Japan AI Guidelines C3.2: 3)(2) Intervention of human judgment

What does C3.2 require?

So that AI outputs do not lack fairness, consider uses in which people step in at appropriate points instead of letting the AI decide alone, with measures so that human judgment is not swayed by automation bias; introduce a process that analyses and deals with the purpose, constraints, requirements and decisions of the AI system or service in a clear and transparent way to check whether bias has arisen; and, mindful of unintended, imperceptible and latent bias (bias not visible in training data, such as sensitive attributes inferred from proxies or intersectional bias), set direction after dialogue with stakeholders of diverse backgrounds, cultures or fields.