If you've been in corporate finance, though, you know how the typical CEO thinks, right?
Quarter to quarter at best, if even that. Pivots whenever revenues endanger their "bonus targets" (those RSU's don't get allocated themselves, you know).
So what happens when any master of business admin sees the promises whispered into their ears, echoing within their head chambers? Right: GenAI and LLM can do basic coding on at least a "junior / associate" level? Let's review our P&L for associate salaries and costs... and I can basically strip away at least 50% -> to be safe, let's keep some of the top performers!
Next thing you know VP of HR gets the call to "cut 15% of the lower level workforce in relevant departments -> you go do the nitty-gritty. I wanna see results in 3 months.
CEO: "Next quarter is coming and I do have an annual earnings review soon. Imagine the great news for our margins!"
You think the same CEO thinks even one step ahead about economic 2nd order, 3rd order etc.? Or they're empathic for the grander societal impact, aka "give a damn about their workforce on a people level"?
If Satya Nadella can slim operations by tooling AI Agents in their coding group you can't have your company not doing it, right, how would that look?
And that, ladies and gents, is how tech layoffs are happening right now across the board.
I must also say these are kind of ironic, because the coding / tech bros thought themselves invincible since they are the crop of the cream, making all these big, beautiful LLMs work.
Basically any western developed market is "consumer-driven" and if consumption drops off a cliff or grinds to a halt slowly and agonizingly, as it already is happening across the western developed world, of course it'll hit tax receipts. Of course it'll go on to hit social security. It'll destroy the little restaurant you so *looooved* going to, but are now scratching your head in front of a "Permanently closed" sign... how did that happen? Alfredo was doing a killer Bolognese, I thought he printed money!
Hyper-financializing everything got us here and it'll unwind... somehow, one way or another. AI is just a nail in the coffin of hyperfinancialization dreams, ie. we can commoditize everything, even chunks of work can now be "bought to spec" from any GenAI.
The most fun part is when you connect the big energy picture to this "dream of AI doing our work" - is 1.1 GWh gonna cut it for meta? How many GWh do you need to commoditize a full work model?
I wonder who the first politician will be banging the drum to tax AI output. Logical next consequence, unless we don't care about societal structures anymore.
Great AI article. Having worked in finance for a long time and having heard a great many "game-changing" sales pitches—like Excel dying in high finance (no), virtual reality becoming a new office (no), and so on—I just don't see AI as a revolutionary change. Based on my experience and what I have read, it is a great efficiency tool with a high error rate that still requires a lot of hand-holding. Let's see what the future brings. My guess: more of the same, with a new tool (AI) in the toolkit.
This is not a good analysis. You have the causality backward. GDP is how much we produce. Gross domestic PRODUCT. The only thing you need to believe in order for GDP to increase 10% while employment falls 20% is that the AI enables us to produce 10% more with only 80% of the workforce. That is plausible on its face.
The equation you trot out just points out that everything we produce has to go somewhere. Either the products are consumed, turned into capital assets, bought by government, or exported. If we produce more, consumption or investment will increase to match what’s being produced. That’s what market clearing prices do. The GDP equation is not a bottoms up build-up to figure out how much we produce. It’s just a way we calculate how much we’ve produced by what we can measure.
I understand where you’re coming from but I think you may not be understanding the calculation.
The equation is the actual, formal math driving GDP. How much we produce isn’t relevant unless it’s consumed. If we produce more but demand does not increase (and the most relevant demand here is consumer spending), than clearing prices will go down (basic supply/demand curve), and the absolute value of consumption will not rise in spite of the increased production.
Government spending would need to increase massively to plug the gap. There’s a growing case for universal basic income, funded by the state, which makes this possible. Unsure if the tech titans would argue for this.
Which is why we will need a UBI or a program that pays people who were displaced by AI or robots their usual salary for a number of years. We will need to let go of the insistence that people must work a 40-hr week.
Hmmm.. if the world became 20% more productive tomorrow, and 10% of workers were directed into unemployment, those still working would be earning much more and returns to capital would also increase a lot.
In this scenario, aggregate consumption levels would probably grow, even with fewer workers overall.
If things do play out as the psycho from Anthropic hopes, i reckon there’ll be some work around protecting him and his cronies from the millions of angry and aimless people whose lives, families and communities he’s played a role in upending.
I’d agree that returns to capital would absolutely increase though I’m less certain on average wages growing in that particular scenario. It’s also not clear to me that even slightly above average wage growth would immediately offset the hit to demand from unemployment going up 2.5x.
If there’s a certain class of skilled workers that are required to get maximum productivity out of AI, they’ll be in a great position to demand a share of the productivity windfall.
They might, but it’s not clear to me that it would be that broad a category relative to all of the many classes that would be negatively impacted. The economy is massive and if you’re disrupting dozens of fields of fairly highly paid individuals the wage gains in what remains would have to be quite a bit to make the aggregate wage increase. If we get some semi-strong AI outcome where attorneys, diagnostic physicians, programmers and host of other highly paid industries see both job losses and decreased wages (which would be true even for people remaining in the field because the job losses would create an at least temporary asymmetry between supply and demand), the wages in that new field growing would need to be really pretty unprecedented.
I’m not saying consumption resulting from increased productivity will definitely offset consumption lost through unemployment, i’m just suggesting its theoretically possible.
My official stance when it comes to predicting what AI does is “what the fuck would I know?”
My suspicion is that we end up with a less just, less free and more violent world. Happy to be proven wrong.
That’s fair. I’m a strong believer that Yogi Berra’s discussion of predictions is definitive. I think the strong case AI future would be one where you see major changes under the hood of GDP that creates productivity growth and high unemployment but it takes a few years for that to result in widespread economic growth, which is what you need to hypothetically get 10% GDP growth. My rough understanding of a decent part of 19th century economics is that that was the pattern during the Industrial Revolution. It may not hold here.
One answer would be this: at 20% unemployment there would be massive surplus labour. That would put downward pressure on wages. The few (thousands) highly skilled AI workers are unlikely to be numerous and highly paid enough to offset the many (millions) of unemployed earning 0. Certainly, gdp per capita would take a hit.
Great piece. Thanks. Add that bots don’t pay taxes, don’t spend money in the community, don’t consume. Infrastructure and public services would become underfunded.
I just don’t see how, with AI serving the ownership class, wealth and power don’t just exponentially consolidate while the rest of the population drifts downward into subsistence.
I agree with you on all of this. I imagine what is going through the heads of CEOs who believe this is possible is that a modest and compromised UBI will become a reality as a bailout for them
Computers are massively overrated. What happened to the ‘paperless office’? Some of our greatest technological achievements such as the moon landings, nuclear power plants and major transportation infrastructure were achieved with little or no computing power and certainly not office computers. And we couldn’t repeat those achievements today without extraordinary expense, probably because people are too busy answering emails.
I forget which economist noted this (Gordon? Cowan?): the effects of the computer revolution are seen everywhere except in GDP statistics. Or something to that effect.
I hate to contradict you but the US has a long history of increasing GDP in the face of decreasing wages. If you think that American wages are buying American products with ‘good as gold’ $$s then I have worrying news to deliver. The tech bros could indeed achieve 10% GDP with 20% unemployment. Why do you think they all support UBI? Because UBI is gov money that replaces wages. How is it financed? Same way the economy’s been financed since Reagan - trade deficit, sale of debt to America’s colonies & geopolitical hegemony. The US could easily force Latam & parts of the Middle East to become its sweatshop (once China detaches), maintain dollar use wherever it has military bases & pay its people UBI while the rest of the world collapses. The FT & Washington Post would continue printing articles about America’s forthright growth & praise their great business successes. And the entire economy could become even more of a hollowed out shell where nothing works, people die on the streets but a handful of companies have vast profits because they are funded by the State (either directly or through UBI). I find it incredible that people don’t see that the US (& Anglo-Saxon deficit countries in general) have been doing this for 40 years already?? Why would it change?
The scenario sounds unlikely but it isn't "mathematically" impossible.
There are a couple channels that could imply higher levels of real consumption.
If we are already assuming vast AI-driven productivity gains, this will likely have some deflationary impact, so prices for goods and services could be lower. This stands to improve consumers' purchasing power even with stable nominal incomes, which complicates the net effect of job losses on households' real income. There is an outcome where this has a positive impact on real GDP growth.
There's also the wealth channel, where the corporate productivity gains flow back to consumers through dividend income and capital gains. i.e. there is a stronger redistribution effect through financial markets.
The new reality could be one of a higher equilibrium unemployment rate and higher consumption.
Also on a more pedantic note, 20% of the population could be unemployed, but the unemployment rate could still mechanically be lower if people are not actively job-seeking (perhaps they are supported by investment income). Not that this counters your points, but that's how I would interpret Amodei's statement that "20% of people don't have jobs". This could play out as a a one-time spike in the unemployment rate, rather than a prediction of a sustained 20% unemployment rate.
Don’t we claim that with every industrial revolution? When machines came to factories, there was also talk that people would only work 4-6 hours or we would need UBI. It’s a cultural issue: as long as we organise society, income and social status based on our professional life, people will find ways to keep working. Perhaps product lifecycles just shorten to erase the increased efficiency of AI.
If we’d truly want to benefit as a society, we’d need to change the role work and career play in our lives first.
If you've been in corporate finance, though, you know how the typical CEO thinks, right?
Quarter to quarter at best, if even that. Pivots whenever revenues endanger their "bonus targets" (those RSU's don't get allocated themselves, you know).
So what happens when any master of business admin sees the promises whispered into their ears, echoing within their head chambers? Right: GenAI and LLM can do basic coding on at least a "junior / associate" level? Let's review our P&L for associate salaries and costs... and I can basically strip away at least 50% -> to be safe, let's keep some of the top performers!
Next thing you know VP of HR gets the call to "cut 15% of the lower level workforce in relevant departments -> you go do the nitty-gritty. I wanna see results in 3 months.
CEO: "Next quarter is coming and I do have an annual earnings review soon. Imagine the great news for our margins!"
You think the same CEO thinks even one step ahead about economic 2nd order, 3rd order etc.? Or they're empathic for the grander societal impact, aka "give a damn about their workforce on a people level"?
If Satya Nadella can slim operations by tooling AI Agents in their coding group you can't have your company not doing it, right, how would that look?
And that, ladies and gents, is how tech layoffs are happening right now across the board.
I must also say these are kind of ironic, because the coding / tech bros thought themselves invincible since they are the crop of the cream, making all these big, beautiful LLMs work.
Basically any western developed market is "consumer-driven" and if consumption drops off a cliff or grinds to a halt slowly and agonizingly, as it already is happening across the western developed world, of course it'll hit tax receipts. Of course it'll go on to hit social security. It'll destroy the little restaurant you so *looooved* going to, but are now scratching your head in front of a "Permanently closed" sign... how did that happen? Alfredo was doing a killer Bolognese, I thought he printed money!
Hyper-financializing everything got us here and it'll unwind... somehow, one way or another. AI is just a nail in the coffin of hyperfinancialization dreams, ie. we can commoditize everything, even chunks of work can now be "bought to spec" from any GenAI.
The most fun part is when you connect the big energy picture to this "dream of AI doing our work" - is 1.1 GWh gonna cut it for meta? How many GWh do you need to commoditize a full work model?
I wonder who the first politician will be banging the drum to tax AI output. Logical next consequence, unless we don't care about societal structures anymore.
Great AI article. Having worked in finance for a long time and having heard a great many "game-changing" sales pitches—like Excel dying in high finance (no), virtual reality becoming a new office (no), and so on—I just don't see AI as a revolutionary change. Based on my experience and what I have read, it is a great efficiency tool with a high error rate that still requires a lot of hand-holding. Let's see what the future brings. My guess: more of the same, with a new tool (AI) in the toolkit.
It being like excel for more than just accounting pretty much sums it up
This is not a good analysis. You have the causality backward. GDP is how much we produce. Gross domestic PRODUCT. The only thing you need to believe in order for GDP to increase 10% while employment falls 20% is that the AI enables us to produce 10% more with only 80% of the workforce. That is plausible on its face.
The equation you trot out just points out that everything we produce has to go somewhere. Either the products are consumed, turned into capital assets, bought by government, or exported. If we produce more, consumption or investment will increase to match what’s being produced. That’s what market clearing prices do. The GDP equation is not a bottoms up build-up to figure out how much we produce. It’s just a way we calculate how much we’ve produced by what we can measure.
I understand where you’re coming from but I think you may not be understanding the calculation.
The equation is the actual, formal math driving GDP. How much we produce isn’t relevant unless it’s consumed. If we produce more but demand does not increase (and the most relevant demand here is consumer spending), than clearing prices will go down (basic supply/demand curve), and the absolute value of consumption will not rise in spite of the increased production.
Government spending would need to increase massively to plug the gap. There’s a growing case for universal basic income, funded by the state, which makes this possible. Unsure if the tech titans would argue for this.
But doesn’t universal basic income solve the riddle.
People are effective paid to be unemployed so they can continue spending.
It might but given that the US doesn't even yet provide universal health care, UBI seems a politically distant idea.
Tbf, during Covid, the US demonstrated it could dump a lot of cash if it wants to
Which is why we will need a UBI or a program that pays people who were displaced by AI or robots their usual salary for a number of years. We will need to let go of the insistence that people must work a 40-hr week.
Hmmm.. if the world became 20% more productive tomorrow, and 10% of workers were directed into unemployment, those still working would be earning much more and returns to capital would also increase a lot.
In this scenario, aggregate consumption levels would probably grow, even with fewer workers overall.
If things do play out as the psycho from Anthropic hopes, i reckon there’ll be some work around protecting him and his cronies from the millions of angry and aimless people whose lives, families and communities he’s played a role in upending.
I’d agree that returns to capital would absolutely increase though I’m less certain on average wages growing in that particular scenario. It’s also not clear to me that even slightly above average wage growth would immediately offset the hit to demand from unemployment going up 2.5x.
Why wouldn’t wages grow?
If there’s a certain class of skilled workers that are required to get maximum productivity out of AI, they’ll be in a great position to demand a share of the productivity windfall.
They might, but it’s not clear to me that it would be that broad a category relative to all of the many classes that would be negatively impacted. The economy is massive and if you’re disrupting dozens of fields of fairly highly paid individuals the wage gains in what remains would have to be quite a bit to make the aggregate wage increase. If we get some semi-strong AI outcome where attorneys, diagnostic physicians, programmers and host of other highly paid industries see both job losses and decreased wages (which would be true even for people remaining in the field because the job losses would create an at least temporary asymmetry between supply and demand), the wages in that new field growing would need to be really pretty unprecedented.
I’m not saying consumption resulting from increased productivity will definitely offset consumption lost through unemployment, i’m just suggesting its theoretically possible.
My official stance when it comes to predicting what AI does is “what the fuck would I know?”
My suspicion is that we end up with a less just, less free and more violent world. Happy to be proven wrong.
That’s fair. I’m a strong believer that Yogi Berra’s discussion of predictions is definitive. I think the strong case AI future would be one where you see major changes under the hood of GDP that creates productivity growth and high unemployment but it takes a few years for that to result in widespread economic growth, which is what you need to hypothetically get 10% GDP growth. My rough understanding of a decent part of 19th century economics is that that was the pattern during the Industrial Revolution. It may not hold here.
One answer would be this: at 20% unemployment there would be massive surplus labour. That would put downward pressure on wages. The few (thousands) highly skilled AI workers are unlikely to be numerous and highly paid enough to offset the many (millions) of unemployed earning 0. Certainly, gdp per capita would take a hit.
Surely you could ask that same question of the economy post-1970s when productivity and real wages began to diverge…
This is simple yet compelling logic. The tech bros did not take MACRO 101, clearly.
Great piece. Thanks. Add that bots don’t pay taxes, don’t spend money in the community, don’t consume. Infrastructure and public services would become underfunded.
I just don’t see how, with AI serving the ownership class, wealth and power don’t just exponentially consolidate while the rest of the population drifts downward into subsistence.
I agree with you on all of this. I imagine what is going through the heads of CEOs who believe this is possible is that a modest and compromised UBI will become a reality as a bailout for them
Computers are massively overrated. What happened to the ‘paperless office’? Some of our greatest technological achievements such as the moon landings, nuclear power plants and major transportation infrastructure were achieved with little or no computing power and certainly not office computers. And we couldn’t repeat those achievements today without extraordinary expense, probably because people are too busy answering emails.
I forget which economist noted this (Gordon? Cowan?): the effects of the computer revolution are seen everywhere except in GDP statistics. Or something to that effect.
I hate to contradict you but the US has a long history of increasing GDP in the face of decreasing wages. If you think that American wages are buying American products with ‘good as gold’ $$s then I have worrying news to deliver. The tech bros could indeed achieve 10% GDP with 20% unemployment. Why do you think they all support UBI? Because UBI is gov money that replaces wages. How is it financed? Same way the economy’s been financed since Reagan - trade deficit, sale of debt to America’s colonies & geopolitical hegemony. The US could easily force Latam & parts of the Middle East to become its sweatshop (once China detaches), maintain dollar use wherever it has military bases & pay its people UBI while the rest of the world collapses. The FT & Washington Post would continue printing articles about America’s forthright growth & praise their great business successes. And the entire economy could become even more of a hollowed out shell where nothing works, people die on the streets but a handful of companies have vast profits because they are funded by the State (either directly or through UBI). I find it incredible that people don’t see that the US (& Anglo-Saxon deficit countries in general) have been doing this for 40 years already?? Why would it change?
One minor point in the CEO’s comments-AI will not cure cancer either (at least in the near-mid term). It could accelerate drug discovery though.
I have much less knowledge of medical research but will take your point!
The scenario sounds unlikely but it isn't "mathematically" impossible.
There are a couple channels that could imply higher levels of real consumption.
If we are already assuming vast AI-driven productivity gains, this will likely have some deflationary impact, so prices for goods and services could be lower. This stands to improve consumers' purchasing power even with stable nominal incomes, which complicates the net effect of job losses on households' real income. There is an outcome where this has a positive impact on real GDP growth.
There's also the wealth channel, where the corporate productivity gains flow back to consumers through dividend income and capital gains. i.e. there is a stronger redistribution effect through financial markets.
The new reality could be one of a higher equilibrium unemployment rate and higher consumption.
Also on a more pedantic note, 20% of the population could be unemployed, but the unemployment rate could still mechanically be lower if people are not actively job-seeking (perhaps they are supported by investment income). Not that this counters your points, but that's how I would interpret Amodei's statement that "20% of people don't have jobs". This could play out as a a one-time spike in the unemployment rate, rather than a prediction of a sustained 20% unemployment rate.
Don’t we claim that with every industrial revolution? When machines came to factories, there was also talk that people would only work 4-6 hours or we would need UBI. It’s a cultural issue: as long as we organise society, income and social status based on our professional life, people will find ways to keep working. Perhaps product lifecycles just shorten to erase the increased efficiency of AI.
If we’d truly want to benefit as a society, we’d need to change the role work and career play in our lives first.
CF South Africa