SAFE AI = SECURE
Has passed test and verification processes and is robust to systemic and well-informed attacks.
SAFE AI = OPEN SOURCE
Methods, code and test-results are accessible for everyone.
SAFE AI = IS SELF-CONSCIOUS
Understands its own role and if uncertainty appears it can e.g. refuse to act.
SAFE AI = CAN KEEP A SECRET
Is designed to respect privacy by being built on the principle of privacy by design.
SAFE AI = HAS CALIBRATED VALUES
Is debugged for stereotypes and bias and understands emotions.
SAFE AI = IS ACCOUNTABLE
Is transparent and communicative with a respect for the “right to explanation”.
SAFE AI = UNDERSTANDS SOCIAL RELATIONS
Understands social relations in addition to a user’s knowledge and competences.
SAFE AI = UNDERSTANDS POWER
Understands data, context and consequences of its actions.
Source:Professor Lars Kai Hansen, DTU Compute, The Technical University of Denmark