1 of 61

What Does ChatGPT Understand?

Alistair Edwards

  • Ownership
  • Water out
  • Water in
  • Sound

2 of 61

3 of 61

What Does ChatGPT Understand?

Alistair Edwards

Honorary Fellow, University of York

4 of 61

What does ChatGPT understand?

Nothing

5 of 61

The End

Questions?

6 of 61

No, seriously

We will look at artificial intelligence (AI)

How it developed leading to things like ChatGPT

Question ‘understanding’ and AI

7 of 61

But first…

A bit of background

8 of 61

Dr Alistair D N Edwards

Education

  • Wilmslow Grammar School
    • Maths, physics & chemistry
  • University of Warwick
    • BSc Physics & Computer Science
  • Pennsylvania State University
    • MSc Computer Science
  • Open University
    • PhD Educational Technology

9 of 61

Dr Alistair D N Edwards

Career

  • University of Manchester Institute of Science & Technology (UMIST)
    • Computer Officer
  • Open University
    • Lecturer
  • University of York
    • Senior Lecturer
    • Honorary Fellow

10 of 61

Career – all in universities

But also ‘real-world’ experience through consultancies

  • Toby Churchill Ltd
  • BT
  • Orange
  • William Hill

11 of 61

Honorary Fellow

Retired in 2018

An active retirement

12 of 61

‘Lecturer’

Teaching

…and research

13 of 61

Teaching

Algorithms and data structures

Human aspects of Computer Science

Web design

Advanced topics in interactive technologies

Crash course on C programming

IT as an enabling technology

Lexical and syntax analysis of programming languages

14 of 61

Research

Human-computer interaction

  • Multimodal interaction

speech

non-speech sounds

tactile interaction

  • Users with particular needs
    • blind people
    • older people
  • Usable security
  • Culture and interaction

15 of 61

What is Artificial Intelligence (AI)?

16 of 61

AI

To me:

Getting a computer to do something that previously could only be done by a person

?

17 of 61

Getting a computer to do something that previously could only be done by a person

A moving target

18 of 61

1950s

Compilers

Translate ‘high-level’ code to machine code

19 of 61

Programming

Do you write programs?

What languages do you use?

‘High-level’ languages

20 of 61

Compilers

function main(void)

{

printf(“Hello world!\n”);

}

21 of 61

1950s

LOAD 100

ADD 101

STORE 103

01000 1100100

01011 1100101

01101 1100111

22 of 61

1950s

ADD

01011

STORE

01101

JUMP

10100

SUB

01110

LOAD

01000

TEQ

10010

STOP

00000

TEST

10001

23 of 61

1950s

Manchester Mark 1

24 of 61

1950s

Compilers

    • Doing something previously only done by people

AI?

25 of 61

AI has experienced highs and lows

26 of 61

1950s

AI is born

Alan Turing

27 of 61

“This is only a foretaste of what is to come, and only the shadow of what is going to be. We have to have some experience with the machine before we really know its capabilities. It may take years before we settle down to the new possibilities, but I do not see why it should not enter any one of the fields normally covered by the human intellect, and eventually compete on equal terms.”

“This is only a foretaste of what is to come, and only the shadow of what is going to be. We have to have some experience with the machine before we really know its capabilities. It may take years before we settle down to the new possibilities, but I do not see why it should not enter any one of the fields normally covered by the human intellect, and eventually compete on equal terms.”

28 of 61

1950s

AI is born

Alan Turing

    • practically invents the computer
    • but also thinks about AI

29 of 61

30 of 61

31 of 61

1950s

The ‘AI winter’

    • a low

1960s

1970s

32 of 61

Japanese Fifth Generation Project

UK Alvey Programme

A high?

1980s

33 of 61

ChatGPT etc

2000s

2010s

2020s

34 of 61

ChatGPT is an LLM

Large language model

How does that work?

35 of 61

Predictive text

‘The boy hits the ball’

36 of 61

The boy hits the ball

37 of 61

How does predictive text work?

Statistics on how words are used

    • by me
    • and everyone else

38 of 61

LLMs are much the same

Based on millions of examples of text on the internet

39 of 61

The Chinese Room

A thought experiment

John Searle

1980

40 of 61

Chinese Room

41 of 61

Chinese Room

Searle understands nothing

But to external observers the room appears to understand

42 of 61

1950s

ADD

01011

STORE

01101

JUMP

10100

SUB

01110

LOAD

01000

TEQ

10010

STOP

00000

TEST

10001

43 of 61

Chinese Room

Searle understands nothing

But to external observers the room appears to understand

44 of 61

ChatGPT is like the Chinese room

It takes inputs and generates outputs

Its ‘book’ is the whole Internet

It understands nothing

45 of 61

Google Translate

Also understands nothing

It works like ChatGPT

    • compares writing in different languages
    • Wikipedia is very useful

46 of 61

47 of 61

It understands nothing

Does it matter?

48 of 61

Statistical AIs are good

…but they cannot explain their reasoning

49 of 61

Other AI technologies can give explanations

Is that important?

50 of 61

What about a medical AI?

AI diagnoses condition A

    • and corresponding treatment

Why did it not diagnose similar condition B?

51 of 61

If a medical AI can explain its reasoning does it understand medicine?

52 of 61

Conclusions

ChatGPT understands nothing

Will computers ever understand anything?

Does it matter if they do?

53 of 61

Question

54 of 61

Question

What does it mean ‘to understand’?

55 of 61

Questions and discussion

AI

or anything else I can help with

56 of 61

Careers

Careers of some of our graduates

57 of 61

Careers

Principal Engineer in the Customer Products Team at the FT

Alice Bartlett

58 of 61

Careers

Co-founder of Ethereum and creator of Polkadot and Kusama - cryptocurrencies

Gavin Wood

59 of 61

Careers

Vice-President for security at Google Cloud

Phil Venables

60 of 61

Careers

  • Web design and development
  • Professors/Lecturers
  • Research students
  • Teachers
  • Social media
  • UX (including Lego!)
  • Software engineering
  • Quality Assurance
  • Risk Advisor
  • Computer Support
  • Analysts
  • Consultant
  • Vicar
  • Sake evangelist

61 of 61

alistair.edwards@york.ac.uk

@ADNEdwards