Zum Generalthema: LegalTech

Explainable Artificial Intelligence – the New Frontier in Legal Informatics

Bernhard Waltl

Zitiervorschlag: Bernhard Waltl / , Explainable Artificial Intelligence – the New Frontier in Legal Informatics, in: Jusletter IT 22. Februar 2018

In recent years, mainstream media coverage on artificial intelligence (AI) has exploded. Major AI breakthroughs in winning complex games, such as chess and Go, in autonomous mobility, and many other fields show the rapid advances of the technology. AI is touching more and more areas of human life, and is making decisions that humans frequently find difficult to understand. This article explores the increasingly important topic of «explainable AI» and addresses the questions why we need to build systems that can explain their decisions and how should we build them. Specifically, the article adds three additional dimensions to capture transparency to underscore the tremendous importance of explainability as a property inherent to machine learning algorithms. It highlights that explainability can be an additional feature and dimension along which machine learning algorithms can be categorized. The article proposes to view explainability as an intrinsic property of an AI system as opposed to some external function or subsequent auditing process. More generally speaking, this article contributes to legal informatics discourse surrounding the so-called «third wave of AI» which leverages the strengths of manually designed knowledge, statistical inference, and supervised machine learning techniques.

Inhaltsverzeichnis

1. Introduction
- 1.1. AI in the Legal Domain and the Expanding Use of Algorithmic Decision Making
- 1.2. The Need for More Algorithmic Transparency
- 1.3. The Opacity of Algorithmic-Decision-Making Software
- 1.4. The Role of Explanations within a Complex Process
2. A Framework for Explanation and Interpretation
- 2.1. Trying to Capture the Essence of Explainability
- 2.2. Expanding Comparability Dimensions of Machine Learning Algorithms for ADM
- 2.3. Explainability as an Intrinsic Property of Machine Learning Algorithms
3. Conclusion
4. References