AI Act passes EU parliament, arguments on implementation and implications heat up

The European Union’s AI Act has been approved by European Parliament by a wide margin. With one more approval, it will establish new rules for facial recognition and other algorithmic applications, which may serve as a blueprint in other parts of the world.

The Act passed with a vote of 523 for and 46 against with 49 abstentions.

Remote biometric identification by law enforcement is allowed only in specific circumstances, and real-time remote biometrics can only be deployed with judicial or administrative authorization, with limits to where and for how long. The Act also includes new limits to biometric data collection practices and uses of inference systems like emotion recognition or predictive policing.

A final review of the Act’s language is pending, as is formal approval by the European Council. Once those hurdles are passed and it is entered into the Journal, the AI Act will enter into force 20 days later, with the various bans, codes of practice and obligations for high-risk systems taking effect over a period of three years.

IT sector trade group Digital Europe notes in its reaction that only 3 percent of the world’s AI unicorns hail from Europe, and private investments into AI are five times higher in China and 14 times higher in the U.S.

“We are pleased to have one set of rules and not 27, but it’s no secret that the AI Act will add extra burdens that companies outside of Europe won’t have,” Digital Europe Director General Cecilia Bonefeld-Dahl writes.

An updated article by Veridas notes that the Act focusses on consent and risk as the key concepts for classifying AI applications. The company also points out the non-remote biometric identification has been classified as low risk.

International influence

Meanwhile, delegates to the Council of Europe are at work on a Convention on Artificial Intelligence, which is on track to be ratified in May, SWI swissinfo.ch reports.

The Convention will provide the first legally-binding set of rules for AI that can be applied outside of Europe, if signed and ratified by foreign governments. The Swiss outlet expresses concern that non-European countries that prioritize AI business interests over human rights will get their way on how the convention is worded.

The Council consists of 46 member states, 27 of which are part of the EU, with another 6 states in observer roles. Switzerland has been a member in the Council of Europe since 1963, but is not an EU member state.

Member states are participating in the drafting process, but civil society groups, along with businesses and academics, cannot. This makes “it impossible for NGOs to correct the compromises previously agreed upon by governments,” says the thinly-veiled editorial.

The Council has a mandate to protect Europe’s vision of human rights, which some civil society groups say is being undermined by foreign business interests.

At issue is whether the Convention should stipulate binding standards for the private sector, and the right of member states to exempt private businesses, either on a temporary (opt-out) or permanent (opt-in) basis. The U.S. and other observer nations like Canada, Britain and Japan have been pushing for the opt-in option, which would allow states to decline the option and keep the Convention rules to the public sector.

The U.S. is unlikely to ratify the Convention any time soon, however, as SWI swissinfo.ch acknowledges.

The Centre for Democracy & Technology Europe makes a similar claim that human rights “ground was lost on human rights in the final negotiations” on the Act (rather than the Convention). CDT Europe says in a statement, however, that the final version is better than the first version.

Australia’s government is considering a fast-track process to approve AI technologies that have already gained approval under regulatory schemes elsewhere, like the AI Act.

A Department of Industry representative told InnovationAus that the department, which is in charge of formulating Australia’s rules for high-risk AI, is watching developments in Europe closely. He suggested the department will take a wait-and-see approach to judging the AI Act’s effectiveness.

Article Topics

AI  |  AI Act  |  biometric identification  |  biometrics  |  data privacy  |  EU  |  facial recognition  |  legislation