Ethics · Checking AI's Work

One risk hides in plain sight. You ask an AI to summarise a report, and back comes a clean answer with a precise figure. It reads perfectly. That figure is nowhere in the report.

Why AI sounds right when it is wrong

An AI does not know when it is right. It predicts the next plausible word, so an invented figure reads exactly like a real one, in the same fluent, confident tone. That is the trap. How sure it sounds tells you nothing about whether it is true. When AI produces something plausible but false, we call it a hallucination, and nothing on its face signals that it happened.

You do not check everything to the same depth. Match the check to the stakes: a throwaway brainstorm needs little, but a number heading into a client report or a board paper needs verifying. AI predicts; you judge. Checking is where that judgement does its work.

A thirty-second check in practice

Say you ask an AI to draft a reply telling a customer your return window. It states thirty days, clearly and confidently. Before you send, you open the actual policy page: it is fourteen days. One look, half a minute, and a wrong promise never reaches the customer. Make that the habit: never let an AI be the final source on a fact that will reach a customer, a client, or a decision-maker. Verify it against something you trust first.

Key Takeaway

An AI answer sounds the same whether it is right or inventing, so the check is what protects you.