Generative AI is not replacing jobs or hurting wages at all, economists claim

Instead of depressing wages or taking jobs, generative AI chatbots like ChatGPT, Claude, and Gemini have had almost no significant wage or labor impact so far – a finding that calls into question the huge capital expenditures required to create and run AI models. In a working paper released earlier this month, economists …

[IanRS]

Another bandwagon

AI is going to be a major transformation factor in the way everything works, just like blockchain was going to be too.

AI works well for pattern matching/recognition, so has potential there. For generative work you may as well employ a parrot - you get recognisable noises copied from elsewhere but no understanding.

[yosemite]

Really?

So, this is based on data from 2023/2024? At the speed the AI industry is moving right now that might as well have been a report from the stone age. Yes, corporations are like super tankers and will take time to adopt and incorporate AI into their practices but, it ain't slowing down.

This report is going to age very quickly

[Jan Ingvoldstad]

Re: Really?

Perhaps they should use a generative language model to confabulate a report from 2026, then.

[PghMike]

When was 2024?

Dude, we're less than four months past 2024. Not sure how that qualifies as 'the stone age'.

[doublelayer]

Re: When was 2024?

You're both right. The data isn't very old, especially given how long it generally takes to create new data. How many people have jobs is often on at least a month's delay as it is, and information about how companies are changing the jobs their employees do is often delayed longer than that if it's explained at all, so the data they've used is probably the latest they have access to. However, if modern LLMs had actually taken over and replaced employees, it wouldn't show up in the last set of data from 2022-2023; the LLMs of that age were much worse, struggling to string a paragraph together.

We won't get the full story until some companies actually try replacing workers with LLMs for a while. Having seen them used, I'm not expecting large changes. While there are a lot of places using them, the quality is still a problem, meaning that companies using them and expecting quality usually need to spend about as much time testing, rerunning, and correcting LLM output as they did doing the thing from scratch in the first place. Various people I know or work with have arrived at different places on the spectrum of how much LLMs are used, and I do know someone who uses LLMs frequently and nonetheless produces good code (he does complain that he has to try five times and then correct manually to get workable code, so I don't know how efficient he is). I'm still waiting to see how badly it fails when a company decides to trust LLM output more readily.

[juice]

Re: Really?

> At the speed the AI industry is moving right now that might as well have been a report from the stone age.

The last I checked, LLM models are still stuck in a dilemma: they're too unreliable for high value work which requires accuracy, and too expensive to use for low-value tasks which don't require accuracy.

I don't see that changing anytime soon.

[Filippo]

Winter is coming.

[Anonymous Coward]

"Winter is coming."

Time to dig up your tulip bulbs before the first frosts of Winter. :)

I imagine the rise of AI pulled the rug from under the careers of more than one cryptocurrency spiv but the retraining required to pimp AI would be pretty minimal - snake oil is snake oil, a substance which if not actually toxic is invariably ineffacious.

If only single digit percentage increases in overall productivity are being observed in practice, the enormous costs (economic and environmental) incurred in constructing these models must fail even the most elementary cost-benefit analysis.

[Shuki26]

Re: "Winter is coming."

I think single digits increase was relevant last year, but in 2025, even more. For sure my job can be made much more efficient with a more mature AI. The various AIs are already allowing me to work quicker and also do less thinking. Layoffs will continue and whoever is left will HAVE TO use it even more to make up for the reduced head count.

[Theodore.S]

Re: "Winter is coming."

| do less thinking.

Use it or lose it... :-)

[Throg]

Wrong Jobs?

Maybe it’s because the jobs listed there actually involve creativity and the ability to think rather than just churn out meaningless text all day?

Let’s think. Which jobs do involve that sort of almost mindless activity? Ah yes. Middle management. Sales and marketing. Estate agents. Maybe even lawyers.

Anyone up for building a B Ark?

[Burgha2]

Re: Wrong Jobs?

"Anyone up for building a B Ark?"

You do know what happened to the last planet that did that, right?

[Neil Barnes]

Re: Wrong Jobs?

We know what happened to the planet where the B-ark _landed_...

[Burgha2]

Re: Wrong Jobs?

"We know what happened to the planet where the B-ark _landed_..."

We actually do know what happened to the planet where the A and C ark people remained

"Fate of Golgafrincham

A notation in the Guide about Golgafrincham after the departure of the B Ark states that the entire remaining population subsequently died from a virulent disease contracted from a dirty telephone."

[Anonymous Coward]

Maybe even lawyers?

"Maybe even lawyers"

It has not gone well for the lawyers who have tried and been caught...

[Philo T Farnsworth]

Re: Maybe even lawyers?

Just wait till we have LLM judges.

[FelixReg]

Productivity numbers have a problem

Use of computers famously didn't show up in productivity numbers in the 1900's, either.

[Anonymous Coward]

Re: Productivity numbers have a problem

Use of computers famously didn't show up in productivity numbers in the 1900's, either.

An interesting observation which a retrospective analysis might be quite revealing assuming 1960-99 is intended rather than 1900-9. ;)

The simplest explanation is that computers didn't alter workplace productivity until they were significant products in themselves.

When PCs and dedicated word processors became sufficiently inexpensive they were deployed to typing pools which increased typist productivity some of which was lost in having to fFaff around with unreliable printers (and early networks or floppy disk sneakernet) but as PC were quickly deployed on to other employees' desks the typing pools rapidly shrank then vanished leaving the remaining employees and the lower echelons of management to spend hours producing documents that a dicto- or steno-typist could produce in minutes. (Not to mention the incompetent stuffing about with Viscalc and Lotus spreadsheets and DBase or FileMaker nonsense from which we have really never escaped - vide today's peddling of AI / nocode.)

In the absence of the advent of the internet and later cell smart phones I am not sure productivity did increase appreciably - certainly industries and workplaces underwent significant structural change but I am not at all certain the net benefit was positive.

The internet and later smartphones only really increased the opportunity and efficiency of flogging existing tangible and intangible† products which also, almost serendipitously, coincided with the tsunami of Chinese tat into western markets which I suspect accounts for a great deal of apparent "productivity" increase.

_{† not sure to which pornography belongs but I have read accounts for 30% of internet traffic. Not that one should believe everything one reads.}

[John Miles]

Re: Documents that a dicto- or steno-typist could produce in minutes

I read a study around the time typists were disappearing - the bottom line was though typist could produced the typed document much quicker, the actual effort of producing the words for the the documents was generally same on a PC by a unskilled typist compared to drafting it by hand and it could be sent out immediately, while sending it to typing pool and need to review what they typed added time. Now add it is much easier to make quick changes compared hand drafts in response to other's reviewing it, it was much quicker to get a letter/document out. Some people make a song and dance out of it on PC, but they'd have done similar writing it by hand.

I've had documents retyped by typing pool/document team - despite having provided it as a word processor file in right format.

[yoganmahew]

Re: Productivity numbers have a problem

Tell that to the airline industry. Automation has cut headcount massively over the years. Productivity in the industry is, IMO, misattributed to plane technologoy or deregulation, but in reality, an airline is a complex set of computer systems with wings. Banking too is mostly IT and has been for a long time.

Having been around since the dawn of the microcomputer and it's regular use in the late 1900s, I can attest to the increase in productivity of tasks such as documentation; it is taken for granted now that failing to do documentation is much less cost saving than failing to do documenation was when it had to go to the typing pool, come back, be proof-read and marked up in pencil, returned to the typing pool and eventually filed in a room full of spiders. Now you have to find project overruns from somewhere else, QA usually.

[Yet Another Anonymous coward]

Re: Productivity numbers have a problem

On the other hand have you seen Boeing's aircraft since they stopped having rooms full of men at drawing boards with slide rules.

[adsp42]

Re: Productivity numbers have a problem

Re: Boeing's aircraft

I'd say it's a coincidence.

It's just the rite of passage, every successful capitalist corporation goes. The fat cats are bleeding it to death until the planes fall out of the sky.

[OverclockedMind]

Re: Productivity numbers have a problem

Interesting. Maybe if they had computers in the 1900's, the black hand group could have taken their rage out by playing Doom instead of shooting the crown price, and the course of the 20th century changed.

[Burgha2]

Had a meeting...

Had one of those group all day talk-fests today. The type where you write things on flip charts and senior managers say how great it was at the end of the day.

At one point someone had the idea of taking a photo of one of those flip charts and getting co-pilot to summarise the top four points.

One was a straight repeat of the largest thing written up. It misinterpreted an acronym, replacing it with something from an entirely unrelated activity and the rest was just recasting apparently randomly selected points in "better" management speak.

The interesting thing was the facilitator seemed to think it was great.

[Yet Another Anonymous coward]

Re: Had a meeting...

So what if you had just done that at the start and all gone to the pub instead?

[Big_Boomer]

So, massive investment but no real effect on productivity,..... so what is the ****ing point? Other than better pattern recognition than humans I have yet to see a single real world use for "AI". Perhaps it will show just how much of a waste of space so many people in business really are but I don't hold out much hope for that either. Even as a toy "AI" is seriously flawed, but at least there it doesn't really matter, except that people believe it's bullsh!t without bothering to check. Just this last weekend a non-technical friend stated a "fact" that he had been told by a certain browser manufacturers "AI" that turned out to be a complete load of hogwash.

[Eclectic Man]

Occupations affected by AI Chatbots

"the labor market impact of AI chatbots on 11 occupations, covering 25,000 workers and 7,000 workplaces in Denmark in 2023 and 2024.

Many of these occupations have been described as being vulnerable to AI: accountants, customer support specialists,

...

"AI chatbots have had no significant impact on earnings or recorded hours in any occupation," the authors state in their paper."

They may be looking at the wrong occupations. Maybe they should check ones that are not being replaced or 'enhanced' by the use of AI, but at ones that have to cope with the effects on people of implementing AI. My experience of AI chatbots is that they are unhelpful, and I may need counselling or professional psychological help after attempting to 'use' them. I'd look for an increase in the hours spent by psychotherapists, CBT practitioners, and bar staff, consoling and sympathising with people who've been 'ChatBotted'.

(Try using the Thames Water Chatbot to get single occupancy tariff.)

OTOH, I recently had a very frustrating experience with '%*& from Customer Services' regarding a bodged job a work person gather company and provided to me in June 2023 (the person used sticky tape to 'repair' a toilet siphon, which is contrary to the Customer Rights Act 2015, but because it is 'out of warranty' and I cannot prove their 'plumber' did it, they have rejected my complaint. This was a 'human' person, without, it seems to me, compassion, empathy, or any interest in future business from me. (Moral: Photograph everything when a work person comes to your home, parts that have failed, the part replacing the party the failed, etc. so you have actual proof of what they did.)

[Johnb89]

But it enables more low quality work!

I'm reminded of the use of powerpoint decades ago, over hand written transparencies: More low quality work. Not better work.

As an example: Person 1 has 3 points to make in an email, and uses AI to write it out long form to look better/longer. Person 2 receives email and uses AI to distil the long text to the salient points. Let's assume both do a good job. Lots of AI use, value add = none.

[Yet Another Anonymous coward]

Re: But it enables more low quality work!

There is an AI tool that generates an entire PowerPoint presentation from a few words in an email.

Then another AI tool at the recipient summarizes the PowerPoint into a few words and emails it to you

A huge productivity improvement

[cookiecutter]

AI has cost peopel their jobs

CEOs are the dumbest breed on the planet and believe all the marketing hype.

AI is a photocopier - a bad one that gives you sheets with something totally different on there that what you put in. However, it's being sold as the new coming of Jesus. And CEOs are dumb arses who believe this stuff.

You have a whole layer of management that literally do nothing. Even in engineering firms, the first people to get the boot are the ACTUAL engineers while there is always money for more Project Managers and MBAs who do nothing but email each other and sit in meetings about meetings.

Customers hate chatbots, but they are rolled out.

AI hallucinates, makes stuff up, but none of this is known by the people who want to look clever & "with it". How many morons have been going to the press with their "We've rolled out AI and fired 1000 people" comments? When you have AI Agents doing SOC work, making medical decisions, etc AND you fire 80% of your staff, the human element is removed too, which removes your controls.

How long until we see someone die because an AI agent made some shit up, or missed something a human would have seen and we get the inevitable investigation and "our thoughts and prayers are with the family right now & we do apologise. no one could have seen this happening"

I mean can you imagine Fushitsu rolling out an AI version of Horizon?

[Anonymous Coward]

AI promises to be as transformational as the Segway.

[DS999]

Of course they haven't

They haven't had long enough to have a real impact. This is like checking in 1995 whether the internet is having a big impact on the bottom line of businesses.

I'm not comparing the potential impact of AI with the impact of the internet, but there was a huge dot com bubble in the mid 90s until 2000 just based on "inflated expectations". Then a big crash when a recession arrived and those expectations weren't realized. The expectations were realized, and then some, but it took a lot longer than 2000 to see the full impact. So the lack of major impact of AI so far, and the huge investments/bubbly valuations of everything "AI" does not have anything to do with whether AI will eventually have a massive impact or if ends up mostly a mirage.

Since we've got a big recession coming thanks to Trump the AI valuations are going to come back to earth, but that won't be "proof" AI will never replace jobs / affect wages in the future any more than the dot com crash proved that the internet was overhyped.

[KenWedin]

AI Definitely Replacing Translators, Editors, and Essay-Writing Tutors Already

Generative AI didn’t just impact my career — it delivered the final blow to a profession already hollowed out by decades of technological and economic indifference.

I was a copyeditor (inaccurately referred to as “proofreader”), a copywriter, a Japanese–English translator in fields ranging from finance and law to medicine and tech, and later, an academic writing and literature tutor. Each of those paths has now been eliminated.

Copyediting, my original field, was already being dismantled in the 1990s when newspapers and other publishers began offshoring editing work to India en masse, prioritizing cost over quality. By the time the Internet displaced print as the dominant medium of reading in the early 2000s, few cared any longer whether any text had been edited at all. That forced me into Japanese translation and eventually private tutoring.

However, potential clients increasingly preferred the costless convenience of Google Translate, regardless of its notorious inaccuracy, to the precision and cultural insight of a professional translator. Then, after a decade of tutoring, even that path was rendered obsolete when ChatGPT began composing competent essays on books my students had not bothered to read. They were overwhelmingly Chinese-Canadian students who cared only about STEM, regarding literary analysis and writing skills as irrelevant.

The economists cited in this article speak confidently of “no wage or labour displacement”, but their data are drawn from formal payrolls. I, like countless others working freelance or in private education, had disappeared from the labour economy long before the Danish tax records could register a ripple.

AI has already decimated my livelihood. It’s not that the needle hasn’t moved — it’s that they’re not holding the right compass.

[teebie]

"commitment to internal education and evangelism"

Evangelism doesn't seem like something you would need to do for a product that was non-shit.