Download "Algorithmic transparency and accountability of digital services" here

How can European law ensure transparency in the obscure field of algorithms? And how can we regulate these mechanisms?

February 2024 marks the full entry into force of the new Digital Services Act (DSA) of the EU. This new report looks at this turning point in the way in which algorithmic systems can be made more transparent and accountable to the user and therefore successfully regulated.

This new report - Algorithmic transparency and accountability of digital services - by the European Audiovisual Observatory has been written by an editorial team from Saarbrücken's Institute of European Media Law.

Following the introduction, chapter two looks at the standard setting work in the field of algorithm good practices carried out by the Council of Europe before the arrival of the DSA. As early as 2017, the Council prepared a study on the impact of algorithms on human rights as well as regulatory implications following on.  A further Council study in 2019 focused on the notion of responsibility and the exercise of digital power without responsibility through advanced digital technologies. Following on from this, the Council of Europe issued a series of standard-setting recommendations during the period 2020 - 2022. These included themes such as the human rights impact of algorithmic systems and the impacts of digital technologies on freedom of expression, for example. The Council is currently working on its “[Framework] Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law”, still in its drafting phase. It is aimed at being a Convention that is open for ratification not only by the CoE member states, but going beyond and therefore having the potential to be a first international baseline agreement on fundamental principles in relation to AI.

Chapter three gives an overview of the European Union regulatory framework pertaining to algorithms. Given that core values of EU law are fundamental rights and transparency, one can easily understand why legislators are interested in algorithms. Secondary EU legislation first started to regulate these in 2019 with the Platform-to-Business Regulation (P2B regulation)  as well as the amended Consumer Rights Directive. The goal here was to achieve transparency concerning online ranking systems (driven by algorithms) of products on offer. Media related EU legislation such as the AVMSD also concentrates on algorithms as used by video sharing platforms, for example. The AVMSD stipulates that VSPs shall establish “transparent and user-friendly” notification systems, systems for explaining how notifications are dealt with, age verification and rating systems, as well as systems for parental control.” The authors then fast forward to the creation of a proposed EU AI Act. The proposal aims at creating harmonised rules for the placing on the market, the putting into service and the use of AI systems in the EU. For the moment, the provisional acceptance of the drafted text by the Coreper was done on 2 February 2024 and the document was made publicly available. The Act is set to be adopted in April 2024 by the plenary of the European Parliament and shortly after by the Council before it will be published in the Official Journal and only then being able to enter into force. 

Chapter four of this report zooms in on the new Digital Services Act, EU legislation which became fully applicable in February 2024. One of the main pillars of the DSA is transparency, whether this be algorithmic transparency or the risk of content manipulation such as disinformation campaigns or deep fakes. The DSA also seeks to minimize the risks of and regulate the area of recommender systems which reply on algorithmically driven suggestions proposing a hierarchy of products or choices to the user. In this case the DSA introduces additional transparency obligations according to which providers of online platforms have to set out in their terms and conditions the main parameters used in their recommender systems. 

Chapter five zooms back out to look at developments beyond Europe as well as those taking place in national law.  The authors cite the work of the OECD in this field. As early as May 2019, OECD countries in the OECD Council adopted the Recommendation on Artificial Intelligence which provided for the first intergovernmental standard on AI and served as a basis for the G20 AI Principles. The UNESCO also published in 2023 an extensive set of Guidelines that address the governance of digital platforms and recommend action both for states and the platforms themselves on issues such as transparency, content moderation, data harvesting, targeted advertising, and the sharing, ranking, and/or removal of content.  The authors then move on to examine countries where obligations concerning platforms have been introduced ahead of the overarching EU AI Act. Germany for example, has had legislation in place since 2020. The German Interstate Media Treaty stipulates that users of intermediary platforms must be able to understand why certain content is shown to them, why it is presented in the given order and why other content is not displayed. The state of play in the UK and the USA is also outlined in this chapter.

Chapter six examines the institutional structures and oversight, in other words, the authorities responsible for overseeing the respect of legislation in this field. Regarding the DSA, for example, a newly created European Board for Digital Services convenes representatives from all of the member states with the Commission chairing this independent advisory group in order to oversee transposition and respect. For the EU member states the situation is clear: besides the setting up of a new surveillance and governance architecture in relation to the AI Act, by 17 February the member states will have had to notify the Commission which national authority or body will take the role of Digital Services Coordinator according to the DSA.

The authors round off by concluding that "transparency alone does not necessarily change practices and behaviours, if it is not followed up on with rules on how algorithmic systems have to be or are prohibited from being designed.[...] In that sense, the activities of the authorities and bodies in charge of overseeing the use of algorithmic systems especially by platforms will be decisive."

A very useful glossary of terminology related to algorithms completes this new report.

A must-read round-up of the current issues concerning algorithm use in Europe and the legislator's approach to ensuring greater transparency in order to regulate these new AI systems.