=Paper=
{{Paper
|id=Vol-2891/xpreface
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-2891/preface.pdf
|volume=Vol-2891
}}
==None==
https://ceur-ws.org/Vol-2891/preface.pdf
XAILA 2020 - the Third Explainable & Responsible AI in Law (XAILA)
Workshop at JURIX 2020 - the 33rd International Conference on Legal
Knowledge and Information Systems
http://xaila.geist.re
Preface
Grzegorz J. Nalepa, Michał Araszkiewicz, Martin Atzmueller, Bart Verheij, Szymon Bobek
In the last several years we have observed a growing interest in advanced AI systems
achieving impressive task performance. However, there has also been an increased awareness
of their complexity and challenging consequences of their possibly limited understandability
to humans. In response, a number of research directions have been initiated. These include
humanized or human-centered AI, as well as ethically aligned, ethically designed, or just
ethical AI. In many of these ideas, the principal concept seems to be the explanatory
capability of the AI system (XAI), e.g. via interpretable and explainable machine learning,
inclusion of human background knowledge and adequate declarative knowledge, that could
provide foundations not only for transparency and understandability, but also for a possible
value alignment and human centricity, as the explanation is to be provided to humans.
Recently, the term responsible AI (RAI) has been coined as a step beyond XAI. Discussion of
RAI has been again strongly influenced by the “ethical” perspective. However, as
practitioners in our fields we are convinced that the advancements of AI are way too fast, and
the ethical perspective much too vague to offer conclusive and constructive results. We are
also convinced that the concepts of responsibility, and accountability should be considered
primarily from the legal perspective, also because the operation of AI-based systems poses
actual challenges to rights and freedoms of individuals. In the field of law, these concepts
should obtain some well-defined interpretation, and reasoning procedures based on them
should be clarified. The introduction of AI systems into the public, as well as the legal
domain brings many challenges that have to be addressed. The catalogue of these problems
include, but is not limited to:
(1) the type of liability adequate for the operation of AI (be it civil, administrative of
criminal liability);
(2) the (re)interpretation of classical legal concepts concerning the ascription of
liability, such as causal link, fault or foreseeability; and
(3) the distribution of liability among the involved actors (AI developers, vendors,
operators, customers etc.).
As the notions relevant for the discussion of legal liability evolved on the basis of observation
and evaluation of human behavior, they are not easily transferable to the new and disputable
domain of liability related to the operation of artificial intelligent systems. The goal of the
workshop is to cover and integrate these problems and questions, bridging XAI and RAI by
integrating methodological AI, as well as the respective ethical and legal perspectives, also
Copyright 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 Inter-
national (CC BY 4.0).
specifically with support of established concepts and methods regarding responsibility, and
accountability. The workshop program included two presentations by invited speakers and
eight by authors presenting their research.
Our first invited speaker was Philipp Hacker (Europa-Universität Viadrina, Frankfurt an der
Oder) who delivered a lecture `AI and Discrimination: Legal Challenges and Technical
Strategies’. The talk focused on the interaction between AI models and liability in the domain
of non-discrimination. The author pointed out that the output of AI models may exhibit bias
toward legally protected groups. In the past, various fairness definitions have been developed
to mitigate such discrimination. Against this background, the talk presented a new model
which allows AI developers to flexibly interpolate between different fairness definitions
depending on the context of the model application. In the second step the talk inquired to
what extent AI developers may risk liability under affirmative action doctrines if they seek to
implement algorithmic fairness measures in their models.
The second invited speaker was Reinoud Baker (LexIQ) who delivered a lecture `Legal
information systems in production’. The speaker presented LexIQ - a Dutch legal tech startup
using data science for legal information services, endorsing the goal to serve citizens,
governments and businesses, for instance by improved access to justice, efficient use of
resources and enhanced compliance. The talk addressed lessons learned from the past 4 years
and focused on the following questions: What can be achieved with modern software and
algorithms? How to make innovative technologies available for legal professionals and even
the wider public? Which challenges are being encountered?
Barbara Gallina presented the paper `Towards Explainable, Compliant and Adaptive Human-
Automation Interaction’ (coauthored with Görkem Pacaci, David Johnson, Steve McKeever,
Andreas Hamfelt, Stefania Costantini, Pierangelo Dell’Acqua, and Gloria-Cerasela Crisan).
The focus is on the responsible design of systems that interact with humans.
Youssef Ennali and Tom van Engers presented the paper `Data-driven AI development: an
integrated and iterative bias mitigation approach’. They discuss bias that leads to
discriminatory decisions, and the identification and prevention of such bias in an iterative
approach aiming at an `unbiased-by-design’ methodology.
Heng Zheng presented the paper `Precedent Comparison in the Precedent Model Formalism:
Theory and Application to Legal Cases?’ (cowritten with Davide Grossi and Bart Verheij).
An approach to case comparison is presented in terms of propositional logic formulas,
allowing for a generalization and refinement of existing approaches.
Bernardo Alkmim presented the paper `Reasoning over Knowledge Graphs in an
Intuitionistic Description Logic’ (with coauthors Edward Hermann Haeusler and Daniel
Schwabe). The paper uses a natural deduction approach to reasoning over the information
modeled in knowledge graphs, with examples in trust, privacy, and transparency.
Annemarie Borg presented the paper `Explaining Arguments at the Dutch National Police’
(coauthored with Floris Bex). The paper addresses a basic framework for the argument-based
explanation of system conclusions in order to give insight into the underlying decision
models and techniques to police analysts and Dutch citizens.
Łukasz Górski, Shashishekar Ramakrishna and Jędrzej Nowosielski presented the paper
`Towards Grad-CAM Based Explainability in a Legal Text Processing Pipeline’. Their
approach adapts an image processing technique, Grad-CAM to explainability in the setting of
legal texts, describing metrics and initial experiments.
Giovanni Sileno presented the paper `Like Circles in the Water: Responsibility as a System-
Level Function’ (cowritten with Alexander Boer, Geoff Gordon and Bernhard Reader). The
paper sketches an approach addressing computational practices that take system environment
and consequences of system use seriously.
Karl Branting presented the paper `Explanation in Hybrid, Two-Stage Models of Legal
Prediction’. In the paper, core explanation tasks in legal decision support for adjudicators and
litigants are identified, a legal prediction model is presented (addressing process initiation
and assessment), and associated development requirements are discussed.
The workshop was concluded with a roundtable discussion in which the invited speakers
were joined by Karl Branting and Enrico Francesconi as the panelists in a lively discussion
with participants.
The workshop organizers would like to thank the Program Committee members for their
work in the review process. We are also grateful to the Chairs of JURIX 2020 - the 33rd
International Conference on Legal Knowledge and Information Systems for providing the
venue for the third edition of the XAILA workshop, following the successful previous
editions which accompanied the JURIX conferences in Groningen (2018) and in Madrid
(2019), respectively. Finally, we would like to thank our inviting speakers, the authors of
papers and all participants to the workshop for their stimulating contributions to the content
of the XAILA2020 workshop. The distinctive feature of XAILA is its interdisciplinary
character and the creation of a common forum for the exchange of results and opinions from
the point of view of such disciplines as legal theory, ethics and Artificial Intelligence and to
efficiently combine the theoretical insights with practical focus. The scope of topics covered
at this third edition and the high level of the presented contributions and accompanying
discussions create a firm basis for the continuation of relevant investigations at the
forthcoming editions of the workshop.
Program committee
Martin Atzmueller, Osnabrück University, Germany
Michał Araszkiewicz, Jagiellonian University, Poland
Kevin Ashley, University of Pittsburgh, USA
Floris Bex, Utrecht University, The Netherlands
Szymon Bobek, AGH University, Poland
Jörg Cassens, University of Hildesheim, Germany
Enrico Francesconi, IGSG-CNR, Italy
Grzegorz J. Nalepa, AGH University, Jagiellonian University, Poland
Jose Palma, University of Murcia, Spain
Juan Pavón, Universidad Complutense de Madrid, Spain
Monica Palmirani, Università di Bologna, Italy
Radim Polčák, Masaryk University, Czech Republic
Ken Satoh, National Institute of Informatics, Japan
Jaromír Šavelka, Carnegie Mellon University, USA
Erich Schweighofer, University of Vienna, Austria
Piotr Skrzypczyński, Poznań University of Technology, Poland
Michal Valco, Constantine the Philosopher University in Nitra, Slovakia
Bart Verheij, University of Groningen, the Netherlands
XAILA 2020 Organizing Committee
Grzegorz J. Nalepa, Jagiellonian University in Kraków, Poland
Bart Verheij, University of Groningen, the Netherlands
Michał Araszkiewicz, Jagiellonian University in Kraków, Poland
Martin Atzmueller, Osnabrück University, Germany