People deserve an answer
When a machine influences a decision that affects someone, the organisation should be able to explain the observed path to that outcome.
People should be able to ask what happened, who was responsible and whether a system acted as intended. Responsible organisations should be able to answer without asking for blind trust.
Machines may take on more responsibility. Accountability must remain human.
Make consequential AI activity understandable to the people who operate it, review it and live with its outcomes.
Read the longer view →It is the ability to reconstruct an action, challenge an explanation and keep responsibility attached to people even as software becomes more autonomous.
When a machine influences a decision that affects someone, the organisation should be able to explain the observed path to that outcome.
A model, vendor or software system can contribute to a decision. Accountability still belongs to the people and institutions that put it into use.
A trustworthy record names its sources, exposes its limits and can be checked by someone other than the team that produced it.
Transparency is the first door. The product underneath is a record of observed AI activity, prepared so another person can examine it and form their own view.
See how the record works →The mission becomes useful when it helps a real person examine a real system and make a better decision.