I learned the true meaning of accountability on a Sunday morning, many years ago, when I had to go into the office to fix a calculation mistake made by a member of my team.

I was an insurance pricing manager at the time. My team performed premium calculations that brought in $2 billion of revenue. It was complex work, spread across multiple staff, and the margin for error was incredibly low.

When my boss spotted an error in a table my team had produced, it was all I could do not to say: "Don't blame me, she's the one who stuffed up."

But that's not how accountability works. I was the manager, so the error was mine.

And that's how I came to be the one sitting in a stuffy office at 9:30am on a Sunday, to ensure the numbers that went out first thing on Monday could be guaranteed to be correct.

Working with AI is exactly the same.

You can delegate responsibility for a task, but you can never delegate the accountability.

If the AI gets it wrong and your stakeholders act on that AI output, you're the one who has to answer for it. Not the tool.

So, the obvious lesson here is to always check AI outputs.

Yeah, we all get that.

However, here's something most conversations about checking AI outputs miss. Something I learned in the aftermath of that Sunday morning.

I never wanted to have to go into the office on a Sunday ever again, so I built a checklist for reviewing my team's work - but I also built my intuition.

I knew our total premiums should come in at around $2 billion. I could recite our top 10 policyholders by heart.

If a number looked off, or an unfamiliar name appeared in the top 10, I knew immediately that something was wrong - before even reviewing the calculation.

That intuition didn't come from the checklist. The checklist was just the record of what I already knew. It came from deep familiarity with the problem: years of working in the domain, understanding what the numbers meant and knowing what "right" felt like.

This is what I think about when I hear people saying they're reviewing AI-created work.

Because if you don't have that internal alarm bell - if you don't know what a correct answer looks like - how will you know there's something off when AI hands you the wrong answer and calls it done?

The accountability doesn't transfer. Neither does the expertise that makes accountability meaningful.

If you don't have that expertise yet, now is the time to start building it.

Talk again soon,

Dr Genevieve Hayes