The wide-ranging proposal, which was leaked ahead of its official release, also promises strict new rules for high-risk AI.
These include algorithms used in policing and recruitment.
Experts say the rules are vague and full of loopholes.
Military use of AI and systems used by authorities to maintain public safety are no exception.
Advertisement
The list of AI systems proposed to be banned includes:
Those designed or used in a way that manipulates human behavior, opinions or decisions…causing a person to act, form opinions or make decisions that are detrimental to them
AI systems used for indiscriminate surveillance are widely used
AI systems used for social scoring
Those that use information or predictions and individuals or groups to target their weaknesses
European policy analyst Daniel Leufer said on Twitter that the definitions are very open to interpretation.
“How do we determine what is harmful to someone? Who assesses this?” he wrote.
For AI deemed high risk, member states must impose more oversight, including the need to designate assessment bodies to test, certify and inspect these systems.
Any company that develops prohibited services or fails to provide correct information could face fines of up to 4% of its global revenue, similar to fines for GDPR violations.
Examples of high-risk AI include:
Systems that establish dispatch priorities for emergency services
Systems that determine admission to or allocation of people to educational institutions
Recruitment algorithms
Those that assess credit worthiness
Used to carry out personal risk assessments
Crime prediction algorithms
The proposals should be “extended to cover all public sector AI systems, regardless of their assigned risk level”, Mr Leufer added.
“This is because people generally do not have a choice as to whether or not to interact with public sector AI systems.”
