William Watson: Chatbots are changing everything and nothing

Yahoo

22-05-2025

The novelist Anthony Horowitz has the 'diary' page in this week's Spectator magazine. The format is amusing occurrences and casual musings as the writer wends his way through his week. Halfway down the column, Horowitz recalls how as a teenager he used to 'slip into' the Old Vic theatre to see plays, experience 'something close to magic' and be left 'breathless.'
But then the next entry abruptly announces: 'The last paragraph was written by ChatGPT.' Horowitz had asked it to write 100 words on theatre tickets in the style of Anthony Horowitz. He then analyzes where it did and didn't succeed. 'If I became breathless in a theatre, I'd expect St. John's Ambulance to remove me quickly,' and so on.
But you could have fooled me! In fact, it did fool me, and it fooled most readers, I'd guess, which was Horowitz's point. We're well beyond the stage where AI produces oohs and aahs just for putting a noun and verb in each sentence and making sure they agree in number. It's now operating at a high level of fluency (though I should add that no bots were abused in the writing of this column.)
It's impossible to imagine where something so powerful will take the world — though I expect Evan Solomon, our new minister of artificial intelligence, will spend lots of money trying. An under-wagered possibility is that after being turned upside down, the world will end up looking more or less the same. This would be consistent with Solow's Paradox, economics Nobelist Robert Solow's famous 1987 observation that: 'You can see the computer age everywhere but in the productivity statistics.'
In a new study of AI adoption in Denmark, Anders Humlum of the University of Chicago and Emilie Vestergaard of the University of Copenhagen quote that line from Solow. And they come to a similar conclusion, as indicated by their study's title, 'Large Language Models, Small Labour Market Effects.' (Would a bot have got to such a taut summary of the message? I think not.)
Not every occupation lends itself to using ChatGPT or similar chatbots. So the researchers look at only 11 where it's likely to be most handy, including: IT, HR and legal professionals, accountants, teachers, journalists and five others. It turns out Denmark is a very digitized place. Every Dane has 'a digital mailbox that Statistics Denmark can use to distribute survey invitations.' Moreover, Danes seem to check their digital mailboxes. The researchers sent out 115,000 survey invitations and got 25,000 completed responses covering 7,000 workplaces — a big sample and good response rate for this sort of thing. (Completed surveys were entered in a draw for a tax-free cash prize though we're not told how much.)
Beyond its paradoxical bottom line, the paper's other (to my mind) surprising finding was just how much chatbots are already being used. Almost two-thirds of workers have used one 'at least a few times' and almost 20 per cent use them daily. Firms 'are now heavily invested,' with 43 per cent of workers 'explicitly encouraged to use them' and only six per cent not allowed to. Firms also do training: 30 per cent of employees have had some. When firms are involved, take-up is higher and there's less of a gender gap. When firms are neutral there's more take-up from men. (One Norwegian study of students found a gender gap largely reflected 'male students continuing to use the tools even when explicitly banned.') Firm-wide investments are 'particularly widespread in journalism and marketing and more limited in teaching.'
Journalists (or their bots!) told the researchers they use AI to brainstorm 'story ideas, angles or interview questions,' to draft content, to fact-check and edit, to summarize 'research materials or interview transcripts' and, surprisingly, to ensure 'AI-generated content abides by journalistic ethics and standards.' They use a bot to decide whether their bot has been ethical?
What's been the effect of using the new technology? Average reported time saving is 2.8 per cent, which seems low, given how powerful the bots are. What do people do with the time they save? Mainly other tasks. Also somewhat more of the same task. And more or longer breaks or leisure time. It seems no one answered 'mindless screen-scrolling' during the freed-up time, though we all know what a problem that now is.
New technology allowing workers to turn to different tasks is a common effect and helps explain why automation typically doesn't displace labour wholesale: firms find new things for their workers to do. Which helps explain the labour market effects, which are: pretty much nothing. The researchers asked people directly whether 'they perceive AI chatbots to have affected their earnings.' No, said 99.6 per cent of respondents.
William Watson: Get the Ozempic! Cabinet has grown by almost two-thirds in two months
William Watson: Free trade is being replaced by crony trade
What people perceive isn't always true, of course. But in this case Denmark's digital connectedness allowed the researchers to check on hours, earnings, total wages, total employment and so on in the firms where bots are used most. And nothing budged.
It's early days yet but the papers' last line and the study's bottom line is that 'two years after the fastest technology adoption ever, labour market outcomes — whether at the individual or firm level — remain untouched.'