Manifesto for real AI transparency and openness
AI transparency realities
The transparency and openness of artificial intelligence, algorithms, and automation, and of the organisations that design, develop and deploy them, poses significant opportunities and benefits, as well as challenges and risks.
AI, algorithmic and automation transparency efforts lack substance
AI transparency and openness principles
Having weighed the limitations and consequences of transparency with organisational, technical, individual, societal and other risks, every organisation designing, developing or deploying AI, algorithmic, or automation systems should adopt the following four principles:
Responsive to feedback, complaints, and appeals
Provide reasonable third-party expert access to data, code, and model
Make third-party audit and investigatory reports public
To all relevant users, including the disabled, the elderly, non-internet users, and other disadvantaged people
Visible at the first point of contact with the system, and across all relevant touchpoints.
October 4, 2021