Commentator
Professor Ida Koivisto, University of Helsinki.
Leader of the assessment
Professor Margrethe Buskerud Christoffersen, Department of Private Law, University of Oslo.
Supervisors
Professor Dag Wiese Schartum, Department of Private Law, University of Oslo.
Heidi Beate Bentzen, Centre for Medical Ethics, University of Oslo.
Abstract
Transparency is a key legal requirement in the EU regulation of artificial intelligence (AI). For example, high-risk AI systems must be designed and developed with an appropriate ‘type and degree’ of transparency. In certain cases, affected parties have the right to be informed about the use of high-risk AI systems, and the right to a ‘meaningful explanation.’ Transparency may seem intuitive on the surface, suggesting that something is open or accessible. Upon closer examination, the meaning of transparency emerges as fragmented and elusive. For example, it is not entirely clear what is meant by the requirement in Article 13(1) of the Artificial Intelligence Act, which states that high-risk AI systems shall be designed with ‘an appropriate type and degree’ of transparency. Transparency may be viewed in a narrow sense, focusing on the disclosure of information or explanations, or in a broader sense, encompassing a range of transparency design features that make the AI system inherently transparent.
The project explores how the concepts of transparent and explainable artificial intelligence are applied in the regulation of AI systems in the European Union (EU). The project is divided into three main components, or research questions.
The first component is conceptual and examines the meaning of transparency and explainability in EU law, by identifying and comparing legal transparency requirements with transparency and explainability applied in computer science. Essentially, the project aims to identify different types or categories of transparency. A conceptual framework is constructed, based on different characteristics or attributes of legal transparency requirements, leading to three distinct types of transparency: transparency as simplicity, observation and explanation.
While the conceptual component focuses on transparency types, the second component examines the degree of transparency. AI systems should be designed with an appropriate "type and degree" of transparency. The question is how the appropriate degree is determined. Two main approaches are examined and compared, the so-called risk- and rights-based approaches. An essential justification or rationale for transparency measures is the risk of harm to "health, safety and fundamental rights," pursuant to Article 9 of the AI Act. The project critically examines how an appropriate degree of transparency can be identified in terms of risk, with a particular focus on how fundamental rights are identified, measured, and balanced against one another.
The third and ultimate component examines the application of transparency and explainability in a concrete setting, namely, the medical sector. The project critically examines the appropriate type and degree of transparency measures for different target groups in the medical sector: domain experts, qualified health personnel, patients and other affected parties.
The overarching aim of the project is to construct a framework for transparent and explainable AI that differentiates, in a meaningful way, between transparency types, guides how the degree of transparency may be determined, and illustrates how an appropriate type and degree of transparency may be applied in practice, with the medical sector as a case example.
Program
- Lars Arnesen - Presentation (30 min.)
- Ida Koivisto - Comments (30 min.)
- Open discussion (30 min.)
- Closed meeting (30 min.)
Guidelines for midway assessments
Guidelines for midway assessments, the PhD programme at the Faculty of Law, University of Oslo
Participate
This event is open to all interested.
Contact
Kristin Koren Viksjø