The European Commission has proposed a new regulation on artificial intelligence. The legal framework is expected to help set global standards for AI, while ensuring EU industry can benefit from the new computing developments and personal privacy is protected.
The regulation takes a gradual approach depending on the perceived level of risk from AI applications. Those with an unacceptable risk, that threaten safety, the livelihoods and fundamental rights of people, will be banned outright.
A number of other applications have been identified as 'high risk' and will be subject to strict controls. This covers AI used in situations such as critical infrastructure and essential systems like banking and credit scoring, justice and law enforcement, systems covering access to employment or education, and immigration management. These applications must come with risk assessment and mitigation systems, data controls and logging systems to ensure traceability, open access to assess compliance, disclosure to end-users and human oversight.
Biometric ID systems, such as facial recognition cameras using AI, are considered high risk. These will be in principle banned under the proposed regulation, but certain situations may permit their use, such as missing person searches, imminent terrorist threats or serious crime investigations. These situations would require prior authorisation by a judge or other independent body and be subject to limits in time, geographic reach and the databases searched.
Other AI applications considered to have limited risk, such as chatbots used for customer service, would be subject only to transparency obligations. So users would have to be informed they are interacting with a machine before the conversation can proceed.
The Commission said the "vast majority" of AI systems fall into the final category of minimal risk. The proposal allows the free use of these applications, such as AI-enabled video games or spam filters, without any specific regulatory obligations. Industry would be invited to develop voluntary codes to help govern low-risk services and provide guidance on any need for future regulation.
In terms of enforcing the proposed regulation, the Commission wants to hand this to existing market surveillance authorities operating in the EU states. It would also create a European Artificial Intelligence Board to support implementation and coordination and help develop new standards for AI.
A second regulation is planned to cover integration of AI in hardware, from robots to lawnmowers, 3D printers, construction machines and industrial production lines. Designed to replace the Machinery Directive, the new Machinery Regulation defines health and safety requirements for the equipment. Businesses will need to perform only one conformity assessment for the two regulations.
The new regulations must still be approved by the European Parliament and Council before they can take effect.
