Beyond the Asymmetric Turing Test Fintan Mallory Rethinking, - - PowerPoint PPT Presentation

beyond the asymmetric turing test
SMART_READER_LITE
LIVE PREVIEW

Beyond the Asymmetric Turing Test Fintan Mallory Rethinking, - - PowerPoint PPT Presentation

Prologue Beyond the Asymmetric Turing Test Beyond the Asymmetric Turing Test Fintan Mallory Rethinking, Reworking and Revolutionising The Turing Test August 1, 2019 Prologue Beyond the Asymmetric Turing Test Prologue Wittgenstein: Turing


slide-1
SLIDE 1

Prologue Beyond the Asymmetric Turing Test

Beyond the Asymmetric Turing Test

Fintan Mallory

Rethinking, Reworking and Revolutionising The Turing Test

August 1, 2019

slide-2
SLIDE 2

Prologue Beyond the Asymmetric Turing Test

Prologue

Wittgenstein: Turing thinks that he and I are using the word ‘experiment’ in two different ways. But I want to show that this is

  • wrong. That is to say, I think that if I could make myself quite

clear, then Turing would give up saying that in mathematics we make experiments. If I could arrange in their proper order certain well- known facts, then it would become clear that Turing and I are not using the word ‘experiment’ differently.

slide-3
SLIDE 3

Prologue Beyond the Asymmetric Turing Test

Prologue

Wittgenstein: Turing thinks that he and I are using the word ‘experiment’ in two different ways. But I want to show that this is

  • wrong. That is to say, I think that if I could make myself quite

clear, then Turing would give up saying that in mathematics we make experiments. If I could arrange in their proper order certain well- known facts, then it would become clear that Turing and I are not using the word ‘experiment’ differently. Turing: I see your point

slide-4
SLIDE 4

Prologue Beyond the Asymmetric Turing Test

Prologue

Wittgenstein: Turing thinks that he and I are using the word ‘experiment’ in two different ways. But I want to show that this is

  • wrong. That is to say, I think that if I could make myself quite

clear, then Turing would give up saying that in mathematics we make experiments. If I could arrange in their proper order certain well- known facts, then it would become clear that Turing and I are not using the word ‘experiment’ differently. Turing: I see your point Wittgenstein: I have no point. If you want to interpret the word ‘experiment’ in a wider sense, then by all means do so. And let us see whether what I have been saying may not be false.

slide-5
SLIDE 5

Prologue Beyond the Asymmetric Turing Test

Types of Test

Litmus Test Identify the presence or absence of a property Shibboleth Test Identify if the candidate is one of us

slide-6
SLIDE 6

Prologue Beyond the Asymmetric Turing Test

Turing’s Paradigm

Not a litmus test ‘The original question ’Can machines think?’ I believe to be too meaningless to deserve discussion.’ Prediction about usage ‘I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.’

slide-7
SLIDE 7

Prologue Beyond the Asymmetric Turing Test

What is Tested?

The Standard Turing Test doesn’t test for mindedness but like-mindedness

slide-8
SLIDE 8

Prologue Beyond the Asymmetric Turing Test

What is Tested?

The Standard Turing Test doesn’t test for mindedness but like-mindedness The recognition of like-mindedness

slide-9
SLIDE 9

Prologue Beyond the Asymmetric Turing Test

What is Tested?

The Standard Turing Test doesn’t test for mindedness but like-mindedness The recognition of like-mindedness This isn’t a simple response-dependent property

slide-10
SLIDE 10

Prologue Beyond the Asymmetric Turing Test

A dialectical tension in the Turing paradigm

Working assumption: Intelligence manifests itself in the ability to distinguish participants

slide-11
SLIDE 11

Prologue Beyond the Asymmetric Turing Test

A dialectical tension in the Turing paradigm

Working assumption: Intelligence manifests itself in the ability to distinguish participants Working method: Test for intelligence measures ability to present

  • neself as intelligent rather than assess intelligence
slide-12
SLIDE 12

Prologue Beyond the Asymmetric Turing Test

A dialectical tension in the Turing paradigm

Working assumption: Intelligence manifests itself in the ability to distinguish participants Working method: Test for intelligence measures ability to present

  • neself as intelligent rather than assess intelligence

Classic objection: Test cannot identify reflexive intelligence (self-consciousness)

slide-13
SLIDE 13

Prologue Beyond the Asymmetric Turing Test

A dialectical tension in the Turing paradigm

Working assumption: Intelligence manifests itself in the ability to distinguish participants Working method: Test for intelligence measures ability to present

  • neself as intelligent rather than assess intelligence

Classic objection: Test cannot identify reflexive intelligence (self-consciousness) Proposal: Take Like-Mindedness seriously! Like-mindedness is an equivalence relation., i.e. it is symmetric, transitive and reflexive

slide-14
SLIDE 14

Prologue Beyond the Asymmetric Turing Test

Symmetry

Symmetry Rxy → Ryx

slide-15
SLIDE 15

Prologue Beyond the Asymmetric Turing Test

Symmetry

Symmetry Rxy → Ryx CAPTCHAS ’Completely Automated Public Turing test to tell Computers and Humans Apart’ are unintelligent (and discriminatory in the pejorative sense)

slide-16
SLIDE 16

Prologue Beyond the Asymmetric Turing Test

Symmetry

Symmetry Rxy → Ryx CAPTCHAS ’Completely Automated Public Turing test to tell Computers and Humans Apart’ are unintelligent (and discriminatory in the pejorative sense) Why this matters: Pride condition If you can’t recognise me as intelligent, I won’t recognise you as intelligent

slide-17
SLIDE 17

Prologue Beyond the Asymmetric Turing Test

Transitivity

Transitivity Rxy ∧ Ryz → Rxz

slide-18
SLIDE 18

Prologue Beyond the Asymmetric Turing Test

Transitivity

Transitivity Rxy ∧ Ryz → Rxz CATS Cats are dumb. Cute - yes - but very stupid

slide-19
SLIDE 19

Prologue Beyond the Asymmetric Turing Test

Transitivity

Transitivity Rxy ∧ Ryz → Rxz CATS Cats are dumb. Cute - yes - but very stupid Why this matters: Trust condition If I recognise you as intelligent, then I recognise your ability to recognise others as intelligent

slide-20
SLIDE 20

Prologue Beyond the Asymmetric Turing Test

Reflexivity

Reflexivity Rxx

slide-21
SLIDE 21

Prologue Beyond the Asymmetric Turing Test

Reflexivity

Reflexivity Rxx The rabbit in the hat If a relation over a set is transitive and symmetric, then it is reflexive

slide-22
SLIDE 22

Prologue Beyond the Asymmetric Turing Test

Reflexivity

Reflexivity Rxx The rabbit in the hat If a relation over a set is transitive and symmetric, then it is reflexive Issues

  • 1. Shift-reflexivity: Rxy → Ryy
  • 2. ‘Self-consciousness exists in itself and for itself, in that, and by

the fact that it exists for another self-consciousness; that is to say, it is only by being acknowledged or recognized.’

  • 3. Self-consciousness is an advantage in the Turing Test anyway
slide-23
SLIDE 23

Prologue Beyond the Asymmetric Turing Test

The Turing Games

1 Judges compete

slide-24
SLIDE 24

Prologue Beyond the Asymmetric Turing Test

The Turing Games

1 Judges compete 2 Round-Robin format

slide-25
SLIDE 25

Prologue Beyond the Asymmetric Turing Test

The Turing Games

1 Judges compete 2 Round-Robin format 3 Two conversations per pairing

slide-26
SLIDE 26

Prologue Beyond the Asymmetric Turing Test

The Turing Games

1 Judges compete 2 Round-Robin format 3 Two conversations per pairing 4 Private-Individual conversations

slide-27
SLIDE 27

Prologue Beyond the Asymmetric Turing Test

The Turing Games

1 Judges compete 2 Round-Robin format 3 Two conversations per pairing 4 Private-Individual conversations 5 Chicanery and deception encouraged

slide-28
SLIDE 28

Prologue Beyond the Asymmetric Turing Test

The Turing Games

1 Judges compete 2 Round-Robin format 3 Two conversations per pairing 4 Private-Individual conversations 5 Chicanery and deception encouraged 6 Graph of recognition produced

slide-29
SLIDE 29

Prologue Beyond the Asymmetric Turing Test

Failure cases

Symmetry A judges B as human but B does not judge A as human Transitivity A judges B as human, B judges C as human but A does not judge C as human

slide-30
SLIDE 30

Prologue Beyond the Asymmetric Turing Test

Virtues

A stronger but tractable test for A.I. research It integrates several areas of current research

slide-31
SLIDE 31

Prologue Beyond the Asymmetric Turing Test

Virtues

A stronger but tractable test for A.I. research It integrates several areas of current research A better thought experiment A machine that passed this would be de facto one of us

slide-32
SLIDE 32

Prologue Beyond the Asymmetric Turing Test

Virtues

A stronger but tractable test for A.I. research It integrates several areas of current research A better thought experiment A machine that passed this would be de facto one of us Connection to the evolution of intelligence Human intelligence evolved in a social context of shifting and signalling affiliation (see Searcy & Nowicki, 2005, McNally & Jackson, 2013)

slide-33
SLIDE 33

Prologue Beyond the Asymmetric Turing Test

Vices

Too hard? The test is difficult to pass but easier than the Total Turing Test

slide-34
SLIDE 34

Prologue Beyond the Asymmetric Turing Test

Vices

Too hard? The test is difficult to pass but easier than the Total Turing Test Recognition is not an equivalence relation The algebra of human interaction is a complex structure

slide-35
SLIDE 35

Prologue Beyond the Asymmetric Turing Test

Vices

Too hard? The test is difficult to pass but easier than the Total Turing Test Recognition is not an equivalence relation The algebra of human interaction is a complex structure Reflexivity = self-consciousness There may be other pre-linguistic recognition relations [Le regard]

slide-36
SLIDE 36

Prologue Beyond the Asymmetric Turing Test

Vices

Too hard? The test is difficult to pass but easier than the Total Turing Test Recognition is not an equivalence relation The algebra of human interaction is a complex structure Reflexivity = self-consciousness There may be other pre-linguistic recognition relations [Le regard] The logocentrism charge The test is still too focussed on language processing and overlooks

  • ther traits of intelligence
slide-37
SLIDE 37

Prologue Beyond the Asymmetric Turing Test

Where next?

Prediction: Within 5 years a machine-coined neologism will have entered human usage

slide-38
SLIDE 38

Prologue Beyond the Asymmetric Turing Test

Where next?

Prediction: Within 5 years a machine-coined neologism will have entered human usage ’They will make mistakes at times, and at times they may make new and very interesting statements, and on the whole the output

  • f them will be worth attention to the same sort of extent as the
  • utput of a human mind.’ Turing, 1948
slide-39
SLIDE 39

Prologue Beyond the Asymmetric Turing Test

Where next?

Prediction: Within 5 years a machine-coined neologism will have entered human usage ’They will make mistakes at times, and at times they may make new and very interesting statements, and on the whole the output

  • f them will be worth attention to the same sort of extent as the
  • utput of a human mind.’ Turing, 1948

‘The second model is fixed as we found that updating the parameters of both agents led to divergence from human language.’ Lewis et. al. 2017 [Facebook Research Lab]

slide-40
SLIDE 40

Prologue Beyond the Asymmetric Turing Test

Tapadh leibh Thank you

slide-41
SLIDE 41

Prologue Beyond the Asymmetric Turing Test

Bibliography Deal or No Deal? End-to-End Learning for Negotiation Dialogues Mike Lewis, Denis Yarats, Yann N. Dauphin, Devi Parikh and Dhruv Batra Turing, ‘Intelligent Machinery,? National Physical Laboratory Report (1948). Reproduced in Copeland, ed., The Essential Turing (New York: Oxford, 2004), pp. 410?32. Turing, ‘Computing Machinery and Intelligence’