When not to use AI

Choosing not to use AI can be a sound operational decision.

Key takeaways

• Do not use AI where accountability cannot be delegated.
• Human relationships and trust may be more valuable than speed.
• Poor evidence becomes more dangerous when expressed fluently.

Consequential decisions

Do not delegate legal, medical, financial, employment or safety decisions to an unaccountable model.

Sensitive relationships

Customers and staff may reasonably expect a real person for complaints, vulnerability, negotiation or complex support.

Poor source conditions

If the required facts are unavailable or unreliable, polished generation can make uncertainty harder to see.

Use a stop test

Stop when the work involves rights, safety, vulnerability, binding commitments or a decision that must be explainable by a qualified person.

Also stop when there is no reliable way to detect a wrong answer before it causes harm.

Choose the better alternative

Use a deterministic rule, approved template, conventional search, calculator, database query or direct conversation when it is clearer and easier to audit.

Action checklist

1. Identify who bears the consequence of an error.
2. Confirm whether a qualified person can review the result.
3. Check whether the source evidence exists.
4. Consider a simpler deterministic method.
5. Document why AI is or is not appropriate.

A sensible next step

Add a short “do not use AI” section to the team’s approved-use policy with examples from real work. Continue with the practical AI recipes, compare the reviewed AI tools or use the AI Finder to narrow your next decision.