SLIDE 1 Artificial Intelligence as Law
Bart Verheij Department of Artificial Intelligence, Bernoulli Institute of Mathematics, Computer Science and Artificial Intelligence www.ai.rug.nl/~verheij
SLIDE 2 Guillotine, Nieuwmarkt, Amsterdam, 1812 (Rijksmuseum RP-P-OB-87.033)
SLIDE 3
SLIDE 4
SLIDE 5
SLIDE 6
TV series Futurama, judge 723 (futurama.fandom.com/wiki/Judge_723)
SLIDE 7
LOI n° 2019-222 du 23 mars 2019 de programmation 2018-2022 et de réforme pour la justice (1) - Article 33
Les données d'identité des magistrats et des membres du greffe ne peuvent faire l'objet d'une réutilisation ayant pour objet ou pour effet d'évaluer, d'analyser, de comparer ou de prédire leurs pratiques professionnelles réelles ou supposées. The identity data of magistrates and members of the registry cannot be reused with the purpose or effect of evaluating, analyzing, comparing or predicting their actual or alleged professional practices.
March 2019
SLIDE 8
May 11, 2019
SLIDE 9
SLIDE 10
NRC Handelsblad June 15, 2019
SLIDE 11
Dutch AI Manifesto, bnvki.org
SLIDE 12 Excellence across all of AI. For all of Europe. With a Human-Centred Focus.
I am a member of the CLAIRE Research Network. Network
Centre
Hub
Centre Centre Centre Centre
SLIDE 13
AI&Law has worked on the design of socially aware explainable responsible AI for decades already
SLIDE 14
AI as Law
SLIDE 15
SLIDE 16
SLIDE 17
SLIDE 18
SLIDE 19
SLIDE 20
CodeX Techindex (Stanford)
SLIDE 21
SLIDE 22
Legal tech exists, is it AI?
SLIDE 23
SLIDE 24
AI & Law is hard
SLIDE 25
Nederland ontwapent The Netherlands disarm
SLIDE 26 Hurdles
- 1. Legal reasoning is rule-guided, rather than rule-
governed.
- 2. Legal terms are open textured.
- 3. Legal questions can have more than one answer,
but a reasonable and timely answer must be given.
- 4. The answers to legal questions can change over
time. Rissland 1988 on Gardner 1987 Harvard Journal of Law and Technology
SLIDE 27 The subsumption model
Facts (given) Legal consequence(s) Rules (given) Montesquieu (1689-1755): The judge as ‘bouche de la loi’
SLIDE 28
The theory construction model
SLIDE 29 Facts (initial version) Evidence (initial version) Legal consequences (initial version) Facts (final version) Evidence (final version) Legal consequences (final version)
The theory construction model
SLIDE 30
AI as Law
SLIDE 31
Artificial Intelligence
AI as mathematics AI as technology AI as psychology AI as sociology AI as law
SLIDE 32
Artificial Intelligence
AI as mathematics Logic Probability theory AI as technology Expert systems Machine learning AI as psychology Cognitive modeling Cognitive computing AI as sociology Multi-agent systems Autonomous robots AI as law ...
SLIDE 33
Toulmin on logic
Logic as mathematics Logic as technology Logic as psychology Logic as sociology Logic as law
SLIDE 34
Law
Law as mathematics Rule following Stare decisis Law as technology Civil law Common law Law as psychology Judicial reasoning Judicial discretion Law as sociology Critical discussion Societal regulation Law as law Rule of law Justice
SLIDE 35
Artificial Intelligence
AI as mathematics Logic Probability theory AI as technology Expert systems Machine learning AI as psychology Cognitive modeling Cognitive computing AI as sociology Multi-agent systems Autonomous robots AI as law Hybrid critical discussion systems
SLIDE 36
SLIDE 37
Topics in AI
Reasoning Knowledge Learning Language
SLIDE 38
Reasoning
Argumentation Defeasibility Inconsistency, incompleteness, uncertainty
SLIDE 39
John is owner Mary is owner Mary is original owner John is the buyer John was not bona fide John bought the bike for €20 Pros Cons
SLIDE 40
SLIDE 41
Grounded extension Stable extension Preferred extension Complete extension
Abstract argumentation semantics (1995)
Dung 1995
SLIDE 42
Grounded extension Stable extension Stage extension Semi-stable extension Preferred extension Complete extension
Abstract argumentation semantics (1996)
Dung 1995 Verheij 1996
SLIDE 43
Argumentation semantics (2003)
DefLog Verheij 2003
Stable Semi-stable Preferred Stage Stable
SLIDE 44
Abstract argumentation (Dung 1995)
Dung’s abstract arguments have internal structure representing support
Abstract version:
SLIDE 45
Case models
SLIDE 46
Knowledge
Argumentation schemes Norms Ontologies
SLIDE 47
SLIDE 48
SLIDE 49
2015
SLIDE 50
Scenario schemes Bex 2009 dissertation
SLIDE 51
SLIDE 52
Learning
Statistical analysis Open data Neural networks
SLIDE 53
Netherlands Criminal Courts Prediction Machine
Predict
SLIDE 54
Netherlands Criminal Courts Prediction Machine
Predict Prediction: The suspect is guilty as charged
SLIDE 55
Netherlands Criminal Courts Prediction Machine
Predict Prediction: The suspect is guilty as charged
CBS: Central Bureau of Statistics in the Netherlands
SLIDE 56
Netherlands Criminal Courts Prediction Machine
Predict Prediction: The suspect is guilty as charged
CBS: Central Bureau of Statistics in the Netherlands
SLIDE 57
Judicial prediction
US supreme court prediction
Prediction method Correct Majority outcome (= always affirm) 60% Majority outcome in past 10 years 67% AI model (Katz, Bommarito, Blackman 2017) 70%
European Court of Human Rights
Prediction method Correct Random guess (prepared dataset) 50% AI model (Aletras et al 2016) 79% AI model (Aletras et al 2016) only using circumstances 73%
SLIDE 58
Neural networks
Bench-Capon ICAIL 1993
Should be 60 for women, 65 for men (difference 5)
SLIDE 59
Cases and rules
Data Knowledge
SLIDE 60
Language
Labeled data Prediction Argument mining
SLIDE 61
2011
SLIDE 62
2014
SLIDE 63
ICAIL 2009
SLIDE 64
SLIDE 65
SLIDE 66
SLIDE 67
SLIDE 68
Feb 11, 2019
SLIDE 69
Vlek 2016 dissertation
SLIDE 70
AI as Law
SLIDE 71
Artificial Intelligence
AI as mathematics Logic Probability theory AI as technology Expert systems Machine learning AI as psychology Cognitive modeling Cognitive computing AI as sociology Multi-agent systems Autonomous robots AI as law Hybrid critical discussion systems
SLIDE 72
Topics in AI
Reasoning Argumentation Formal semantics Knowledge Schemes and norms Commonsense Learning Rules and cases Explainability, responsibility Language Interpretation Understanding
SLIDE 73
www.ai.rug.nl/~verheij/oratie
SLIDE 74
SLIDE 75
Cases and rules
Data Knowledge
SLIDE 76
TV series Futurama, judge 723 (futurama.fandom.com/wiki/Judge_723)
SLIDE 77
AI as Law
SLIDE 78
Conclusion
AI&Law is more relevant than ever. AI&Law has worked on the design of socially aware, explainable, responsible AI for decades already. AI&Law addresses the hardest problems across the breadth of AI (reasoning, knowledge, learning, language). AI&Law inspires ideas for new solutions (argumentation, schemes and norms, rules and cases, interpretation).
SLIDE 79 Artificial Intelligence as Law
Bart Verheij Department of Artificial Intelligence, Bernoulli Institute of Mathematics, Computer Science and Artificial Intelligence www.ai.rug.nl/~verheij
SLIDE 80 Further reading
Verheij, B. (2018). Arguments for Good Artificial
- Intelligence. Groningen: University of Groningen.
Inaugural lecture. http://www.ai.rug.nl/~verheij/oratie/. details pdf
