The European Commission has proposed a draft of a regulation for guaranteeing a trustworthy and reliable artificial intelligence

On 21.04.2021 the European Commission proposed a system of measures, aimed at creating guarantees for the development of trustworthy and reliable artificial intelligence and securing the leading role of the European Union in this process. A part of the proposed measures is also a draft of a regulation to define the requirements for systems, which rely on artificial intelligence.

With the proposed regulation draft the use of artificial intelligence is strictly prohibited, when it can be a risk for the safety and rights of humans. Examples for such unacceptable risks is the use of artificial intelligence for manipulation of human behavior to circumvent free will or for allowing “social scoring” by governments.

For high-risk systems – which include such, related to transport infrastructure, education and vocational training, employment, essential public and private services, migration and border control, administration of justice and democratic process – the following guarantees for the use of trustworthy and reliable artificial intelligence are required:

– adequate risk assessment and mitigation measures;

– requirements for quality of the datasets feeding the systems to minimize the risks of discriminatory or wrong decisions;

– logging of the activity;

– availability of detailed documentation to ensure transparence and accountability of the used systems before the authorities;

– clear and detailed information for the user

– guarantees for the constant human supervision;

– high level of security and robustness of the applied technological solutions.

For cases where the risk is limited less measures are required, which allow the user to be aware in due time, that artificial intelligence is applied, to decide whether to continue to use the procedure or service in question. This is applicable for instance for chabots.

For certain cases it is accepted, that the risk from the application of artificial intelligence is minimal and no specific protection measures are required – such as video-games or spam filters.

The responsibility with regard to the activity of artificial intelligence is assigned to its creators, as well as to the persons, who are using it. Respectively the control for observing the requirements of the regulation for a trustworthy and reliable artificial intelligence is entrusted to the national market surveillance authorities, whereas the creation of a European Artificial Intelligence Board is also proposed, along with the adoptionm of codes of conduct  and the creation of regulatory sandboxes for responsible innovation.

The specific requirements of the regulation shall be clear after the final text is adopted.