EU bill classifies artificial intelligence systems according to four risk levels, with riskier applications to face tougher requirements regarding transparency and data sharing