The European Commission released the final draft of its proposal to regulate artificial intelligence and machine learning Wednesday, making it the first major governing body to propose serious limits on the use of artificial intelligence.
The proposal suggests banning the use of all AI systems that pose a serious risk to human health and human rights, including tools that would allow "social scoring" of populations by governments or systems that might direct children to perform dangerous activities. The law also would label a large swath of systems as "high-risk" and require strict rules and obligations before they can be put into use.
Some of the systems that qualify as high-risk include all biometric identification (facial recognition, genetic markers and iris scan databases, for example), as well as systems used for employment, immigration, education and law enforcement. These high-risk systems would be required to undergo a risk assessment, provide documentation and human oversight and verify the quality of the datasets used to train them, among other requirements.
The regulation lists a few categories that qualify as low or minimal risk (which would require less oversight), including chatbots and video game and spam filters.
The proposed law will face years of debate and votes in the European Council and European Parliament before it could pass in some form.