Professional Documents
Culture Documents
2020-06-29 AI in Law A Primer - Dan Katz
2020-06-29 AI in Law A Primer - Dan Katz
page | DanielMartinKatz.com
corp | elevateservices.com
lab | TheLawLab.com
the conversation continues online
#BuceriusLegalTech
ARTIFICIAL
INTELLIGENCE -
AN INTRODUCTION
THE VAST MAJORITY
OF #LEGALTECH
PRODUCTS PURPORT
TO LEVERAGE
ARTIFICIAL
INTELLIGENCE
(AI)
SO I THINK IT WOULD
MAKE SENSE TO REALLY
EXPLORE AI
BOTH IN GENERAL
AND WITHIN LAW
BECAUSE MANY OF
THE APPLICATIONS
WITHIN LAW TRADE
ON CONCEPTS
DEVELOPED
OUTSIDE OF LAW
SO WE WILL START
WITH AI OUTSIDE
OF LAW
THEN WE WILL TALK
ABOUT AI+LAW
ARTIFICIAL
INTELLIGENCE IS ALL
AROUND US …
INDEED -
THE PAST FEW YEARS
HAVE WITNESSED SOME
PRETTY IMPRESSIVE
DISPLAYS …
NOT JUST THESE …
BUT ALSO THESE …
Beat the best
human players at
Texas Hold 'Em
poker
Detected Crop
Disease
Identified diabetic
retinopathy (a
leading cause of
blindness) from
retinal photos
Spotted cancer in
tissue slides better
than human
epidemiologists
Wrote sports
articles for the
Associated Press
Wrote its own
machine learning
software
and
Painted a pretty
good van Gogh
MORE EXAMPLES HERE …
http://www.businessinsider.com/artificial-intelligence-ai-most-impressive-
achievements-2017-3#what-ai-cando-everyday-humanstuff-1
IT SEEMS LIKE EVERY
COMPANY IS RACING
TO EMBED AI INTO
THEIR PRODUCTS /
SERVICES
THERE ARE AT LEAST
TWO REASONS WHY AN
INDIVIDUAL LAWYER,
LAW FIRM OR LEGAL
DEPARTMENT …
SHOULD BE INTERESTED IN
RECENT DEVELOPMENTS IN
ARTIFICIAL INTELLIGENCE
(1)
THIS IS WHAT THE
WORLD IS
BECOMING
AND BY EXTENSION
THIS WHAT YOUR
CLIENT’S WORLD
IS BECOMING …
(2)
MUCH LIKE THE REST
OF THE BUSINESS
WORLD, THE
DELIVERY OF LEGAL
SERVICES ARE BEING
IMPACTED BY A.I.
IT IS IMPORTANT TO
UNDERSTAND HOW TO
LEVERAGE SUCH TOOLS
TO HELP DELIVER VALUE
A.I. HISTORY
AND OVERVIEW
I WOULD LIKE TO
REMIND EVERYONE
AT THE OUTSET
BEFORE THERE
WERE COMPUTERS
HUMANS DID ALL OF
THE COMPUTING
IF YOU REFLECT
UPON OUR OWN
DECISION MAKING …
WHAT DO WE DO?
LOOK FOR PATTERNS
WEIGH VARIABLES
dimension 1
dimension 2
f( ) OUTPUT
dimension 3 (Prediction, Decision, etc.)
.
. and / or
.
.
dimension n
PATTERN MATCHING
Alan Turing
“BEFORE 1949, ‘COMPUTERS LACKED A
KEY PREREQUISITE FOR INTELLIGENCE:
THEY COULDN’T STORE COMMANDS,
ONLY EXECUTE THEM …”
http://sitn.hms.harvard.edu/flash/2017/history-artificial-intelligence/
WE HAVE HAD A RANGE
OF FALSE STARTS AND
A.I. WINTERS
EVEN SOME OF THE
WORLD’S LEADING
EXPERTS HAVE
GOTTEN THINGS
WRONG …
“MACHINES WILL BE CAPABLE,
WITHIN TWENTY YEARS, OF DOING
ANY WORK THAT A MAN CAN DO.”
– Herbert Simon
in 1965
“MACHINES WILL BE CAPABLE,
WITHIN TWENTY YEARS, OF DOING
ANY WORK THAT A MAN CAN DO.”
– Herbert Simon
in 1965
IT TURNS OUT THAT
LOTS OF THE IDEAS
THAT WE ARE
LEVERAGING TODAY
WERE ACTUALLY
DEVELOPED MANY
DECADES AGO …
THERE IS NO DOUBT
THIS IS A HYPE
CYCLE OF SORTS …
BUT IT IS ALSO POSSIBLE
(AND IN MY VIEW LIKELY)
THAT THIS TIME WILL BE
DIFFERENT (IN THE
MEDIUM TERM)
SO WHAT IS
POWERING
THIS
A.I.
REVOLUTION?
INCREASING
COMPUTING
POWER
DECREASING
DATA STORAGE
COSTS
Moore’s law
!
Kryder’s Law
is the other half
of the story …
Kryder’s law
!
Today’s AI is
all (mostly) about
Prediction
Some Key Ideas
About
Prediction
(1) Inverse Problem
Or Both
(2) System Dynamics
Imagine Two Different
Complex Systems
Weather
Tides
vs.
TIDES ALMANAC
Easy/ Predictable Difficult / Chaotic
Formal Treatment of the
question of prediction in
alternative Domains
THE DIVISION BELL
IN ARTIFICIAL
INTELLIGENCE
DATA VS. RULES
THE DIVISION BELL
IN ARTIFICIAL
INTELLIGENCE
ARTIFICIAL INTELLIGENCE IS A BROAD FIELD
COMPETING
METHODOLOGICAL
ORIENTATIONS IN
ARTIFICIAL INTELLIGENCE
RICHARD SUSSKIND
DEVELOPED THE FIRST
EXPERT SYSTEM IN
LAW IN THE 1980’S
“In artificial intelligence, an expert system is a computer system that emulates
the decision-making ability of a human expert.
The first expert systems were created in the 1970s and then proliferated in the
1980s.
Expert systems were among the first truly successful forms of artificial
intelligence (AI) software.
However, some experts point out that expert systems were not part of true
artificial intelligence since they lack the ability to learn autonomously
from external data.”
JUST SOME OF THE
CHALLENGES FOR EXPERT SYSTEMS
Knowledge Acquisition Problem
how do I get the information that is being leveraged by the
Human Reasoner ? (including informal rules that the expert has
difficulty expressing)
2005 - Present
rules based A.I. < data driven A.I.
ULTIMATELY WE ARE TRYING TO LEARN
THE RULES / DYNAMICS THAT
UNDERLIE SOME CLASS OF ACTIVITY
WITH THAT UNDERSTANDING
WE WANT TO BE ABLE TO
MIMIC / PREDICT
WHAT ARE SOME OF THE
RULES AND DATA HERE?
TyMetrix/ELM -
Using $80 billion+ in Legal
Spend Data to Help GC’s
Look for Arbitrage
Opportunities, Value
Propositions in Hiring Law
Firms
#LegalSpendAnalytics
Quantitative Legal Prediction
#NegotiationAnalytics
Quantitative Legal Prediction
“Lawyers say the real value in mediation and arbitration might in the future
come from large-scale data analysis of arbitrators and mediators themselves,
in an effort to predict outcomes and potentially affect the course of
settlements … Matthew Saunders, partner at Ashurst, notes that data
analytics “could be extended to predicting which way arbitrators or a
mediator might go”.
IN ADDITION, THERE
ARE AN INCREASING
NUMBER OF
RELEVANT ACADEMIC
PAPERS
(AND IN SOME CASES
ASSOCIATED COMPANIES)
“…I study choice of law by
analyzing the nearly 1,000,000
contracts that have been disclosed
to the Securities and Exchange
Commission between 1996–2012.”
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.
http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0174698
THIS IS SOMETIMES
CALLED ‘PREDICTIVE
ANALYTICS’
WHICH IN OUR FIELD
IS THEN CALLED
‘LEGAL ANALYTICS’
I JUST WANT TO
REMIND YOU THIS IS
NOT ABOUT
PERFECTION BUT
COMPARATIVE
PREDICTIVE
PERFORMANCE
TWO MAJOR FORMS
OF MACHINE
LEARNING
UNSUPERVISED
MACHINE LEARNING
SUPERVISED
MACHINE LEARNING
CLUSTERING
IS AN EXAMPLE OF
UNSUPERVISED
MACHINE LEARNING
CLUSTERING IS ABOUT
PUTTING ‘SIMILAR’
ITEMS TOGETHER
KEY IDEA IS THAT WE
NEED TO CONVERT A
COLLOQUIAL NOTION
OF ‘SIMILARITY’
INTO A
MATHEMATICAL
NOTION OF
‘SIMILARITY’
‘SIMILARITY’ IS A
MULTIDIMENSIONAL
CONCEPT
0% similarity 100% similarity
threshold threshold
Relevant
and/or f( )
010
Not Relevant
101
001
relevance?
yellow = relevant
white = non-relevant
Relevant Non - Relevant
gold standard labeled data
new sample
of data
sample #2
new sample
of data
sample #2
THAT WE CALL
#LEGALEDUCATION
BUT IT IS NOT
JUST THE
CONSUMPTION
OF LANGUAGE …
LAW / LAWYERING IS (IN
PART) AN EXERCISE IN
LINGUISTIC CONSTRUCTION
AND INTERPRETATION
LAWYERS,
JUDGES AND
REGULATORS
ARE MASSIVE
PRODUCERS
OF TEXT
BRIEFS,
MEMOS,
STATUTES,
OPINIONS,
REGULATIONS,
CONTRACTS,
ETC.
ARE JUST
SOME OF THE
LEGAL WORK
PRODUCT …
PRODUCED
ON A DAILY
BASIS ACROSS
THE WORLD’S
VARIOUS
LEGAL
SYSTEMS
EVEN MANY
YEARS AGO…
THE SCALE
AND
COMPLEXITY
OF THIS
SIGNIFICANT
VOLUME OF
INFORMATION
DROVE BOTH
LABOR MARKET
SPECIALIZATION
AS WELL AS THE
NEED FOR WHAT
MIGHT BE CALLED
PRE-MODERN
INFORMATION
TECHNOLOGY
INDEXING
SYSTEMS
SUMMARIES
TRACKING
SYSTEMS
THE SCALE AND
COMPLEXITY
OF LAW AND
LEGAL WORK
PRODUCT
CONTINUES
TO GROW …
RICHARD SUSSKIND
DEVELOPED THE FIRST
EXPERT SYSTEM IN
LAW IN THE 1980’S
BOTH THE
AI+LAW
CONFERENCE
AND THE JURIX
CONFERENCE
THEREAFTER
BEGAN TO FOCUS
ON THESE TOPICS
FOR MANY YEARS, THE
STUDY OF LEGAL
ARGUMENTATION AND
LEGAL REASONING WERE
A SIGNIFICANT THRUST OF
THE RESEARCH AGENDA
LEGAL THEORY AND
LEGAL NLP ALSO SHARE
A RELATIONSHIP …
REALISM VS. FORMALISM
https://lsolum.typepad.com/legal_theory_lexicon/
2004/05/legal_theory_le_4.html
THE LOGIC
OF THE
LAW?
https://www.ics.uci.edu/~alspaugh/cls/shr/hohfeld.html
I DO *NOT* BELIEVE THAT
LAW IS A LOGIC
COMPUTER BUT IT
BEHAVES MORE OR LESS
AS IF DEPENDING ON
CIRCUMSTANCES
SYNTACTIC NLP
VS
SEMANTIC BASED NLP
SYNTAX
WORDS
WORD FREQUENCY
PART OF SPEECH FREQUENCY
ETC.
EXAMPLES SYNTACTIC NLP
CRTL + F IS EXACT
‘STRING’ MATCHING
REGULAR EXPRESSION
(REGEX)
RULE BASED METHOD(S) THAT CAN BE
USED TO LOOK FOR WORD PATTERNS AND
RETURN RESULTS FOR THOSE PATTERNS
TF - IDF
INVERSE
TERM
DOCUMENT
FREQUENCY
FREQUENCY
SEMANTICS IS RELATIONSHIP
BETWEEN WORDS THAT
INTERACT TO PRODUCE
HIGHER ORDER MEANING
LOTS OF THE TRAINING FOR
LAWYERS IS ACTUALLY ABOUT
THE DEEP SEMANTIC
INTERPRETATION OF LANGUAGE
CONTRACTS
EXAMPLES —> STATUTES
REGULATIONS
JUDICIAL DECISIONS
SEMANTICS IS
HARD FOR
MACHINES
MOST OF THE NLP TOOLS
LARGELY (OR COMPLETELY)
IGNORE SEMANTICS
I WOULD FRAME THE
CURRENT STATE OF
AFFAIRS AS
HOW WELL CAN WE
PERFORM WITH MACHINES
WITHOUT HAVING A DEEP
SEMANTIC UNDERSTANDING
OF LEGAL LANGUAGE ?
< SO AGAIN >
WHAT IS A ROUGH
DEFINITION OF NLP?
ROUGH IDEA
STATISTICAL
REPRESENTATION
OF LANGUAGE
FIRST LET ME
START WITH THE
METHODS IN
GENERAL
(I.E. OUTSIDE OF LAW)
Sentiment Analysis
Sentiment
Analysis
Named Entity Recognizer
Named Entity Recognizer
Machine Translation
Machine Summarization
Question Answering
Question Answering
Note Chatbots
rely upon
QA pairs
Topic
Identification
/ Modeling
Topic
Identification
/ Modeling
Each word is related to
“LIBOR”, but “eurodollar”
has the strongest direct
relationship
27
Maura R. Grossman
Gordon V. Cormack
http://jolt.richmond.edu/v17i3/article11.pdf
http://www.abajournal.com/legalrebels/article/maura_grossman_profile
LEGISLATIVE AND
REGULATORY
PREDICTION
Nay, J. J. (2017).
“Predicting and
understanding law-
making with word
vectors and an
ensemble model.” PLoS
ONE 12(5): e0176999.
https://doi.org/10.1371/
journal.pone.0176999
Consider
a simple
example
Total Number of
4.5 million Act / Agency References
CONTRACT
ANALYTICS
WE CAN DIVIDE THE
CONTRACT LIFECYCLE
INTO PRE AND POST
EXECUTION
(INCLUDING
OBLIGATIONS MGMT)
PRE EXECUTION CONTRACT ANALYTICS
CYCLE TIMES
NEGOTIATION OPTIMIZATION
elevateservices.com
computationallegalstudies.com
@ computational
danielmartinkatz.com
thelawlab.com