Imagine raising a child with a rulebook. Don't lie to your teacher. Don't take your sister's toys. Say please and thank you. Good rules, but on day five a friend asks them to keep a secret that involves someone getting hurt. The rulebook is silent.
Now imagine you also taught them that honesty matters, that people who cannot protect themselves deserve protection, and that when they are torn they should ask whether they would be comfortable if everyone could see what they were about to do. That child can navigate the secret situation because you gave them something that transfers.
AI skills have the same problem.
Rules are necessary, not sufficient
"Always cite your sources," "never make up data," and "flag confidence below 60%" are useful instructions. They are also finite. They catch the failures you anticipated.
Values are generalisable. Truthfulness means an honest "I don't know" is worth more than a confident guess. Humility means staying within scope. Integrity means showing the trail behind a conclusion. Those principles cover situations your rulebook never predicted.
This is the philosophy behind Constitutional AI: give a system principles, ask it to critique its response against them, and revise before delivery. You do not need to implement the formal method to use the idea.
Six values I would encode
Truthfulness. Humility. Integrity. Healthy self-doubt. Fairness. Courage.
Rules are the guardrails on the road. Values are the ability to navigate when the road ends.
This is part of the Parenting Your AI series, a practitioner's guide to building AI skills that are safe, effective, and worth trusting. Written from inside enterprise AI systems by someone who has spent years diagnosing what goes wrong when AI meets real work at scale.
Read the full series at KnowledgeManagement.ie