The Digital Age and the EC proposal for Artificial Intelligence (AI) Regulation

26.05.2021

Considering that the innovative, technological development of Artificial Intelligence (AI) has globally skyrocketed throughout the years, it has become very important for the European Commission (EC) to take a closer look at the safety of AI technology.

On 21 April 2021 the European Commission proposed safety rules and actions, focused on Artificial Intelligence, with the purpose of harmonising the AI systems safety rules. Based on the Commission’s 2020 study “A Europe fit for the Digital Age”, the Commission proposed the following regulations:

A proposal for the first legal code ever
drafted on this subject

A proposal for the adapted
Machinery Directive 2006/42/EC

What are the aims of the proposed Regulations?

By proposing the AI Regulation the Commission aims to ensure the safety of people and businesses, as well as to inspire the growth and innovation of AI. It is, therefore, important that the right standards are developed to make sure that AI is safe and secure. This policy is further extended into the proposal for the Machinery Regulation.

In short, the AI Regulation will centre its attention on the safety risks of AI systems, while the Machinery Regulation will also focus on the safe integration of AI systems into machinery.

What is AI?

Artificial Intelligence (AI) refers to software applications that generate data (output) which influence its surroundings. These data may include content (e.g., of a website), predictions (e.g., weather forecasts), decisions (e.g., after voting online), and recommendations (e.g., after taking an exam). An AI system can be either a component of a product (imbedded software) or a product on its own (stand-alone software).

The AI Regulation proposal in a nutshel

In our daily routines we are often confronted with Artifical Intelligence in all kinds of areas, such as transport (e.g., your car gadgets), cyber security (e.g., on your computer), or health (e,g., a robot-assisted surgical device). Since there are also risks involved in the use of these systems, the EC aims to set the highest safety standards possible, at the same time tending to make Europe a global centre of first-rate AI.

AI definition from a risk-based approach

In view of the various risks involved in the application of the different AI products, while targeting at minimizing future problems with the AI (a future-proof approach), the Commission has made the following risk-based classification of AI technology:

Unacceptable risk High risk Limited risk (low risk)Minimal risk

That presents a clear threat to a person’s safety, livelihood or right, e.g., applications that manipulate human behaviour, such as an interactive toy that evokes dangerous behaviour in children.

Which is a safety component (imbedded or not) that can present serious risks to the safety and health or fundamental rights of a person, e.g.:

critical infrastructure that could present danger to a person’s life, e.g., in transport;
educational training, determining a person’s educational level or professional career, e.g., test scores;
safety components of products, e.g., in robot-assisted automotive industry;
employment management, e.g., CV-software for recruitment procedures;
essential private and public services, e.g., keeping credit scores for citizens linked with the possibility to request a loan;
law enforcement that may influence a person’s basic rights, e.g., the analysis of evidence;
migration, asylum and border control management, e.g., the verification of travel documents;
the administration of justice and democratic processes, e.g., when applying a legal procedure to a specific course of action.

That requires transparency, e.g., a software application with a chat option, i.e., a person should be aware that he is communicating with a software program.

Used in most applications, such as video games.

Conformity assessment of AI systems

Based on the classification in Table 1, the EC has outlined the necessary steps to be taken in case a manufacturer or provider intends to place an AI system on the market and/or put it into service.

Unacceptable riskHigh riskLimited risk (low risk) and Minimal risk

Title II, article 5 of the AI Regulation proposal refers to “prohibited ArtificIal Intelligence practices” that cause physical or psychological/manipulative harm (please be referred to Table 1). These AI systems will be banned, which means that these will not be placed on the market or put into use.

According to Title III, Chapter V, article 43, paragraph 1, one out of two types of conformity assessment procedures may be applicable to a high-risk AI system:

a conformity assessment procedure based on internal control (Annex VI),

a conformity assessment procedure, requiring a Notified Body and based on assessment of the quality management system and assessment of the technical documentation (Annex VII).

The option depends on the type of AI system, as well as whether or not a Notified Body is involved.

Both types of conformity procedures are broadly discussed in Chapter V, as well as the accompanying Annexes.

According to Title IX, article 69, paragraph 1, the EC and the Member States will support businesses that intend to place AI systems other than high-risk AI systems on the market, in drawing up codes of conduct , in order to have them voluntarily apply the mandatory requirements of high-risk AI systems (Title III, Chapter 2) to their non-high-risk AI systems.

What is next?

The European Commission opted for a regulation instead of a directive. This will enable a speedier and more accurate legal implementation. Contrary to a directive that first must be transposed by enforcing similar laws and rules in each separate Member State, a regulation will be applicable immediately in all Member States. In this manner, the EC aims to maintain the consistency and conformity of AI technology.

The AI Regulation proposal will now (together with the Machinery Regulation proposal) be discussed by the European Parliament.

Certification Company will keep track of further developments and publish further updates on the development and implementation of AI Regulation.