Anyone downplaying the impact of AI is looking at things from a practical perspective rather than getting inside the heads of decision makers.
It doesn't matter whether AI is ready to do peoples' jobs. All that matters is whether leaders who make budgets *think* it is. If they all get around a table and decide "We can cut 500 heads this year, AI will cover the productivity loss" or "We can keep the workforce flat, AI will help our workers be more productive" then that will be the reality.
This is quite condescending towards "decision makers".
Bad decisions lead to bad outcomes. Outside of SV, businesses haven't burned to the ground precisely because those decision makers have limited the impact of AI to stuff nobody was doing anyway like writing notes and filling in gaps for documentation.
Everyone saying AI is an excuse: <whatever> is always an excuse. Companies build up marginal people over time: people who aren’t overtly fire-worthy, but who aren’t core contributors either. The pressure builds up, like the conditions for an avalanche, over time. When it’s at a critical point an inciting event can be relatively minor. And then the list gets made, and if there’s no one who says, “We really need Bob,” Bob goes on the list.
It can be about the proximal cause, but it doesn’t have to be at all.
All of that said, AI is going to directly cause job loss, I’m calling it now. Not as much as the doomsayers predict, but more than most people expect.
> All of that said, AI is going to directly cause job loss, I’m calling it now. Not as much as the doomsayers predict, but more than most people expect.
Unless there is some limit to model development we can't currently foresee, plain economics will see to it that white collar job losses will be close to total. Likewise blue collar if we don't find a limit to spatial AI and robotics development.
The problem with all these discussions is that no-one rubbishing the job-apocalypse forecasts can say why or how progress will peter out - beyond pointing to economic limits ("it's a bubble") which won't apply over longer terms. Given the pace of progress the last few years, and this inability to say why job losses won't scale with the tech, anyone ruling them out is either wish thinking, or showing a staggering failure of imagination.
If there's a reason the losses will be "Not as much as the doomsayers predict", say what it is.
> If there's a reason the losses will be "Not as much as the doomsayers predict", say what it is.
OK, I'll make an attempt:
1. AI capabilities have obviously exploded at an amazing rate over the past few years, but I think most people in the field view a lot of the "Bobby grew a ton by age 13, he'll obviously be 100 feet tall in a few years"-type analysis of a few years ago to be wrong. Or, at least, people see limits to current AI tech, and that completely new/unknown approaches will eventually be needed. Of course, AI never really gets worse, and I can easily see a lot of problems (e.g. hallucination rates) being greatly improved even with just existing tech.
2. I think tons of jobs will get obliterated. I think you'd have to be insane to go into radiology as a med student right now. Tons of people currently make their living driving, and robots can already do a lot of that. More broadly, there are already lots of jobs that are basically "data in, one unambiguously correct answer out" that AI will excel at. Creative jobs will also be affected. I read a report recently about how AI dramas are all the rage in China, and they're already displacing jobs for actors.
3. But I disagree that losses will be "close to total". There will still be a strong desire for humans to actually decide on the "what do we make?", even if it's mostly made by the AI/robots. For a particular depressing and macabre analogy, think of the American South during slavery. Even though most of the actual labor was done by slaves (in the analogous case AI/robots), there were still jobs directing the work to be done.
So I guess I'm in the "it will be a shit show of epic proportions" camp, but that's still not as bad as some of the worst doomsaying I've seen.
3 seems the strongest of these arguments. The 'other techs plateaued' argument ignores that this is the first tech ever to convert electricity into thought and agency. There isn't a precedent for AI, and until intelligence stops scaling with compute, any assumption of a limit - that may not even exist - being reached in the few years left before jobs are wiped out is arbitrary faith.
I agree though, that business leadership roles will still survive - with some industries, wherever some principle or vision needs to be maintained - with the normal little adjustments humans might prefer to feel out for themselves. Perhaps also politicians, sportsmen, escorts, priests, anyone involved in spiritual and new age therapy. But this is still close to total. And aside from ownership/leadership which can earn in power and influence, it isn't clear how any of these jobs would be paid.
Hinton was probably right, even in 2016. When a med student chooses their residency, they want to choose a career that will be around in 40 years. The tech obviously wasn't there in 2016, but it is tantalizingly close today. I have a family member who is a radiologist who works for a group that deploys AI tools as an adjunct, and is was pretty eye opening the first time that tool caught a critical finding he missed.
Interestingly, there is currently a huge shortage of radiologists because the tech (but, more importantly, the regulatory framework) isn't quite there yet, but again people choosing a medical specialty aren't looking at today or a year or two out, they want a career that will sustain them into old age after investing years and hundreds of thousands in training. People are worried at what the landscape will look like in 5 years, let alone 20, 30 or 40.
The limit to job loss is completely unrelated to AI capabilities. Rather it is social.
There is a breaking point where if enough people end up jobless it will lead to genuine bloody uprisings. I won't pretend to know where exactly that point is, but I am more than happy to state that it is before "nobody has a job anymore" is reached.
The pace of progress is precisely why many people qualitatively assume the curve will flatten soon: J curves are generally (obviously not always) unsustainable.
You're arguing that the limits will appear because they usually do. (Correct my paraphrase if this is unfair.) Apart from being blind faith, this argument is oblivious to the fact that capability so far has scaled directly with compute and that the experts developing the models expect that to continue.
AI will certainly cause job loss. It already has (regardless of to what degree it was an excuse for something else in many cases). I agree with what you say, though.
The big question to me is whether the people who lost those jobs will have better opportunities in the future. That's kind of up to all of us.
I’m working at a major fortune 100 company, they don’t even have MS foundry set up. There are probably ~10 ai use cases or less in the company now. So much hinges on figuring out who will grant permissions to create a resource or app ID…. The companies jobs are safe because they are sooo far behind and self-limiting.
You need to look at this in context of longer timeframes; recent college graduates have had higher unemployment rates since 2019, well before AI, and the diff has increased since covid. I suspect it's more about a decline in overall growth and hiring and AI now gives a convenient excuse, but the trend has been in play for a while.
According to the Federal Reserve data, the unemployment rate for recent grads only started to exceed the rate for all workers around 2022 and has been widening since.
It's hard to separate impact of workers being replaced by AI from an impact of a recession (or stagnation). It's not obvious how AI impacts employees with various experience: on one hand a senior is better at spotting AI hallucination one other hand a junior using AI can do much more than a junior was able to do a couple years ago for a lower (if adjusted for inflation) than a couple years ago salary.
I struggle to see how WFH, especially as that was far more common from 2020 to 2023 than 2023 to 2026
Rather than the post-covid slump we've seen globally
> WFH makes supervision, monitoring, and on-the-job learning harder
It makes it different. In many ways it makes it easier, if you have the right supervisors and mentors working in the right way.
The larger impact would be hotdesking. Going to an office and not sitting anywhere in your team makes collaboration harder than working from home.
The requirement to move job to progress in remuneration harms retention, and thus reduces willingness to invest in a junior, but it's the expectation to move job after 2-3 years.
It’s mostly manufactured hype to keep the AI bubble going.
Nearly all the tech layoffs are simply companies trimming fat that was there the whole time. Outside the tech bubble folks have increasing disdain towards AI and can smell AI generated content from a mile away.
The tech is cool, and useful, but massively overhyped. Now there’s a mad rush for companies to IPO before the music stops.
I think this is spot-on. AI frankly isn't good enough to replace people en mass. It may never be. The limits of the technology seem self evident. It's useful, but it's not magic.
> I think this is spot-on. AI frankly isn't good enough to replace people en mass.
It doesn't actually need to be good enough for people to make decisions like it is good enough. My SO left her previous job because a McKinsey spreadsheet had fanciful notions of how deeply they could do job cuts in her org. The remaining team was so overwhelmed that most people quit up to some ridiculously high management level.
So you might have people replaced en-mass, due to leaders believing the hype. Maybe they correct for it a little, but not enough to dial in to the reality.
What's that saying that goes something like "People tend to vastly overestimate the impact of technology in the short run, and vastly underestimate its impact in the long run." A lot of these comments here are making feel disdain for both extremes of the debate. I agree that AI has been used as an excuse for a lot of layoffs, but it also unambiguously has resulted in real layoffs - the Washington Post ran an article years ago, a few months after ChatGPT came out, about how some copywriters had been replaced by AI.
The technology isn't magic, it has limitations, but it's not that hard to see a large percentage of current jobs being made obsolete by AI - the BLS has already identified many such roles; https://www.inc.com/soren-kaplan/the-bureau-of-labor-statist...
I don't think it will kill all white color jobs, but on the other side, declaring it a nothing burger seems like a lot of willful blindness given it already has replaced some people and AI's capabilities have grown legions since then.
Anyone downplaying the impact of AI is looking at things from a practical perspective rather than getting inside the heads of decision makers.
It doesn't matter whether AI is ready to do peoples' jobs. All that matters is whether leaders who make budgets *think* it is. If they all get around a table and decide "We can cut 500 heads this year, AI will cover the productivity loss" or "We can keep the workforce flat, AI will help our workers be more productive" then that will be the reality.
This is quite condescending towards "decision makers".
Bad decisions lead to bad outcomes. Outside of SV, businesses haven't burned to the ground precisely because those decision makers have limited the impact of AI to stuff nobody was doing anyway like writing notes and filling in gaps for documentation.
Everyone saying AI is an excuse: <whatever> is always an excuse. Companies build up marginal people over time: people who aren’t overtly fire-worthy, but who aren’t core contributors either. The pressure builds up, like the conditions for an avalanche, over time. When it’s at a critical point an inciting event can be relatively minor. And then the list gets made, and if there’s no one who says, “We really need Bob,” Bob goes on the list.
It can be about the proximal cause, but it doesn’t have to be at all.
All of that said, AI is going to directly cause job loss, I’m calling it now. Not as much as the doomsayers predict, but more than most people expect.
It already has, the stuff we used to have teams for, like translations and content asset creation for CMSes, are nowadays mostly done by AI.
Check any modern CMS, this is now a basic feature.
> All of that said, AI is going to directly cause job loss, I’m calling it now. Not as much as the doomsayers predict, but more than most people expect.
Unless there is some limit to model development we can't currently foresee, plain economics will see to it that white collar job losses will be close to total. Likewise blue collar if we don't find a limit to spatial AI and robotics development.
The problem with all these discussions is that no-one rubbishing the job-apocalypse forecasts can say why or how progress will peter out - beyond pointing to economic limits ("it's a bubble") which won't apply over longer terms. Given the pace of progress the last few years, and this inability to say why job losses won't scale with the tech, anyone ruling them out is either wish thinking, or showing a staggering failure of imagination.
If there's a reason the losses will be "Not as much as the doomsayers predict", say what it is.
> If there's a reason the losses will be "Not as much as the doomsayers predict", say what it is.
OK, I'll make an attempt:
1. AI capabilities have obviously exploded at an amazing rate over the past few years, but I think most people in the field view a lot of the "Bobby grew a ton by age 13, he'll obviously be 100 feet tall in a few years"-type analysis of a few years ago to be wrong. Or, at least, people see limits to current AI tech, and that completely new/unknown approaches will eventually be needed. Of course, AI never really gets worse, and I can easily see a lot of problems (e.g. hallucination rates) being greatly improved even with just existing tech.
2. I think tons of jobs will get obliterated. I think you'd have to be insane to go into radiology as a med student right now. Tons of people currently make their living driving, and robots can already do a lot of that. More broadly, there are already lots of jobs that are basically "data in, one unambiguously correct answer out" that AI will excel at. Creative jobs will also be affected. I read a report recently about how AI dramas are all the rage in China, and they're already displacing jobs for actors.
3. But I disagree that losses will be "close to total". There will still be a strong desire for humans to actually decide on the "what do we make?", even if it's mostly made by the AI/robots. For a particular depressing and macabre analogy, think of the American South during slavery. Even though most of the actual labor was done by slaves (in the analogous case AI/robots), there were still jobs directing the work to be done.
So I guess I'm in the "it will be a shit show of epic proportions" camp, but that's still not as bad as some of the worst doomsaying I've seen.
3 seems the strongest of these arguments. The 'other techs plateaued' argument ignores that this is the first tech ever to convert electricity into thought and agency. There isn't a precedent for AI, and until intelligence stops scaling with compute, any assumption of a limit - that may not even exist - being reached in the few years left before jobs are wiped out is arbitrary faith.
I agree though, that business leadership roles will still survive - with some industries, wherever some principle or vision needs to be maintained - with the normal little adjustments humans might prefer to feel out for themselves. Perhaps also politicians, sportsmen, escorts, priests, anyone involved in spiritual and new age therapy. But this is still close to total. And aside from ownership/leadership which can earn in power and influence, it isn't clear how any of these jobs would be paid.
> I think you'd have to be insane to go into radiology as a med student right now.
Hinton said the same thing in 2016. Maybe it is finally different this time?
People also said you would be crazy to go into tech after the dot-com crash.
Hinton was probably right, even in 2016. When a med student chooses their residency, they want to choose a career that will be around in 40 years. The tech obviously wasn't there in 2016, but it is tantalizingly close today. I have a family member who is a radiologist who works for a group that deploys AI tools as an adjunct, and is was pretty eye opening the first time that tool caught a critical finding he missed.
Interestingly, there is currently a huge shortage of radiologists because the tech (but, more importantly, the regulatory framework) isn't quite there yet, but again people choosing a medical specialty aren't looking at today or a year or two out, they want a career that will sustain them into old age after investing years and hundreds of thousands in training. People are worried at what the landscape will look like in 5 years, let alone 20, 30 or 40.
Why do you think the jobs directing the work will be dine by us instead of by huge data centers with manager ais?
The limit to job loss is completely unrelated to AI capabilities. Rather it is social.
There is a breaking point where if enough people end up jobless it will lead to genuine bloody uprisings. I won't pretend to know where exactly that point is, but I am more than happy to state that it is before "nobody has a job anymore" is reached.
The pace of progress is precisely why many people qualitatively assume the curve will flatten soon: J curves are generally (obviously not always) unsustainable.
You're arguing that the limits will appear because they usually do. (Correct my paraphrase if this is unfair.) Apart from being blind faith, this argument is oblivious to the fact that capability so far has scaled directly with compute and that the experts developing the models expect that to continue.
> that the experts developing the models expect that to continue.
We must be listening to different experts then. One small example, Apple's widely discussed paper on the limits of current approaches: https://machinelearning.apple.com/research/illusion-of-think...
AI will certainly cause job loss. It already has (regardless of to what degree it was an excuse for something else in many cases). I agree with what you say, though.
The big question to me is whether the people who lost those jobs will have better opportunities in the future. That's kind of up to all of us.
I’m working at a major fortune 100 company, they don’t even have MS foundry set up. There are probably ~10 ai use cases or less in the company now. So much hinges on figuring out who will grant permissions to create a resource or app ID…. The companies jobs are safe because they are sooo far behind and self-limiting.
Unemployment rates for recent college graduates stand at around 5.6%, well above the level for all workers.
I think we just found the first evidence of AI's expected influence on the labor market.
You need to look at this in context of longer timeframes; recent college graduates have had higher unemployment rates since 2019, well before AI, and the diff has increased since covid. I suspect it's more about a decline in overall growth and hiring and AI now gives a convenient excuse, but the trend has been in play for a while.
https://www.newyorkfed.org/research/college-labor-market#--:...
According to the Federal Reserve data, the unemployment rate for recent grads only started to exceed the rate for all workers around 2022 and has been widening since.
It's been creeping up even before the release of chatgpt.
https://www.economist.com/content-assets/images/20250621_FNC...
It's hard to separate impact of workers being replaced by AI from an impact of a recession (or stagnation). It's not obvious how AI impacts employees with various experience: on one hand a senior is better at spotting AI hallucination one other hand a junior using AI can do much more than a junior was able to do a couple years ago for a lower (if adjusted for inflation) than a couple years ago salary.
> It's hard to separate impact of workers being replaced by AI from an impact of a recession (or stagnation).
Doubly so since both phenomena interact: recessions are the time when employers will push hard for disruptive automation.
A recent paper believes work from home reducing entry level gigs fits the data better, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6787638.
I hate it, but the X thread is the easiest review piece I can find, https://x.com/pj_lambert/status/2057477629528150369.
https://xcancel.com/pj_lambert/status/2057477629528150369
https://xcancel.com/pj_lambert/status/2057477629528150369
I struggle to see how WFH, especially as that was far more common from 2020 to 2023 than 2023 to 2026
Rather than the post-covid slump we've seen globally
> WFH makes supervision, monitoring, and on-the-job learning harder
It makes it different. In many ways it makes it easier, if you have the right supervisors and mentors working in the right way.
The larger impact would be hotdesking. Going to an office and not sitting anywhere in your team makes collaboration harder than working from home.
The requirement to move job to progress in remuneration harms retention, and thus reduces willingness to invest in a junior, but it's the expectation to move job after 2-3 years.
Looking at this graph
https://upload.wikimedia.org/wikipedia/commons/a/ab/Unemploy...
I think you'd struggle to draw any conclusions about AI.
Note your quote was "all workers", not "workers of same age with or without a degree"
In the aftermath of 2008, recent graduates hit 7-8%, but their contemporaries without a degree hit 15%
It’s mostly manufactured hype to keep the AI bubble going.
Nearly all the tech layoffs are simply companies trimming fat that was there the whole time. Outside the tech bubble folks have increasing disdain towards AI and can smell AI generated content from a mile away.
The tech is cool, and useful, but massively overhyped. Now there’s a mad rush for companies to IPO before the music stops.
I think this is spot-on. AI frankly isn't good enough to replace people en mass. It may never be. The limits of the technology seem self evident. It's useful, but it's not magic.
> I think this is spot-on. AI frankly isn't good enough to replace people en mass.
It doesn't actually need to be good enough for people to make decisions like it is good enough. My SO left her previous job because a McKinsey spreadsheet had fanciful notions of how deeply they could do job cuts in her org. The remaining team was so overwhelmed that most people quit up to some ridiculously high management level.
So you might have people replaced en-mass, due to leaders believing the hype. Maybe they correct for it a little, but not enough to dial in to the reality.
What's that saying that goes something like "People tend to vastly overestimate the impact of technology in the short run, and vastly underestimate its impact in the long run." A lot of these comments here are making feel disdain for both extremes of the debate. I agree that AI has been used as an excuse for a lot of layoffs, but it also unambiguously has resulted in real layoffs - the Washington Post ran an article years ago, a few months after ChatGPT came out, about how some copywriters had been replaced by AI.
The technology isn't magic, it has limitations, but it's not that hard to see a large percentage of current jobs being made obsolete by AI - the BLS has already identified many such roles; https://www.inc.com/soren-kaplan/the-bureau-of-labor-statist...
I don't think it will kill all white color jobs, but on the other side, declaring it a nothing burger seems like a lot of willful blindness given it already has replaced some people and AI's capabilities have grown legions since then.