When AI Saves Time, Who Gets the Time Back?

Written by Louize Clark

When AI saves time, who gets that time back?

By Louize Clark

It sounds like a simple question, and most of us have already felt the answer, even if we haven't said it out loud. You finish something in half the time it used to take, and before you've even noticed, something else has filled the space: another email, another meeting, another job that was waiting for a gap. So what really happens to our time when AI gives some of it back? And what happens to the part of it we were supposed to spend checking what AI has given us?

I've been thinking about this since a social worker told me something this week that I can't stop coming back to. She has been carrying a caseload of fifty-five. About six months ago she started using AI to take some of the pressure off, and she told me she was finally coming out of burnout. For the first time in a long while, she could see some light at the end of the tunnel.

I'd want that for her whether or not she gets through a single extra case. If you've been going home every night feeling like there's still a day's work waiting for you, getting some of that time back matters. You might finish on time for once. You might have a bit more of yourself left when you walk through the door.

But we do tend to fill time when it becomes available, and organisations are very good at it. If a team is getting through the work more quickly, it's easy to see why a manager would allocate another case or book another visit. Nobody would call that a decision about AI. It just looks like good management. And somewhere in that same day, someone still has to check what the AI has produced.

The Ada Lovelace Institute's research on AI transcription in social care shows how easily this happens. One manager wanted to test whether the tools could allow social workers to go from one home visit a day to two. I can see why you'd want to try that, because more time with families is the whole point. But a second visit also means a second set of notes, and those notes still need checking. In the same research, social workers described spending anything from a few minutes to an hour reviewing what the tools produced. So when the day is planned around two visits, how much of that hour has been allowed for?

It isn't only social care. In last Friday's AI Law Report we covered Flux Drive v Rong, a US case in which one side argued that the other side's lawyers should have billed fewer hours because they had used AI and online research tools. The appeals court found no evidence that the hours were excessive and pointed to the work the lawyers still had to do, including verifying the authorities and arguments they relied on. It's easy to see how much quicker something is to produce. You have to look at the rest of the job to know how much time you've actually saved.

And the rest of the job lands on a person, usually at the worst possible moment.

You've got five minutes before a meeting and a document you haven't read. You ask AI to summarise it, and what comes back makes sense. You've used the same tool all week and it's been good. You have time to read the summary, but have you got time to open the original and see what it has left out? Say one of the points doesn't look quite right. Now you have another decision to make. Do you walk into the meeting and say you need more time before you can give an answer, when people are waiting on you? And if you do that more than once, how will it go down with the person who manages you?

This is where I think the conversation about keeping a human in the loop needs to spend a lot more time. We're usually clear about who is supposed to check the work. We're much less clear about what that person can actually do on a busy afternoon, when they've been making decisions all day and the next meeting is about to start. A tool that has been helpful all week is easy to trust at that point, and research on automation bias has found that trust, workload and time pressure all play a part in how readily people rely on automated advice. That isn't carelessness. It's what happens to any of us when there's more to check than there is time to check it.

Then there's what happens if they do check. At the University of Bedfordshire last week, I ran the hands-up exercise I wrote about in my last paper, where people keep a hand raised only while they can honestly say yes. When I asked them to keep their hands up if their teams had enough information and time to review what AI was producing properly, hands came down. When I asked whether their targets and the way they managed people supported employees who questioned AI, nobody had a hand left in the air.

Think about two people doing the same job. One goes back to the source document, finds that something doesn't match and sends it back to be looked at again. The other reads the summary, sees nothing obviously wrong and approves it. At the end of the month, if you're looking at how much work they got through, the second person looks quicker, and the first might be asked why they're behind. So what do you think they'll do next time, when they're already behind and something looks a bit off?

They still know how to check. They may even have the authority to stop the work. But they also know who they'll have to explain it to on Friday. I wonder how often we tell people to use their judgement and then measure them in a way that makes it harder to do exactly that.

Which brings me back to the social worker, and the relief she told me about. I want her to keep it. I also want her to be able to stop and read something properly when it doesn't look right, without having to find that time at home or explain why she's done one fewer visit. Both of those come out of the same hour, and at the moment I'm not sure anyone has decided which one gets it.

If we're relying on people to make that call, perhaps we should start by asking them what happens on the days when they do.

This is the first paper in a series on the human in the loop. I'll be looking at what people need to know to question AI, what happens when they do, and what changes when they spend years working this way.

AI Policies UK helps organisations understand how AI changes people, systems and operations, including what people need to exercise judgement at work. Get in touch at info@aipolicies.uk.

Sources

Louize Clark
Next

Nobody Kept Their Hand Up