European Union Artificial Intelligence Act: What Is It and What Can We Expect?

by Kanerva Jalas

In 2018, the European Union (EU) set global standards for what constitutes personal data, and what level of protection should be expected from the ever more powerful technological firms. Therefore, the expectations for a similar development are anticipated to take place when the EU AI Act enters into force in the near future. In late April 2021, the EU Commission proposed a Regulation - a piece of legislation which is directly applicable in all Member States - for the government of Artificial Intelligence (AI). This post will delve into the legal, technological and global implications of the proposed piece of legislation.

The discussion of the impact of AI in the EU began in 2017 when the European Parliament directed the Commission to analyse the possible implications of the technology in the Union and requested 'a draft EU Framework’. The first indication for the proposal for an EU-level legislative piece in the field of AI was made in 2019, at the leader’s meeting consisting of heads of government and countries of the 27 Member States. This was also seen from the political direction plans of the newly elected President of the Commission, Ursula von Der Leyen, which included an intention to initiate legislative development considering the ethical developments. These guidelines did not only aim to propose the AI Act but also included proposals (and adoptions) for several other legislative pieces such as the Digital Markets Act (DMA), the Data Governance Act, the Digital Services Act (DSA), and the Machinery Regulation. Furthermore, in 2020 and 2021, the European Parliament adopted four resolutions for the legislation, some focusing on the adoption assessing the ethics of AI and its use, others for the possibility of accountability and the possibility to bring forth civil liability claims, and the effects of AI in the field of intellectual property. The draft AI Act takes the most central approach to regulate AI and the developments in the field. Furthermore, the legislation is the first instance where horizontal regulation is proposed for AI.

From a substantive standpoint, Commission preparatory impact assessment addressed key points which were identified to be addressed in the legislation: first, the risks posed by the use of AI for citizens and safety; second, the possibility for the use of AI to result in infringements of rights; third, whether governments are prepared and have sufficient resources to address AI development; fourth, the uncertainty of AI legislative effects on its usage in companies; fifth, the economical impact of the legislation on the EU due to diminished competitive ability, and lastly, effects of fragmented legislation within the Union for the internal market. As stated in the Proposal for the Regulation, the legislation would be based on Article 114 of the Treaty on the Functioning of the European Union (TFEU), which allows for harmonisation of Union Member States legislation on the grounds of ‘establishment and functioning of the internal market’. Fulfilling this requirement, the Proposal states that the primary aim of the legislation is to make sure that the single market remains functional. Additionally, the Proposal pin-points ‘specific aims’ of the Regulation including 1) the AI systems within the internal market are in accordance with pre-existing EU legislation and are safe, 2) to promote innovation and investment in AI, while ensuring legal certainty, 3) safeguard fundamental rights and ensure effective legal recourse and 4) to ensure the functioning of the internal market and counteract possible fragmentation within it. Another legislative basis for the Proposal is Article 16 TFEU, which enshrines the right of individuals to the protection of their personal data. According to the Commission, Article 16 is a necessary basis for the legislation due to its content being based on the protection of personal data, specifically imposing restrictions on the use of AI for ‘remote biometric identification’.

The Proposal defines that the Regulation would take a ‘technology-neutral definition’ for AI. According to most scientific communities, there is no agreed definition for AI systems, but instead, the term is used in a broad manner to describe a multitude of different computer programs or technologies, such as machine learning, or natural language processing. The definition proposed for the Regulation originates from an Expert Group, highlighting a commonly adopted view within scientific academia which states that AI ‘refers to systems that display intelligent behaviour by analysing their environment and taking actions - with some degree of autonomy - to achieve specific goals’. As the core of the Proposal, the Commission aims to adopt a ‘risk-based approach’ identifying four different categories. First, there is a category of ‘low or minimal risk’, the stage at which no obligations would be imposed on the actors employing the AI system. Second, there are those systems with ‘limited risk’, which would require the imposition of transparency obligations. These concern AI systems which either 1) ‘interact with humans’, 2) observe emotion or determine biometric data or 3) which can create deep fake content. The third category includes AI systems with a ‘high risk’. If an AI system which qualifies as ‘high-risk’ is released on the internal market, it must comply with certain requirements, and be subjected to an ‘ex-ante conformity assessment’. Whether a system falls under this category depends on both, the function which will be performed by the system as well as the purpose for which it is intended to be used. Finally, there are prohibited AI systems and practices, which target AI uses that can be considered to be contrary to EU values, for example, due to violations of fundamental rights of individuals. More specifically, uses which would include ‘subliminal techniques’ and be manipulative; exploit vulnerable groups or individuals; which would be employed for ‘social scoring’ by governments or authorities, or wide-range remote biometric identification are considered to pose an unacceptable risk and are prohibited in the EU area.

The practical governance effects of the AI Act would include the establishment of a European Artificial Intelligence Board at the Union level, which would include representatives from the Union States as well as the Commission. The aim of the board would be to ensure the presence of a harmonised implementation of the Act and cooperation at the Union level. Additionally, Member States would implement national supervisory authorities to exercise oversight of the application of the Act rules. To facilitate proper compliance, the supervisory authorities would be accompanied by market surveillance authorities to analyse whether operators are complying with requirements imposed for ‘high-risk’ AI usage. Finally, the Act would authorise the issuance of administrative fines, for a maximum amount of 6 per cent of total annual turnover in the case of non-compliance.

For now, the AI Act Proposal includes the introduction of a first-ever attempt at harmonisation of Member State legislation for AI. Support has been shown for the EU’s attempt to regulate the fast-developing area, while criticisms have been posed for the Act’s harmonising effects and possible future hindering of other areas of regulation such as carbon emissions. For now, it will be left to be seen how the legislative proposal develops further in the EU legislative process.

Sources:

Commission, Independent High-Level Expert Group on Artificial Intelligence, ‘A Definition of AI: Main Capabilities and Disciplines’ (2019): https://digital-strategy.ec.europa.eu/en/library/definition-artificial-intelligence-main-capabilities-and-scientific-disciplines, accessed on the 8th of April.

Proposal for a Regulation of the European Parliament and of the Council Laying Down Harmonised Rules on Artificial Intelligence and Amending Certain Union Legislative Acts COM(2021) 206 final.

Consolidated version of the Treaty on the Functioning of the European Union [2012] OJ C326/47.

Commission, ‘Impact Assessment accompanying the Proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence’ (Staff Working Document) SWD(2021) 84 final.

The AI Act, ‘What is the EU AI Act?’: https://artificialintelligenceact.eu/, accessed on the 8 of April 2023.

European Parliament, ‘Artificial Intelligence Act’ (Briefing EU Legislation in Progress, 2021).