AI Didn’t Take Our Jobs. It Turned Them Into Checking – Social Europe

This post was originally published on this site.

KEY INSIGHTS
  • Verification is the new job: From IKEA’s call centres to Norwegian newsrooms and software teams, humans now check what machines produce instead of producing it themselves.
  • The checking goes unmeasured: In a survey of more than 2,000 American workers, 57 per cent of managers had already fixed or redone colleagues’ AI-reliant work.
  • Reliability breeds inattention: US regulators urged airlines in 2013 to have pilots fly by hand more often because automation had let their skills erode.
  • Answers displace learning: Turkish students who practised with an AI tutor that gave answers did worse on the exam than those who never used it, while a hints-only version did no harm.
  • Better tools carry a hidden bill: Once checking becomes ritual, the thinking stops without anyone deciding that it should.

When IKEA introduced its customer-service bot in 2021, the bot steadily took over the routine questions; by this summer it was handling nearly three-quarters of the customers who used it. IKEA did not lay off the 8,500 people in its call centres. It retrained them, over two years, to handle the hard cases and to sell interior design over video.

This is the story everyone cites, and it deserves to be. But look at what the retained humans now do: they take the calls the machine could not resolve, which is to say they handle what the machine got wrong or could not judge. They are the check.

That is the pattern I keep seeing, and it is not the one we were promised. The promise was that AI would do the work and we would do something better with the time. What is happening in many places is quieter. The machine does the work, and we check it. The checking has become the job.

Norwegian newspapers were early to put AI-written summaries at the top of their articles. In September 2023 the country’s largest paper, VG, ran a summary reporting that the footballer Erling Haaland had been shot. He had not. He had posed as a Viking for a photographer, and the model had misread a word.

The sports editor apologised and said the paper had never intended to kill him in print. The lesson newsrooms drew was not to stop summarising but to put a person between the model and the page; at the papers I work with, an editor now reads every summary before it is published. The journalists did not lose their jobs to the summariser. They acquired a new one: reading what it wrote.

A friend who runs software teams tells me the same thing has become the norm in programming. The engineers generate the code, then they verify it, then they test it. Some of his people call the job boring now. The programmer who used to build is the inspector of a building that appeared overnight.

Making Has Become Cheap, Checking Has Not

I am not above this. I write research grant applications for a living, in a second language, and AI has made that work easier than it has ever been. Before these tools I would write three to five versions of a proposal before submitting it. This year I wrote one with an assistant at my side, and there were more than a hundred versions.

I tuned, I read, I asked the model to criticise what it had helped me write, I asked colleagues the same, and when the two disagreed I had to decide who was right. Did I get faster? I never measured the time. What I know is that I spent most of it not writing but verifying.

The surveys say I am typical. In a poll of more than 2,000 American workers this spring, 57 per cent of managers said they had already had to fix or redo work that a colleague had produced by leaning too heavily on AI, and 77 per cent of respondents said they now review AI-assisted work more carefully than work done by hand. Making things has become cheap. Checking them has not, because a check still takes a person, and a person still takes the time it takes.

Reliable Tools Are the Dangerous Ones

So far this is an argument about cost, and it has a comforting ending: the tools will get better, the errors will become rarer, and the checking will shrink until it is a formality. I think that ending is the dangerous part.

Consider the editor who reads AI summaries all day and finds nothing wrong in any of them for a month. She does not stop reading, but the eye slides. The summary that was always right has earned a trust that no summary deserves, and the one error in a thousand arrives at a reader who has stopped expecting it. Pilots learned this first: in 2013 US regulators urged airlines to have crews fly by hand more often, because the systems that had made flying safer had also let the skill of flying erode.

The same thing happens inside a person who is learning. A perfect answer is a reason to trust the answer, not a reason to understand it. In an experiment published last year, secondary-school students in Turkey who practised maths with an AI tutor improved dramatically during practice and then did worse on the exam, without the tutor, than students who had never used it. A version of the tutor that gave hints instead of answers did no such harm. The difference that mattered was not accuracy. It was who did the thinking.

That is the bargain we are making in every workplace that has turned production into verification. While the tools are imperfect, the checking keeps us thinking; it is tedious, but we are still in the work. When the tools become reliable enough, the checking becomes a ritual and then a line in a policy, and the thinking stops without anyone deciding that it should. We will have kept the humans in the loop and emptied the loop of anything human.

I do not want worse tools. I want us to be honest about what the better ones cost. Part of the bill is money. Most of it is harder to count: the engineer who no longer knows why the code works, the editor who no longer reads, the professor who has a hundred versions and is not sure what he learned from any of them.

IKEA did the right thing by keeping its people. The harder question is what the rest of us keep when the checking, too, is done for us.

Leave a Reply

Your email address will not be published. Required fields are marked *