Machine Learning Meets Public Policy Edward W. Felten Kahn - - PowerPoint PPT Presentation

▶
machine learning meets public policy
SMART_READER_LITE
LIVE PREVIEW

Machine Learning Meets Public Policy Edward W. Felten Kahn - - PowerPoint PPT Presentation

Sep 20, 2023 12 likes •905 views

Machine Learning Meets Public Policy Edward W. Felten Kahn Professor of Computer Science and Public Affairs Director, Center for Information Technology Policy Princeton University "AI is probably the most important thing humanity has

slide-1
SLIDE 1

Machine Learning Meets Public Policy

Edward W. Felten Kahn Professor of Computer Science and Public Affairs Director, Center for Information Technology Policy Princeton University

slide-2
SLIDE 2
slide-3
SLIDE 3
slide-4
SLIDE 4
slide-5
SLIDE 5
slide-6
SLIDE 6
slide-7
SLIDE 7

"AI is probably the most important thing humanity has ever worked on. I think

  • f it as something more

profound than electricity

  • r fire.”

Sundar Pichai, Google CEO 24 Jan 2018

7

slide-8
SLIDE 8

People who are affected by AI/ML deserve some say in how it is used.

slide-9
SLIDE 9

Decisions will be made. What is our role in the decisions?

slide-10
SLIDE 10
slide-11
SLIDE 11
slide-12
SLIDE 12
slide-13
SLIDE 13
slide-14
SLIDE 14
slide-15
SLIDE 15
slide-16
SLIDE 16

Truth

slide-17
SLIDE 17

Truth

slide-18
SLIDE 18
slide-19
SLIDE 19

Politics is not a search for truth.

slide-20
SLIDE 20

a feature, not a bug

slide-21
SLIDE 21

Democracy is not a search for truth.

slide-22
SLIDE 22

an algorithm for resolving disagreements

slide-23
SLIDE 23

no question is undecidable

slide-24
SLIDE 24

all questions are decidable in O(1) time

slide-25
SLIDE 25

no need to decide underlying facts

slide-26
SLIDE 26

no need for a coherent explanation

slide-27
SLIDE 27

and yet …

slide-28
SLIDE 28

individual legislators seem

slide-29
SLIDE 29

individual legislators seem logically inconsistent

slide-30
SLIDE 30

individual legislators seem logically inconsistent indifferent to truth

slide-31
SLIDE 31

Politicians behave that way for a reason.

slide-32
SLIDE 32

Consider the following model …

slide-33
SLIDE 33

P = universe of proposals = {p0, p1, p2, …} Assume proposals are independent. A bill is a subset of P.

slide-34
SLIDE 34

Voter vi supports proposal pj iff Ui(pj) > 0 Voter i has utility function Ui(.) Define: pj passes iff majority of voters support pj

slide-35
SLIDE 35

Given two disjoint bills B1, B2: Ui(B1 U B2) = Ui(B1) + Ui(B2) Assumption:

slide-36
SLIDE 36

Given two disjoint bills B1, B2: If Vi supports B1 and Vi supports B2, Then Vi supports B1 U B2 Corollary:

slide-37
SLIDE 37

Given two disjoint bills B1, B2: If B1 passes and B2 passes, Then B1 U B2 passes Theorem:

slide-38
SLIDE 38

Given two disjoint bills B1, B2: If B1 passes and B2 passes, Then B1 U B2 passes Non-Theorem:

slide-39
SLIDE 39

Voter

B1 B2 B1 U B2

Alice

1

  • 2
  • 1

Bob

  • 2

1

  • 1

Charlie

1 1 2

slide-40
SLIDE 40
  • utputs not logically consistent
slide-41
SLIDE 41

Let’s generalize the model …

slide-42
SLIDE 42

partition voters into districts

slide-43
SLIDE 43

partition voters into districts legislature: one rep per district

slide-44
SLIDE 44

partition voters into districts legislature: one rep per district rep supports B iff majority of constituents support B

slide-45
SLIDE 45

implications

slide-46
SLIDE 46

legislator not logically consistent

slide-47
SLIDE 47

legislator not logically consistent supports B1, supports B2

slide-48
SLIDE 48

legislator not logically consistent supports B1, supports B2 might not support B1 U B2

slide-49
SLIDE 49

legislator doesn’t care about the facts

slide-50
SLIDE 50

individual legislators seem logically inconsistent indifferent to truth

slide-51
SLIDE 51

legislative strategy

slide-52
SLIDE 52

Consider the following problem…

slide-53
SLIDE 53

Given: bill B amendments A1, …, An

(mutually disjoint)

Amendment Problem

Can you add a subset of the Ai to B, to make an amended proposal that will pass?

slide-54
SLIDE 54

Given: bill B amendments A1, …, An

(mutually disjoint)

Amendment Problem

Can you add a subset of the Ai to B, to make an amended proposal that will pass?

NP-complete!

slide-55
SLIDE 55

Ted Stevens

Nobody knew this could be so complicated.

slide-56
SLIDE 56

real world: even more complicated

slide-57
SLIDE 57

voters not self-consistent

slide-58
SLIDE 58

legislators make deals

slide-59
SLIDE 59

administrative agencies

slide-60
SLIDE 60

indirectly accountable to voters

slide-61
SLIDE 61

locally, system will look irrational

slide-62
SLIDE 62

It’s complicated.

slide-63
SLIDE 63

what to do?

slide-64
SLIDE 64

“It’s not okay to not know how the Internet works.”

slide-65
SLIDE 65

“It’s not okay to not know how government works.”

slide-66
SLIDE 66

good decisions

slide-67
SLIDE 67
slide-68
SLIDE 68
slide-69
SLIDE 69
slide-70
SLIDE 70

just the facts

slide-71
SLIDE 71
slide-72
SLIDE 72
slide-73
SLIDE 73

dictate the decision

slide-74
SLIDE 74
slide-75
SLIDE 75

Who elected you?

slide-76
SLIDE 76

download your brain

slide-77
SLIDE 77

be useful

slide-78
SLIDE 78

your knowledge + their preferences

slide-79
SLIDE 79

your knowledge + their knowledge and preferences

slide-80
SLIDE 80

get their knowledge and preferences structure the decision space

slide-81
SLIDE 81

Knowledge Knowledge Preferences

slide-82
SLIDE 82

engagement over time mutual trust

slide-83
SLIDE 83

role of our community

slide-84
SLIDE 84

need boots on the ground

slide-85
SLIDE 85

create a career path

slide-86
SLIDE 86

build incentives

slide-87
SLIDE 87

This is important!

slide-88
SLIDE 88

We need to be in the room.

slide-89
SLIDE 89

Machine Learning Meets Public Policy

Edward W. Felten Kahn Professor of Computer Science and Public Affairs Director, Center for Information Technology Policy Princeton University