notice: passing the Total Turing Test needs almost every topic in this course.
The Turing Test in Practice
MODULE 2 · TURING TEST
1966
1991-2019
2014
2025
- ELIZA
- Simple patterns; some users believed it understood them.
- Loebner Prize
- Yearly contest. No machine won the gold prize.
- Eugene Goostman
- Posed as a 13-year-old boy; fooled 33% of judges using excuses.
- GPT-4.5 (UC San Diego)
- With a persona, judged human 73% of the time in a 3-party test.
Big question: modern chatbots can now pass short Turing Tests. Does that prove they think and understand? Many experts say no.
CAPTCHA: The Reverse Turing Test
MODULE 2 · TURING TEST
Full name: Completely Automated Public Turing test to tell Computers and Humans Apart (2003).
Twist: AI now solves many picture CAPTCHAs, so sites also watch your mouse movement and behaviour.
Slide 19
MODULE 3
Can Machines Really Think?
The Chinese Room argument and the limits of the Turing Test
The Chinese Room (John Searle, 1980)
MODULE 3 · CAN MACHINES THINK?
Input
Inside the room
Output
Questions written in Chinese are pushed into the room through a slot.
A man who knows NO Chinese follows a huge English rule book: 'if you see these symbols, send back those'.
Perfect Chinese answers come out. People outside think someone inside understands Chinese.
Searle: computers handle syntax, not semantics. Passing the test ≠ understanding.
Replies to Searle and Weaknesses of the Turing Test
MODULE 3 · CAN MACHINES THINK?
Systems reply
Searle's answer
The man does not understand, but the WHOLE system might - like one brain cell vs the whole brain.
Let the man memorise the whole book. He still understands no Chinese.
Link: this is why our course measures AI as rational agents (L01, L04), not by fooling people.
Activity: Trace an Addition Machine
CLASS ACTIVITY · PAIRS · 9 MIN
- Q1. Unary numbers: 11 = 2, 111 = 3. What should the answer be?
- Q2. Trace 11+111 step by step: state, cell, read, action.
- Q3. What is the final tape? How many steps (from step 0)?
- Q4. What does 111+1 become?
- Q5. Why does the machine need state q2?
Start tape (head on the first 1, state q0):
␣
+
␣
␣
Common Mistakes
WATCH OUT
WRONG IDEA
CORRECT IDEA
✗ A Turing machine is a real, physical computer.
✓ It is an imaginary model (thought experiment) of computing from 1936.
✗ The tape is short, like RAM.
✓ The tape is unlimited (infinitely long) in both directions.
✗ The head can jump to any cell.
✓ The head moves only ONE cell at a time: Left, Right or Stay.
✗ Passing the Turing Test proves a machine thinks.
✓ It shows it can imitate human chat. Searle: imitation ≠ understanding.
✗ CAPTCHA tests computers' intelligence.
✓ In CAPTCHA the computer is the judge and tests whether YOU are human.
Recap Quiz
CHECK YOUR UNDERSTANDING · 5 MIN
- Q1. Turing introduced his machine in the paper of year:
- (a) 1936 (b) 1950 (c) 1956 (d) 1980
- Q2. In rule (q1, 1) → (0, L, q1) the machine:
- (a) Writes 1, moves R (b) Writes 0, moves L (c) Halts (d) Writes 0, moves R
- Q3. The bit flipper turns 1011 into:
- (a) 1100 (b) 0100 (c) 1011 (d) 0011
- Q4. 'No program can decide for every program whether it stops' is the:
- (a) Church-Turing thesis (b) Halting problem (c) Chinese Room (d) Turing Test
- Q5. The Chinese Room argument was given by:
- (a) Alan Turing (b) John McCarthy (c) John Searle (d) Joseph Weizenbaum
- Q6. In a CAPTCHA, the judge is:
- (a) A human (b) A computer (c) A robot arm (d) No one
Homework and Next Lecture
BEFORE THE NEXT CLASS
- Task 1: Trace Machine 2 (increment, S10) on input 1001. Show every step as a table and give the final tape.
- Task 2: Design a Turing machine that replaces every 1 with 0 and leaves 0s unchanged. Write its rule table and trace it on 1101.
- Task 3: In 5-6 lines: does passing the Turing Test prove understanding? Use the Chinese Room in your answer.
- Read: Kaplan Ch 4; Russell & Norvig Section 1.1.1 and 28.1-28.2 (philosophy of AI).
NEXT LECTURE
- Lecture 4:
- Intelligent Agents: PEAS, Environments and Agent Types