Slide 1

Slide 1 text

Experts, Crowds + Algorithms Applied to SCOTUS daniel martin katz The Three Forms of (Legal) Prediction blog | ComputationalLegalStudies.com edu | illinois tech - chicago kent law lab | TheLawLab.com page | DanielMartinKatz.com

Slide 2

Slide 2 text

TODAY I WOULD LIKE TO BEGIN WITH A PRACTICAL SET OF LEGAL AND LAW RELATED PROBLEMS

Slide 3

Slide 3 text

AS THESE ARE THE CLASSIC QUESTIONS THAT CLIENTS POSE ON AN ONGOING BASIS …

Slide 4

Slide 4 text

No content

Slide 5

Slide 5 text

HOW MUCH IS THIS GOING TO COST ?

Slide 6

Slide 6 text

HOW LONG WILL THIS MATTER TAKE ?

Slide 7

Slide 7 text

ARE WE GOING TO WIN ? WHAT IS OUR EXPECTED LIABILITY?

Slide 8

Slide 8 text

No content

Slide 9

Slide 9 text

IF YOU REFLECT UPON IT …

Slide 10

Slide 10 text

LAW-LAW LAND HAS MANY PREDICTION PROBLEMS

Slide 11

Slide 11 text

THE ONLY QUESTION IS ON WHAT BASIS THOSE PREDICTIONS WILL BE OFFERED

Slide 12

Slide 12 text

No content

Slide 13

Slide 13 text

I WOULD LIKE TO REMIND EVERYONE AT THE OUTSET

Slide 14

Slide 14 text

BEFORE THERE WERE COMPUTERS

Slide 15

Slide 15 text

HUMANS DID ALL OF THE COMPUTING

Slide 16

Slide 16 text

IF YOU REFLECT UPON OUR OWN DECISION MAKING …

Slide 17

Slide 17 text

HERE ARE JUST A FEW COGNITIVE PROCESSES …

Slide 18

Slide 18 text

LOOK FOR PATTERNS WEIGH VARIABLES MAKE CONCEPTUAL LEAPS (using analogical reasoning)

Slide 19

Slide 19 text

ABSTRACTION OF A PROJECTING WEIGHTS INTO A DECISION f( ) dimension 1 dimension 2 dimension 3 . . . . dimension n OUTPUT (Prediction, Decision, etc.) and / or INPUTS

Slide 20

Slide 20 text

PATTERN MATCHING evolutionary biology is an algorithm which privileges good pattern matching

Slide 21

Slide 21 text

PATTERN MATCHING Biology is why you have trouble with Pie Charts

Slide 22

Slide 22 text

PATTERN MATCHING But are very good at interpreting distances

Slide 23

Slide 23 text

ANALOGICAL REASONING Lawyers are particularly good at this task

Slide 24

Slide 24 text

No content

Slide 25

Slide 25 text

IN ALL OF HUMAN HISTORY …

Slide 26

Slide 26 text

WE HAVE DEVELOPED THREE APPROACHES TO PREDICT OUTCOMES …

Slide 27

Slide 27 text

IN LAW OR MORE BROADLY …

Slide 28

Slide 28 text

THOSE THREE APPROACHES ARE EXPERTS CROWDS ALGORITHMS

Slide 29

Slide 29 text

No content

Slide 30

Slide 30 text

MORE SIMPLY STATED … WE CAN ASK EXPERT CROWD ALGORITHM TO PREDICT SOMETHING

Slide 31

Slide 31 text

IN LAW, BOTH HISTORICALLY AND STILL TODAY HUMAN DECISION MAKING IS THE DOMINANT FORM OF COMPUTATION

Slide 32

Slide 32 text

IN THE PAST DECADE THERE HAS BEEN AT LEAST SOME EXPLORATION AT ALTERNATIVES

Slide 33

Slide 33 text

LETS CALL IT THE ROBOT LAWYER THESIS …?

Slide 34

Slide 34 text

No content

Slide 35

Slide 35 text

OR MORE REALISTICALLY IT SHOULD BE CALLLED THE USE OF ALGORITHMS TO INTERROGATE PATTERNS IN DATA

Slide 36

Slide 36 text

REMEMBER AN ‘ALGORITHM’ IS JUST A RECIPE FOR A MACHINE TO COMPLETE A TASK

Slide 37

Slide 37 text

A.I.

Slide 38

Slide 38 text

ARTIFICIAL INTELLIGENCE

Slide 39

Slide 39 text

THIS IS A TOPIC FOR WHICH I AM OFTEN CALLED UPON TO COMMENT

Slide 40

Slide 40 text

AND FOR WHICH A FULL FLEDGED PRESENTATION COULD BE OFFERED …

Slide 41

Slide 41 text

https://www.slideshare.net/Danielkatz/ artificial-intelligence-and-law-a-primer ACCESS MORE HERE

Slide 42

Slide 42 text

BUT AT A HIGH LEVEL LET ME JUST SAY …

Slide 43

Slide 43 text

ARTIFICIAL INTELLIGENCE IS A BROAD FIELD

Slide 44

Slide 44 text

LETS LOOK AT THESE SPECIFIC SUBFIELDS

Slide 45

Slide 45 text

data driven AI rules based AI

Slide 46

Slide 46 text

No content

Slide 47

Slide 47 text

DATA DRIVEN A.I.

Slide 48

Slide 48 text

DATA DRIVEN A.I. = MACHINE LEARNING NATURAL LANGUAGE PROCESSING

Slide 49

Slide 49 text

No content

Slide 50

Slide 50 text

THE ULTIMATE GOAL IS TO PREDICT SOME CLASS OF LEGAL OUTCOMES

Slide 51

Slide 51 text

HERE ARE JUST A FEW USE CASES IN LAW

Slide 52

Slide 52 text

#Predict Relevant Documents Data Driven EDiscovery/Due Diligence (Predictive Coding)

Slide 53

Slide 53 text

#Predict Relevant Documents Data Driven EDiscovery/Due Diligence (Predictive Coding) #Predict Contract Terms/Outcomes Data Driven Transactional Work

Slide 54

Slide 54 text

#Predict Relevant Documents Data Driven EDiscovery/Due Diligence (Predictive Coding) #Predict Rogue Behavior Data Driven Compliance #Predict Contract Terms/Outcomes Data Driven Transactional Work

Slide 55

Slide 55 text

#Predict Relevant Documents #Predict Case Outcomes / Costs Data Driven Legal Underwriting Data Driven EDiscovery/Due Diligence (Predictive Coding) #Predict Rogue Behavior Data Driven Compliance #Predict Contract Terms/Outcomes Data Driven Transactional Work

Slide 56

Slide 56 text

#Predict Relevant Documents #Predict Case Outcomes / Costs Data Driven Legal Underwriting Data Driven EDiscovery/Due Diligence (Predictive Coding) #Predict Rogue Behavior Data Driven Compliance #Predict Contract Terms/Outcomes Data Driven Transactional Work #Predict Regulatory Outcomes Data Driven Lobbying, etc.

Slide 57

Slide 57 text

No content

Slide 58

Slide 58 text

TODAY I WANT TO TAKE THESE AND OTHER IDEAS AND DISCUSS THEIR APPLICATION TO SCOTUS PREDICTION

Slide 59

Slide 59 text

SCOTUS PREDICTION IS A PASTIME

Slide 60

Slide 60 text

EVERY YEAR, LAW REVIEWS, MAGAZINE AND N E W S PA P E R A R T I C L E S , T E L E V I S I O N A N D RADIO TIME, CONFERENCE PANELS, BLOG P O S T S , A N D T W E E T S A R E D E V O T E D T O QUESTIONS SUCH AS: H O W W I L L T H E C O U R T R U L E I N T H I S PARTICULAR CASE? IN THE DIRECTION OF WHICH PARTY WILL AN INDIVIDUAL JUSTICE VOTE?

Slide 61

Slide 61 text

JUDICIAL PREDICTION IS ALSO THE LAW + POLITICAL SCIENCE GRAIL QUEST

Slide 62

Slide 62 text

YOU COULD START WITH HOLMES AND THE LEGAL REALISTS BUT THESE WERE *NOT* REALLY SCIENTIFIC EFFORTS

Slide 63

Slide 63 text

FRED KORT, PREDICTING SUPREME COURT DECISIONS MATHEMATICALLY: A QUANTITATIVE ANALYSIS OF THE “RIGHT TO COUNSEL” CASES, 51 AMER. POL. SCI. REV. 1 (1957). 1957 S. SIDNEY ULMER, QUANTITATIVE ANALYSIS OF JUDICIAL PROCESSES: SOME PRACTICAL AND THEORETICAL APPLICATIONS, 28 LAW & CONTEMP. PROBS. 164 (1963). 1963 A CO U P L E O F E A R LY E F F O RT S

Slide 64

Slide 64 text

COMPUTERWORLD JULY 1971 PROGRAM WRITTEN IN FORTRAN (THE 91% PREDICTION MARK WAS LIMITED TO CERTAIN CASES) HAROLD SPAETH

Slide 65

Slide 65 text

JEFFREY A. SEGAL, PREDICTING S U P R E M E C O U R T C A S E S PROBABILISTICALLY: THE SEARCH AND SEIZURE CASES, 1962-1981, 78 AMERICAN POLITICAL SCIENCE REVIEW 891 (1984) 1984 A N I M P O RTA N T L AT E R E F F O RT

Slide 66

Slide 66 text

Columbia Law Review (2004) Theodore W. Ruger, Pauline T. Kim, Andrew D. Martin, Kevin M. Quinn Legal and Political Science Approaches to Predicting Supreme Court Decision Making The Supreme Court Forecasting Project: B U T T H I S WA S T H E PA P E R T H AT I N S P I R E D O U R E F F O RT S 2004

Slide 67

Slide 67 text

No content

Slide 68

Slide 68 text

EXPERTS

Slide 69

Slide 69 text

Columbia Law Review October, 2004 Theodore W. Ruger, Pauline T. Kim, Andrew D. Martin, Kevin M. Quinn Legal and Political Science Approaches to Predicting Supreme Court Decision Making The Supreme Court Forecasting Project:

Slide 70

Slide 70 text

experts

Slide 71

Slide 71 text

Case Level Prediction Justice Level Prediction 67.4% experts 58% experts From the 68 Included Cases for the 2002-2003 Supreme Court Term

Slide 72

Slide 72 text

these experts probably overfit

Slide 73

Slide 73 text

they fit to the noise and not the signal

Slide 74

Slide 74 text

No content

Slide 75

Slide 75 text

if this were finance this would be trading worse than S&P500

Slide 76

Slide 76 text

#NoiseTrading

Slide 77

Slide 77 text

#BuffetChallenge

Slide 78

Slide 78 text

#BuffetChallenge

Slide 79

Slide 79 text

like many other forms human endeavor law is full of 
 noise predictors …

Slide 80

Slide 80 text

from a pure forecasting standpoint

Slide 81

Slide 81 text

the best known SCOTUS predictor is

Slide 82

Slide 82 text

No content

Slide 83

Slide 83 text

the law version of superforecasting

Slide 84

Slide 84 text

No content

Slide 85

Slide 85 text

CROWDS

Slide 86

Slide 86 text

not enough crowd based decision making in institutions

Slide 87

Slide 87 text

No content

Slide 88

Slide 88 text

“Software developers were asked on two separate days to estimate the completion time for a given task, the hours they projected differed by 71%, on average. When pathologists made two assessments of the severity of biopsy results, the correlation between their ratings was only .61 (out of a perfect 1.0), indicating that they made inconsistent diagnoses quite frequently. Judgments made by different people are even more likely to diverge.”

Slide 89

Slide 89 text

No content

Slide 90

Slide 90 text

No content

Slide 91

Slide 91 text

FA N TA SY S COT U S I S A S CO O L A S I T S O U N D S

Slide 92

Slide 92 text

FA N TA SY S COT U S WA S F O U N D E D B Y J O S H B L AC K M A N I N 2 0 0 9

Slide 93

Slide 93 text

P R I Z E S A N D S P O N S O R S H I P H AV E VA R I E D B U T T H E R E H AV E B E E N T E N S O F T H O U S A N D S O F $ $ A N D P R I Z E S D I ST R I B U T E D OV E R T H E Y E A R S

Slide 94

Slide 94 text

No content

Slide 95

Slide 95 text

No content

Slide 96

Slide 96 text

U S E R S C R E AT E A LO G I N A N D ACC E S S T H E S I T E

Slide 97

Slide 97 text

O N A CA S E B Y CA S E B A S I S , U S E R S CA N E N T E R T H E I R R E S P E C T I V E P R E D I C T I O N S

Slide 98

Slide 98 text

U S E R S A R E F R E E TO C H A N G E T H E I R P R E D I C T I O N S U N T I L T H E DAT E O F F I N A L D E C I S I O N

Slide 99

Slide 99 text

https://fivethirtyeight.com/features/ obamacares-chances-of-survival-are-looking- better-and-better/ ( S O M E T I M E S I N I N T E R E ST I N G WAY S A S S H O W N A B OV E ) U S E R S A R E F R E E TO C H A N G E T H E I R P R E D I C T I O N S U N T I L T H E DAT E O F F I N A L D E C I S I O N

Slide 100

Slide 100 text

F O R E AC H CA S E , W E A R E A B L E TO T R AC K TO P E R F O R M A N C E O F P L AY E R S A N D CO M PA R E I T TO T H E O U TCO M E O F T H E CA S E S

Slide 101

Slide 101 text

We can generate Crowd Sourced Predictions

Slide 102

Slide 102 text

No content

Slide 103

Slide 103 text

S U M M A R Y STAT I ST I C S

Slide 104

Slide 104 text

6 S COT U S T E R M S

Slide 105

Slide 105 text

4 2 5 L I ST E D CA S E S 6 S COT U S T E R M S

Slide 106

Slide 106 text

7 2 8 4 U N I Q U E PA RT I C I PA N T S 4 2 5 L I ST E D CA S E S 6 S COT U S T E R M S

Slide 107

Slide 107 text

7 2 8 4 U N I Q U E PA RT I C I PA N T S 4 2 5 L I ST E D CA S E S 6 3 6 8 5 9 P R E D I C T I O N S 6 S COT U S T E R M S

Slide 108

Slide 108 text

W E H AV E A S I G N I F I CA N T A M O U N T O F P L AY E R T U R N OV E R WO R K I N G W I T H R E A L DATA ( ~ 3 % O F M A X PA RT I C I PAT I O N )

Slide 109

Slide 109 text

S O M E T I M E S F O L K S C H A N G E T H E I R VOT E S 6 3 6 8 5 9 5 4 5 8 4 5 F I N A L P R E D I C T I O N S OV E R A L L P R E D I C T I O N S WO R K I N G W I T H R E A L DATA

Slide 110

Slide 110 text

T H E N U M B E R O F P L AY E R S H A S D E C L I N E D B U T T H E E N G AG E M E N T R AT E H A S I N C R E A S E D WO R K I N G W I T H R E A L DATA

Slide 111

Slide 111 text

W E B E L I E V E T H I S I S * N OT * R E A L LY S U RV I VO R S H I P B I A S B U T R AT H E R A R E V E L AT I O N M E C H A N I S M ( I . E . YO U ST I C K A R O U N D I F T H I N K YO U A R E G O O D AT T H E U N D E R LY I N G TA S K )

Slide 112

Slide 112 text

S U M M A R Y DATA

Slide 113

Slide 113 text

No content

Slide 114

Slide 114 text

CROWD MODELING PRINCIPLES

Slide 115

Slide 115 text

C R O W D S O U R C I N G C R O W D S O U R C I N G D O E S * N OT * R E F E R TO A S P E C I F I C T E C H N I Q U E O R A LG O R I T H M

Slide 116

Slide 116 text

C R O W D S O U R C I N G G E N E R A L LY R E F E R S TO A P R O C E S S O F AG G R E G AT I O N A N D / O R S E G M E N TAT I O N O F I N F O R M AT I O N S I G N A L S

Slide 117

Slide 117 text

VA R I O U S S I G N A L TY P E S T H E I N P U T S I G N A L S CA N A S S U M E M A N Y D I F F E R E N T F O R M S I N C LU D I N G F R O M M O D E L S O R I N D I V I D UA L S O R S E N S O R S ( O R S O M E CA S E S E V E N OT H E R C R O W D S )

Slide 118

Slide 118 text

CROWD OF INDIVIDUALS The most well know approach involves extracting ‘wisdom’ from crowds where crowds are built from individual people

Slide 119

Slide 119 text

CROWD OF SENSORS Note crowds need not be composed of humans but could be networked IT systems Decentralized Distributed Ledgers -or- Oracles -or- IOT sensors with Crowdsourcing Validation #Blockchain #InternetofThings #Crypto

Slide 120

Slide 120 text

#CRYPTO CROWD Thanks to Team Augur for the Shoutout

Slide 121

Slide 121 text

Random Forest Model Breiman, L.(2001). Random forests. Machine learning, 45(1), 5-32. Grow a set of differentiated trees through bagging and random substrates (predict using a consensus mechanism) C R O W D O F M O D E L S

Slide 122

Slide 122 text

No content

Slide 123

Slide 123 text

A S W E R E V I E W E D T H E C R O W D S O U R C I N G L I T E R AT U R E …

Slide 124

Slide 124 text

W E O B S E RV E D T H AT I T WA S D I F F I C U LT TO A P P LY T H E P R I N C I P L E S TO C R O W D S S U C H A S O U R S

Slide 125

Slide 125 text

C R O W D S O U R C I N G I S ‘ U N D I S C I P L I N E D ZO O O F M O D E L S ’ J E S S I CA F L AC K P R O F E S S O R S A N TA F E I N ST I T U T E D E C . 2 7 , 2 0 1 7 ( V I A T W I T T E R )

Slide 126

Slide 126 text

W E AG R E E … A N D T H U S I N T H E PA P E R W E J U ST STA RT E D OV E R …

Slide 127

Slide 127 text

A N D AT T E M P T E D TO B U I L D C R O W D S F R O M F I R ST P R I N C I P L E S …

Slide 128

Slide 128 text

No content

Slide 129

Slide 129 text

CROWD CONSTRUCTION FRAMEWORK

Slide 130

Slide 130 text

W E O U T L I N E A G E N E R A L F R A M E WO R K F O R CO N ST R U C T I N G C R O W D S F R O M F I R ST P R I N C I P L E S

Slide 131

Slide 131 text

I N T H E C L A S S I C CO N D O R C E T J U R Y S E T T I N G , M O D E L S TY P I CA L LY U S E P R E D I C T I O N S F R O M A L L PA RT I C I PA N T S

Slide 132

Slide 132 text

H O W E V E R , M O D E L S CA N A L S O TA K E I N TO ACCO U N T I N F O R M AT I O N ( S I G N A L S ) F R O M S O M E S U B S E T O F PA RT I C I PA N T S ( D E F I N E D U S I N G E I T H E R I N C LU S I O N R U L E S O R E XC LU S I O N R U L E S )

Slide 133

Slide 133 text

No content

Slide 134

Slide 134 text

E X P E R I E N C E P E R F O R M A N C E R A N K STAT I ST I CA L T H R E S H O L D I N G W E I G H T I N G C R O W D CO N ST R U C T I O N R U L E S

Slide 135

Slide 135 text

C R O W D CO N ST R U C T I O N R U L E S T H E I N T E R AC T I O N O F T H E S E R U L E S Y I E L D S * N OT * A N I N D I V I D UA L M O D E L B U T R AT H E R A M O D E L S PAC E

Slide 136

Slide 136 text

T H E M O D E L S PAC E M O D E L S PAC E F E AT U R E S 2 7 7 , 2 0 1 P OT E N T I A L M O D E L S

Slide 137

Slide 137 text

T H E M O D E L S PAC E 1 + ( 2 8 · 9 9 · 1 0 0 ) = 2 7 7 2 0 1 F I R ST T E R M I S T H E S I M P L E ST C R O W D S O U R C I N G M O D E L W I T H N O S U B S E T O R W E I G H T I N G R U L E S S E CO N D T E R M CO R R E S P O N D S TO 2 8 M O D E L S ( CO M B I N AT I O N O F P E R F O R M A N C E T H R E S H O L D / W E I G H T I N G R U L E S ) F O R E AC H CO M B I N AT I O N O F 9 9 R A N K A N D 1 0 0 E X P E R I E N C E T H R E S H O L D S

Slide 138

Slide 138 text

No content

Slide 139

Slide 139 text

MODEL TESTING & RESULTS

Slide 140

Slide 140 text

W E S I M U L AT E T H E P E R F O R M A N C E O F * E AC H * O F T H E 2 7 7 , 2 0 1 P OT E N T I A L C R O W D M O D E L S M O D E L ( S ) ACC U R ACY

Slide 141

Slide 141 text

A LT H O U G H I T I S A L A R G E M O D E L S PAC E W E D O H I G H L I G H T T H E P E R F O R M A N C E O F F O U R E X A M P L E M O D E L S ( A N D T H E N U L L M O D E L )

Slide 142

Slide 142 text

B A S E L I N E A LWAY S G U E S S R E V E R S E N U L L M O D E L

Slide 143

Slide 143 text

M O D E L 1 A L L C R O W D S I M P L E M A J O R I TY

Slide 144

Slide 144 text

M O D E L 2 F O L LO W T H E L E A D E R W I T H N O T H R E S H O L D I N G

Slide 145

Slide 145 text

M O D E L 3 F O L LO W T H E L E A D E R W I T H E X P E R I E N C E T H R E S H O L D I N G ( X P = 5 )

Slide 146

Slide 146 text

M O D E L 4 M A X I M U M ACC U R ACY ( T H E R E A R E AC T UA L LY S E V E R A L M O D E L CO N F I G U R AT I O N S W H I C H O F F E R R O U G H LY E Q U I VA L E N T P E R F O R M A N C E ) E X P E R I E N C E T H R E S H O L D O F 5 C R O W D S I Z E I S CA P P E D AT 2 2 E X P O N E N T I A L W E I G H T W I T H A L P H A O F 0 . 1

Slide 147

Slide 147 text

CASE LEVEL CUMULATIVE ACCURACY

Slide 148

Slide 148 text

JUSTICE LEVEL CUMULATIVE ACCURACY

Slide 149

Slide 149 text

DISTRIBUTION OF JUSTICE LEVEL MODEL ACCURACY

Slide 150

Slide 150 text

No content

Slide 151

Slide 151 text

W E S I M U L AT E T H E P E R F O R M A N C E O F * E AC H * O F T H E 2 7 7 , 2 0 1 P OT E N T I A L C R O W D M O D E L S R O B U ST N E S S O F P E R F O R M A N C E

Slide 152

Slide 152 text

ROBUSTNESS VISUALIZED T H I S I S A L L R E L AT I V E TO T H E N U L L M O D E L ( O F A LWAY S G U E S S R E V E R S E )

Slide 153

Slide 153 text

ROBUSTNESS VISUALIZED T H E CO N TO U R P LOT F L AT T E N S T H E D I M E N S I O N A L I TY O F T H E S PAC E ( E AC H C E L L I S T H E AV E R AG E M O D E L P E R F O R M A N C E OV E R A L L OT H E R M O D E L PA R A M E T E R AT E AC H E X P E R I E N C E , R A N K CO M B O )

Slide 154

Slide 154 text

ROBUSTNESS VISUALIZED J U ST I C E L E V E L CA S E L E V E L

Slide 155

Slide 155 text

T H E K E Y I D E A

Slide 156

Slide 156 text

N OT A L L M E M B E R S O F C R O W D A R E M A D E E Q UA L

Slide 157

Slide 157 text

W E M A I N TA I N A ‘ S U P E R C R O W D ’ W H I C H I S T H E TO P N O F P R E D I C TO R S U P TO T I M E T- 1

Slide 158

Slide 158 text

the ‘supercrowd’ outperforms the overall crowd (and even the best single player)

Slide 159

Slide 159 text

H T T P S : / / A R X I V. O RG / A B S / 1712 . 0 3 84 6 H T T P S : / / PA P E R S . S S R N . C O M / S O L 3 / PA P E R S . C F M ? A B S T R AC T _ I D = 3 0 8 5710

Slide 160

Slide 160 text

No content

Slide 161

Slide 161 text

ALGORITHMS

Slide 162

Slide 162 text

http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0174698 Katz DM, Bommarito MJ II, Blackman J (2017), A General Approach for Predicting the Behavior of the Supreme Court of the United States. PLoS ONE 12(4): e0174698.

Slide 163

Slide 163 text

Our algorithm is a special version of random forest (time evolving) http://journals.plos.org/ plosone/article?id=10.1371/ journal.pone.0174698 available at RESEARCH ARTICLE A general approach for predicting the behavior of the Supreme Court of the United States Daniel Martin Katz1,2*, Michael J. Bommarito II1,2, Josh Blackman3 1 Illinois Tech - Chicago-Kent College of Law, Chicago, IL, United States of America, 2 CodeX - The Stanford Center for Legal Informatics, Stanford, CA, United States of America, 3 South Texas College of Law Houston, Houston, TX, United States of America * [email protected] Abstract Building on developments in machine learning and prior work in the science of judicial pre- diction, we construct a model designed to predict the behavior of the Supreme Court of the United States in a generalized, out-of-sample context. To do so, we develop a time-evolving random forest classifier that leverages unique feature engineering to predict more than 240,000 justice votes and 28,000 cases outcomes over nearly two centuries (1816-2015). Using only data available prior to decision, our model outperforms null (baseline) models at both the justice and case level under both parametric and non-parametric tests. Over nearly two centuries, we achieve 70.2% accuracy at the case outcome level and 71.9% at the jus- tice vote level. More recently, over the past century, we outperform an in-sample optimized null model by nearly 5%. Our performance is consistent with, and improves on the general level of prediction demonstrated by prior work; however, our model is distinctive because it can be applied out-of-sample to the entire past and future of the Court, not a single term. Our results represent an important advance for the science of quantitative legal prediction and portend a range of other potential applications. Introduction As the leaves begin to fall each October, the first Monday marks the beginning of another term for the Supreme Court of the United States. Each term brings with it a series of challenging, important cases that cover legal questions as diverse as tax law, freedom of speech, patent law, administrative law, equal protection, and environmental law. In many instances, the Court’s decisions are meaningful not just for the litigants per se, but for society as a whole. Unsurprisingly, predicting the behavior of the Court is one of the great pastimes for legal and political observers. Every year, newspapers, television and radio pundits, academic jour- nals, law reviews, magazines, blogs, and tweets predict how the Court will rule in a particular case. Will the Justices vote based on the political preferences of the President who appointed them or form a coalition along other dimensions? Will the Court counter expectations with an unexpected ruling? PLOS ONE | https://doi.org/10.1371/journal.pone.0174698 April 12, 2017 1 / 18 a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 OPEN ACCESS Citation: Katz DM, Bommarito MJ, II, Blackman J (2017) A general approach for predicting the behavior of the Supreme Court of the United States. PLoS ONE 12(4): e0174698. https://doi. org/10.1371/journal.pone.0174698 Editor: Luı ´s A. Nunes Amaral, Northwestern University, UNITED STATES Received: January 17, 2017 Accepted: March 13, 2017 Published: April 12, 2017 Copyright: © 2017 Katz et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. Data Availability Statement: Data and replication code are available on Github at the following URL: https://github.com/mjbommar/scotus-predict-v2/. Funding: The author(s) received no specific funding for this work. Competing interests: All Authors are Members of a LexPredict, LLC which provides consulting services to various legal industry stakeholders. We received no financial contributions from LexPredict or anyone else for this paper. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

Slide 164

Slide 164 text

T H E S O U R C E CO D E F O R O U R A LG O PA P E R I S AVA I L A B L E O N

Slide 165

Slide 165 text

No content

Slide 166

Slide 166 text

W E CA L L O U R A LG O PA P E R A ‘ G E N E R A L’ A P P R OAC H

Slide 167

Slide 167 text

B E CAU S E W E A R E N OT I N T E R E ST E D I N A LO CA L LY T U N E D M O D E L B U T R AT H E R A M O D E L T H AT CA N ‘ STA N D T H E T E ST O F T I M E ’

Slide 168

Slide 168 text

G E N E R A L S COT U S P R E D I C T I O N 243,882 28,009 Case Outcomes Justice Votes 1816-2015

Slide 169

Slide 169 text

G E N E R A L S COT U S P R E D I C T I O N 70.2% accuracy at the case outcome level 71.9% at the justice vote level 1816-2015

Slide 170

Slide 170 text

W E P R E D I C T * N OT * A S I N G L E Y E A R B U T R AT H E R ~ 2 0 0 Y E A R S ( 1 8 1 6 - 2 0 1 5 ) O U T O F S A M P L E

Slide 171

Slide 171 text

N O W I T I S WO RT H N OT I N G T H AT P R E D I C T I O N O R I E N T E D PA P E R S A R E C U R R E N T LY S W I M M I N G AG A I N ST M A I N ST R E A M S O C I A L S C I E N C E ( A N D L AW )

Slide 172

Slide 172 text

CAU S A L I N F E R E N C E I S T H E H A L L M A R K O F M O ST Q UA N T O R I E N T E D L AW + S O C I A L S C I E N C E S C H O L A R S H I P

Slide 173

Slide 173 text

I T I S B E ST S U I T E D TO P O L I CY E VA LUAT I O N ( S U C H A S D O E S T H I S PA RT I C U L A R P O L I CY I N T E RV E N T I O N AC H I E V E I T S STAT E D O B J E C T I V E S )

Slide 174

Slide 174 text

O R I N STA N C E S W H E R E E STA B L I S H I N G L I N K S B E T W E E N CAU S E A N D E F F E C T A R E C R I T I CA L

Slide 175

Slide 175 text

B U T T H E R E I S A N A LT E R N AT I V E PA R A D I G M # P R E D I C T I O N

Slide 176

Slide 176 text

M AC H I N E L E A R N I N G P R E D I C T I V E A N A LY T I C S ‘ I N V E R S E ’ P R O B L E M B -S C H O O L CO M P S C I P H Y S I C S P R E D I C T I O N

Slide 177

Slide 177 text

Andrew D. Martin, Kevin M. Quinn, Theodore W. Ruger & Pauline T. Kim, Competing Approaches to Predicting Supreme Court Decision Making, 2 Perspectives on Politics 761 (2004). “the best test of an explanatory theory is its ability to predict future events. To the extent that scholars in both disciplines (social science and law) seek to explain court behavior, they ought to test their theories not only against cases already decided, but against future outcomes as well.”

Slide 178

Slide 178 text

https://www.computationallegalstudies.com/2017/08/28/legal-analytics-versus-empirical- legal-studies-causal-inference-vs-prediction/ https://www.slideshare.net/Danielkatz/legal-analytics-versus-empirical- legal-studies-or-causal-inference-vs-prediction-redux M O R E O N T H AT TO P I C H E R E

Slide 179

Slide 179 text

No content

Slide 180

Slide 180 text

T H E R E I S G R O W I N G I N T E R E ST I N T H E P R E D I C T I O N C E N T R I C A P P R OAC H

Slide 181

Slide 181 text

“There are two cultures in the use of statistical modeling to reach conclusions from data. One assumes that the data are generated by a given stochastic data model. The other uses algorithmic models and treats the data mechanism as unknown …. If our goal as a field is to use data to solve problems, then we need to move away from exclusive dependence on data models and adopt a more diverse set of tools.” Leo Breiman, Statistical modeling: The two cultures (with comments and a rejoinder by the author), 16 Statistical Science 199 (2001) Note: Leo Breiman Invented Random Forests

Slide 182

Slide 182 text

No content

Slide 183

Slide 183 text

No content

Slide 184

Slide 184 text

No content

Slide 185

Slide 185 text

No content

Slide 186

Slide 186 text

No content

Slide 187

Slide 187 text

Susan Athey, The Impact of Machine Learning on Economics http://www.nber.org/chapters/c14009.pdf

Slide 188

Slide 188 text

3 5 5 S C I E N C E 6 3 2 4 3 F E B R UA R Y 2 0 1 7 S P E C I A L I S S U E O N P R E D I C T I O N

Slide 189

Slide 189 text

No content

Slide 190

Slide 190 text

EXPERTS, CROWDS, ALGORITHMS

Slide 191

Slide 191 text

P R E D I C T I O N I S N OT N E C E S S A R I LY # M L A LO N E B U T R AT H E R S O M E E N S E M B L E O F E X P E RT S , C R O W D S + A LG O R I T H M S

Slide 192

Slide 192 text

http://www.sciencemag.org/news/ 2017/05/artificial-intelligence-prevails- predicting-supreme-court-decisions Professor Katz noted that in the long term …“We believe the blend of experts, crowds, and algorithms is the secret sauce for the whole thing.” May 2nd 2017

Slide 193

Slide 193 text

IN MANY INSTANCES BLENDS OF INTELLIGENCE WILL OUTPERFORM A SINGLE STREAM OF INTELLIGENCE

Slide 194

Slide 194 text

THE PSEUDOCODE OF OUR TIMES …

Slide 195

Slide 195 text

HUMANS + MACHINES HUMANS OR MACHINES >

Slide 196

Slide 196 text

crowd forecast learning problem is to discover how to blend streams of intelligence algorithm forecast ensemble method ENSEMBLE MODEL we can use machine learning methods and metadata such as case topic, lower court as well as crowd metadata to ‘learn’ the conditional weights to apply to the input signals

Slide 197

Slide 197 text

No content

Slide 198

Slide 198 text

THERE ARE LOTS OF INTERESTING APPLICATIONS ONCE YOU CAN PREDICT SOMETHING …

Slide 199

Slide 199 text

AND BY PREDICT … I AM TALKING ABOUT PREDICTION AT THE PORTFOLIO LEVEL …

Slide 200

Slide 200 text

I HAVE BEEN VERY INTERESTED IN THE OVERLAP BETWEEN LEGAL TECH FIN TECH

Slide 201

Slide 201 text

Fin(Legal)Tech Conference finlegaltechconference.com @Illinois Tech - Chicago Kent College of Law

Slide 202

Slide 202 text

https://computationallegalstudies.com/2017/10/24/10-legal-tech-lessons-dollars-doughnuts-fin-legal-tech-via-aba-journal/

Slide 203

Slide 203 text

nearly 75+ videos and counting TheLawLabChannel.com

Slide 204

Slide 204 text

No content

Slide 205

Slide 205 text

Law on the Market

Slide 206

Slide 206 text

When we would present this work on #SCOTUS Prediction folks would ask us “why do I care about marginal improvements in prediction ? “

Slide 207

Slide 207 text

Well at a very minimum — if you could predict the cases you could perhaps trade on them in the relevant securities market …

Slide 208

Slide 208 text

In other words, given our ability to offer forecasts of judicial outcomes, we wondered if this information could inform an event driven trading strategy ?

Slide 209

Slide 209 text

http://arxiv.org/abs/1508.05751 available at http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2649726

Slide 210

Slide 210 text

We call this idea “Law on the Market” LOTM

Slide 211

Slide 211 text

A Motivating Example Myriad Genetics NASDAQ: MYGN Market Cap of ~$3 billion+

Slide 212

Slide 212 text

Myraid Genetics “Myriad employs a number of proprietary technologies that permit doctors and patients to understand the genetic basis of human disease and the role that genes play in the onset, progression and treatment of disease.”

Slide 213

Slide 213 text

Myraid Genetics “Myriad was the subject of scrutiny after it became involved in a lengthy lawsuit over its controversial patenting practices” which including the patenting of human gene sequences ....

Slide 214

Slide 214 text

No content

Slide 215

Slide 215 text

June 13, 2013 Supreme Court Offers this Decision ~10:05am

Slide 216

Slide 216 text

Initial Media Reports and Initial Trading 11:48am

Slide 217

Slide 217 text

Initial Media Reports Early Afternoon “In early afternoon trading Thursday, Myriad shares were up 5.4 percent, or $2.36, at $35.73.”

Slide 218

Slide 218 text

Final Media Reports

Slide 219

Slide 219 text

Final Media Reports

Slide 220

Slide 220 text

9:30am Open

Slide 221

Slide 221 text

10:00am SCOTUS

Slide 222

Slide 222 text

11:05am MYGN Trades UP

Slide 223

Slide 223 text

2:15pm MYGN is Off its Daily Peak but still up

Slide 224

Slide 224 text

Day 1 Close MYGN is Off Nearly 10% from Open and 20% from Daily High

Slide 225

Slide 225 text

Day 2 the Sell Off Continues

Slide 226

Slide 226 text

A Good Time to Buy an Option :)

Slide 227

Slide 227 text

No content

Slide 228

Slide 228 text

No content

Slide 229

Slide 229 text

SO THESE EXAMPLES REPRESENT A FORM OF EXISTENCE PROOF …

Slide 230

Slide 230 text

BUT PERHAPS THEY ARE RARE AND ANACHRONISTIC CASES …?

Slide 231

Slide 231 text

No content

Slide 232

Slide 232 text

ONE OBVIOUS CHALLENGE IS THE PROSPECT THAT THIS INFORMATION IS ALREADY INCORPORATED INTO THE PRICE OF THE RELEVANT SECURITY #EfficientMarketHypothesis #Fama #EMH

Slide 233

Slide 233 text

IN ALLIED FIELDS OF HUMAN ENDEAVOR, THERE ARE FAIRLY RAPID MARKET RESPONSES TO CHANGES IN THE INFORMATION ENVIRONMENT

Slide 234

Slide 234 text

THIS ALL PRESUPPOSES A RIGOROUS INFORMATION AND MODELING ENVIRONMENT — THAT IS GENERALLY LACKING IN QUESTIONS OF LEGAL PREDICTION #QuantitativeLegalPrediction #LegalAnalytics #FinLegalTech

Slide 235

Slide 235 text

No content

Slide 236

Slide 236 text

BUT PERHAPS THEY ARE RARE AND ANACHRONISTIC CASES …?

Slide 237

Slide 237 text

No content

Slide 238

Slide 238 text

Theoretical + Empirical Questions

Slide 239

Slide 239 text

Market Incorporation Hypothesis Are judicial decisions already reflected in the share price ? (If this were true - we would rarely see market move post decision)

Slide 240

Slide 240 text

How General Are These Specific Examples? Theoretical + Empirical Questions (In other words, is this a general phenomenon ?)

Slide 241

Slide 241 text

What is the nature of the signal incorporation environment ? (In other words, what are the dynamics associated with does the market response ?) Theoretical + Empirical Questions

Slide 242

Slide 242 text

No content

Slide 243

Slide 243 text

METHODS

Slide 244

Slide 244 text

(1) Coding / PreProcessing (2) Candidate LOTM Events (3) Formal Evaluation Using CAPM (market model of returns) (4) Evaluate Speed of Incorporation and Related Informational Dynamics

Slide 245

Slide 245 text

(1) Coding / PreProcessing We reviewed and coded 1,363 total cases decided over the period in questions. We asked a simple question - could this case plausibly impact a publicly traded security ?

Slide 246

Slide 246 text

All Data & Code is Available Here^ ^Other than the WRDS Data which is *not* open source but can be obtained from Wharton https://github.com/mjbommar/law-on-the-market https://wrds-web.wharton.upenn.edu/wrds/

Slide 247

Slide 247 text

(3) Formal Evaluation Using CAPM (market model of returns)

Slide 248

Slide 248 text

Abnormal Returns Common approach is to use index as baseline and seek to identify statistically significant deviations from that baseline We want to isolate the effect of the event from other general market movements

Slide 249

Slide 249 text

This Paper Leverages 5 Minute Data -5 Days, +5 Days A KEY POINT much higher frequency than most papers in literature

Slide 250

Slide 250 text

No content

Slide 251

Slide 251 text

SOME RESULTS

Slide 252

Slide 252 text

Summary Results

Slide 253

Slide 253 text

Market Cap

Slide 254

Slide 254 text

Some Additional Cases

Slide 255

Slide 255 text

(4) Evaluate Speed of Incorporation and Related Informational Dynamics

Slide 256

Slide 256 text

Speed of Information Incorporation

Slide 257

Slide 257 text

This is very slow … Perhaps the real action is in the options market ?

Slide 258

Slide 258 text

In conclusion, we believe that this research raises many questions and justifies a range of future work in the area

Slide 259

Slide 259 text

Future Work Real Trading Strategy Analysis Other Classes of Litigation Events 8k’s and Docket Arbitrage Higher and Lower Order Analysis Litigation Reserves, etc.

Slide 260

Slide 260 text

Litigation Funding

Slide 261

Slide 261 text

Other Classes of Litigation Events Litigation Funding and Reserves Litigation Reserves

Slide 262

Slide 262 text

FT Big Read Feb 7, 2019

Slide 263

Slide 263 text

No content

Slide 264

Slide 264 text

TODAY I HAVE GIVEN YOU AN OVERVIEW OF THREE APPROACHES TO LEGAL PREDICTION

Slide 265

Slide 265 text

AND DEMONSTRATED THEIR APPLICATION TO ONE PARTICULAR PROBLEM (#SCOTUS)

Slide 266

Slide 266 text

HOWEVER, I HOPE IT IS CLEAR THAT THESE TECHNIQUES CAN BE APPLIED MORE BROADLY

Slide 267

Slide 267 text

TO A WIDE RANGE OF PREDICTION PROBLEMS ACROSS THE LEGAL INDUSTRY AND BEYOND …

Slide 268

Slide 268 text

No content

Slide 269

Slide 269 text

Daniel Martin Katz @ computational computationallegalstudies.com danielmartinkatz.com illinois tech - chicago kent college of law @ thelawlab.com