Professional work has traditionally rested on two roles. A preparer produces the work. A reviewer examines it. When questions arise, the reviewer can ask how a figure was reached, which evidence was used and why a particular judgment was made. Two minds engage with the work, each answerable for a different part of it.
Now some organizations are placing AI in the preparer’s seat, reassured by a popular phrase: “human in the loop.” The workflow diagram barely changes. The machine prepares, the human reviews. Yet something important has disappeared.
AI can answer questions about its output, but its explanation is another generated output. It may describe a plausible path to the result without providing a verifiable record of the path actually taken. Whatever it says, the machine bears no consequence if the work is wrong. The preparer’s chair appears occupied, but no accountable person is sitting in it.
Consider a chef asked to approve a finished dish. A great palate may judge its taste. It cannot reconstruct what happened in the kitchen: where the ingredients came from, what was substituted, which shortcuts were taken or what should never have entered the pot. Yet the reviewer is now expected to certify the whole process from the finished plate.
So what is the basis of review when the reviewer did not prepare the work and has no accountable preparer to question?
Tasting the output is insufficient. Keeping a human in the loop cannot mean placing one at the end merely to approve the result. The human must examine the evidence, assumptions and material path to the answer closely enough to make the judgment their own. AI may perform most of the work, but accountability for the conclusion remains with the person who accepts it.
When AI prepares the work, the human in the loop is no longer the second accountable mind in the process. That human is the only one.
Frank Ng is a retired NASDAQ CEO, who co-authors this column with his son Ryan after publishing their book Hey AI, Let’s Talk!
