Regulation of AI: A close look at the EU's AI regulation (AI Act)

In mid-June 2023, the EU Parliament agreed on a system for regulating AI. Read here to find out exactly what the AI regulation says, how companies are affected by the development and use of AI systems, who is in favour of it and what points of criticism have been raised.

For over two years, the European Union has been endeavouring to regulate artificial intelligence (AI). With the approval of the "AI Act" in the European Parliament, this goal is now within reach. Among other things, the MEPs are calling for an extension of the list of bans proposed by the EU Commission, including biometric classification systems based on sensitive characteristics and preventive police systems. Once an agreement has been reached, talks can be initiated with the EU countries.

AI regulation: when AI falls into the wrong hands

For a EU AI Regulation it's high time, because: "AI is already omnipresent in our everyday lives," says Carina Zehetmaier, President of Woman in AI Austria, at the Deep Dive Meet-ups of INDUSTRIEMAGAZIN. She sees huge potential in the application of AI and believes that it will generate considerable added value.

However, it also points to possible Risks of abuse which should be curbed by regulation. A recent case from China shows what can happen if AI falls into the wrong hands. A company there has applied for a patent to specifically identify the Uyghur minority using AI and facial recognition. "We can only speculate what the goal was," says Zehetmaier.

She adds that problematic use of AI is now occurring more frequently, such as the example of Amazon, which used AI for recruitment. Based on the available data, the system concluded that a disproportionate number of male employees in the company would be positive. As a result, not a single woman was invited to a job interview.

Zehetmaier therefore sees AI as a reflection of society: "AI is not objective or neutral. We tend to be more tolerant of machines than humans. AI ultimately reflects our beliefs and views."

Until 2026 the AI Regulation (AI Act) is due to come into force. The law has cleared the first major hurdle in the European Parliament in Brussels and the details are now being discussed further in the trialogue. Zehetmaier emphasises how the law will affect the economy: "The law not only affects companies that produce AI systems, but every company that uses AI."

The EU aims to regulate the application and not the technology. A further aim is to achieve global ethical standards as has already been done with the General Data Protection Regulation has been achieved. However, even strict regulation of AI will not be able to completely prevent misuse.

Carina Zehetmaier, President of Woman in AI Austria

The AI Regulation Act in detail

Which Key points now includes the new law on Regulation of AI? The proposed legislation follows a risk-based model where regulation depends on the risk posed by the AI. It bans AI systems with unacceptable risks, such as those that could be used for social scoring. All AI applications are to be categorised according to their risk. Depending on their classification, providers must fulfil certain safety and transparency standards.

What risk classifications are there?

1. Unacceptable risk: particularly harmful AI applications that violate EU values, for example because they infringe fundamental rights. This applies in particular to social scoring (the assessment of social behaviour by authorities), the exploitation of children's vulnerability or the use of technology to exert subliminal influence.
2. High risk: AI systems that adversely affect people's safety or their fundamental rights (protected by the Charter of Fundamental Rights of the European Union).
3. Low risk: Special transparency obligations are imposed on certain AI systems, for example if there is a clear risk of manipulation, such as through the use of chatbots. Users should be aware that they are dealing with a machine.
4. Minimal risk: All other AI systems can be developed and used in compliance with generally applicable law.

MEPs also want AI systems that can influence elections or election results or pose a significant threat to human health, safety, fundamental rights or the environment to be added to the high-risk list. The Commission's draft provides for specially adapted rules for the reasonable use of generative AI, such as ChatGPT.

To optimise the Promoting innovation in AIthe parliamentarians are proposing exemptions for research activities and the use of so-called AI real-world laboratories. Florian Tursky (ÖVP), State Secretary for Digitalisation, expressed his satisfaction with the result and emphasised the urgency of rapid regulation of AI.

The roadmap of the AI Act

Global relevance for AI regulation

The newly introduced EU AI Regulation becomes Worldwide relevance as it extends to companies that provide or use AI systems within the EU. Even if providers are based outside the EU, such as in the UK or the US, the law would apply if their services are used in the EU.

A Critical point A key issue in the discussions was determining which AI applications should be banned due to their unacceptable risk. Initially, a ban on AI-supported tools for monitoring interpersonal communication was considered, but this proposal was rejected. Instead, the use of software for biometric identification was restricted.

This identification software, which was originally only prohibited for real-time applications, may now only be used for serious criminal offences and with prior judicial approval. In addition, the use of AI-supported emotion recognition software in the areas of law enforcement, border management, the workplace and educational institutions has been prohibited.

The current draft law does not yet contain any guidelines for requirements for developers of generative AI such as GPAIS (General Purpose AI Systems) - categorisation as high-risk is currently being discussed in politics. Violations of the law could result in fines of up to 30 million euros or six per cent of global profits, whichever is higher.

Criticism of the EU's AI regulation