Facial recognition complicates consensus on European AI law

Historic debate airs in the European Parliament for the last step before the EU approves next year the first major legislation designed to regulate artificial intelligence.

Oliver Thansan
Oliver Thansan
13 June 2023 Tuesday 11:08
9 Reads
Facial recognition complicates consensus on European AI law

Historic debate airs in the European Parliament for the last step before the EU approves next year the first major legislation designed to regulate artificial intelligence. It has been an arduous debate for months, with many agreements on the need to set the lines in the use of such a powerful technology, but also with some ideological disagreements on aspects such as the use of facial recognition in public spaces. If the proposal is approved today in the Chamber, the European Council -composed of the governments-, the Commission and the Parliament, will have to agree on a common text for the new AI law.

Although there is a widespread conviction in the Eurochamber that artificial intelligence needs regulation as soon as possible, the final drafting of the regulation will be a difficult task. The ban on facial recognition in public areas will be one of the biggest bottlenecks. Several members of the European People's Party intervened yesterday to ask for at least two exceptions to the rule that the police cannot use it in their investigations.

The populists propose – and will insist on the negotiation of the text that will begin if the proposal is approved today – that there must be exceptions to the ban on facial recognition through AI. Two of those proposed by some MPs were that the police could use AI identification in cases of terrorism and in cases of missing children. EU interior ministers have previously expressed concern that they believe the proposed law may unduly restrict the use of this technology by security forces.

The Eurochamber proposal envisages banning facial identification in public spaces in real time and also in recordings, although with the exception of judicial investigations for serious crimes. It is the most controversial point, but it is not the only one.

While the progressive forces demanded a veto on the fact that AI is used for biometric identification and warned that this is how it is used in China and Russia, the conservatives appealed to the democratic nature of the European proposal and hope introduce the exception to the law that allows the police to use technology in certain cases.

The executive vice-president of the Commission, Margrethe Vestager, urged the deputies to approve this rule today and expressed her conviction that "a balance" will be found between the veto on the recognition of people in public spaces and the prosecution of crimes.

The biometric identification of the AI ​​can also be applied to other aspects, as recalled yesterday by some deputies who demanded an exhaustive definition of everything that should be vetoed. One of the distinguishing features of the Eurochamber proposal is that the regulations establish obligations for providers of AI services according to the degree of risk they pose.

In this assessment, the European standard proposes to consider as an "unacceptable risk" those that involve the "cognitive manipulation of the behavior of specific vulnerable people or groups", such as toys that encourage dangerous behavior in children; the social score, a "classification of people based on their behaviour, socio-economic status or personal characteristics", and the controversial biometric identification.

AI systems that affect security and fundamental rights will also be considered “high risk”. All products that use it must be in a database of the European Union.

Regarding generative AI, such as ChatGPT, Midjourney or Anthropic, the Parliament proposes that it fulfills transparency requirements, such as disclosing that the content has been generated using this technology, preventing it from generating illegal content and publishing summaries of the data protected by copyright 'author used to train the system.

Finally, in the gradation of risks, the Parliament considers certain AI systems to be "limited" and will have to meet transparency requirements, so that users can make informed decisions. The proposal is for people to be aware of when they are interacting with a system of this kind. After today's vote, the text could be ready for final approval at the end of this year.