Every selection process makes a claim: that the people chosen were chosen for reasons connected to the program. Fairness is whether that claim survives inspection.
Three things make it survive. Criteria fixed before review begins. Evidence attached to every score. A record of who decided what, and when.
AI helps with the second one. When a score points to the exact sentences that produced it, a coordinator can agree, disagree, or overrule, and the reasoning stays visible either way.
What AI must never do is decide. A model can surface, summarise, and structure. The choice belongs to the person accountable for the outcome.