Europe now leads the world in AI regulation

Risk categorisation

The law categorises AI-based applications into four groups based on the level of risk associated with their use: unacceptable risk, high risk, limited risk, and low risk.

Unacceptable risks encompass AI systems capable of manipulating and coercing human behaviour to alter decisions or influence public opinion.

Using biometric information or physical characteristics (such as height, gait, eye colour, facial features, voice patterns etc) in public spaces for the purpose of suppressing specific groups is also now unacceptable.

The law also identifies government-run social assessment systems (or 'social scoring') as an unacceptable risk.

These are used by some governments to compile data on an individual's risk levels by monitoring their financial, social, and criminal behaviour.

High-risk applications in AI might include those stemming from the use of self-driving cars, medical applications, or CV-scanning tools that rank job applicants.

The legislation lists any AI system used as a safety component in the management of critical digital infrastructure, such as road traffic, water, gas, and electricity, as high-risk.  

It also deems AI systems to be high-risk if they help assess and determine someone's employment, health interventions, creditworthiness, insurance, or benefits. Limited and minimal risks are associated with entertainment systems such as games or systems generating text or videos.

The law provides a grace period for companies to align with its provisions. Applications posing unacceptable risks must comply within six months of the law's enactment.

Additionally, lawmakers are granted nine months to identify best practices and establish a robust model for companies and AI developers.

Benefitting humanity

EU legislators want AI to be harnessed for the good of humanity, to serve and enhance human welfare while upholding fundamental rights and minimising risk.

In this sense, Europe can differ from the US, where priorities can include material gains and technological proliferation to boost productivity.

While the EU's law generally prohibits the use of biometric identification systems, it does allow for certain exceptions under strict conditions, provided that judicial and administrative safeguards are in place.

Such technology may be permissible within specified parameters and timelines, for instance, to locate missing people or to pre-empt threats to public security.

These provisions set boundaries and establish the European Union as a global advocate for human rights in technology. Firms that disregard the bloc's ethical standards can now expect to be penalised.

Enshrining honesty

The EU law places a significant emphasis on ensuring transparency in AI applications. For instance, content generated by AI must now be disclosed as such. This should help stop the spread of 'deepfakes' and false reports.

It underscores the importance of categorising applications based on their associated risks and obligating developers to furnish essential information. AI systems must also adhere to the requirements of EU intellectual property rights.

The Computer and Communications Industry Association (CCIA) Europe, a lobby group, wants to monitor the law's impact on AI productivity and warns of possible ramifications for competitiveness.

It cautioned against burdening innovative developers with disproportionate compliance costs and additional administrative requirements, such as personal data.

CCIA Europe's competition policy manager told Euronews that a competitive AI market would be more advantageous for EU consumers than prematurely imposing additional regulations, which could stifle innovation and cooperation.

The EU law stands as a model and reference point for numerous countries grappling with the intricate task of regulating AI and mitigating its risks. It is also a vehicle through which the European Union can export the values and standards of its member states in areas like human rights and equality.

Acting as template

Other countries are seeking to do something similar. In September 2021, Brazil's Congress passed a bill that creates a legal framework for AI (it still needs to pass through the Senate).

The EU legislation will set a precedent, with similar measures expected in the US, which continues to trail the EU in regulating the AI sector.

Two years after their publication, the voluntary pledges of AI companies remain mere statements of intentions. A year earlier, in 2021, the White House drafted an AI Bill of Rights. It may have been well-meaning, but nothing was binding.

Late last year, President Biden issued an executive order promoting "the safe and reliable development and use of AI." It established new AI safety and security standards, prioritised privacy protection, and advocated for innovation and competition.

In terms of sector regulation, it helped establish the likely US direction, yet it was still not binding, so it did not compel action and adherence, as the new EU law does.

With no AI laws passed by Congress, Biden's administration has been limited to what essentially amounts to guidelines, leaving big US tech companies to formulate their own policies and police their own products.

Alternative pathways

In March, the UN General Assembly passed a landmark resolution aimed at promoting the safe use of AI, urging countries to avoid AI applications that contravene international law or threaten human rights.

Although not legally binding, its international adoption could catalyse action and encourage countries to enact their own legislation governing the use of AI.

Critics say contentious issues like the use of AI in military applications were not covered, nor was the cosy relationship between big tech firms and governments.

The US supported the resolution, which sparked debate. According to Forbes, a business magazine, Washington's backing was with a view to counterbalancing China's growing influence in the field of AI.

The country that establishes the first workable legal framework for AI is likely to shape the conditions and values underpinning its use, which Washington recognises. 

Consequently, there is the opportunity to promote Western values such as freedom and democracy in AI applications, in contrast to the authoritarianism, surveillance, and censorship now so well-established in China.

In areas such as legal frameworks, the development of AI is becoming increasingly subject to traditional geopolitical rivalries between the US and China.

But the technology itself is advancing far faster than the efforts of governments and international organisations to regulate it. As soon as a law is drafted, it is out of date.

Elon Musk, the entrepreneur who owns electric car maker Tesla and social media site X, was an early backer of OpenAI, which created the popular generative AI algorithm ChatGPT. 

Read more: Dissecting discord: The unravelling of the Musk-Altman OpenAI partnership

He has joined other experts to issue warnings about the alarming evolution of the technology. Its speed and scope of development mean that it will soon surpass human intelligence, adding to the complexity of its regulation.

Furthermore, the nature of the technology means that it is not delineated by geographical boundaries, unlike other regulated industries. AI is an area that shows the limits of sovereignty.

The need for legal safeguarding is now both urgent and important, because authorities and companies abusing such powerful technology could cause untold problems. 

In the race between innovation and regulation, the EU has fired the first serious shot for the latter.