The US sees 3 quarters of productivity growth up 3.2% YoY. Now, it needs to ensure strong employment, growing fixed invests & a stable supply side to spearhead with AI to unlock productivity gains.
Lol...you are right Ryan. I tried hard to find an emoji for a muffin, and then I gave up and settled for the cupcake. It was getting late at night, so I made the hard creative choice. Glad you enjoyed the post.
You are right. Even with fiscal recklessness and growing populism, the US is still stronger than the rest of the world when it comes to strength in innovation, financial markets and military. Partly why, it continues to stay resilient, even when we see parts of Europe in recession and a mixed picture in Asia. I don't necessarily believe that US will remain at the center of the world order forever, but it still has a good shot in maintaining its stronghold over the next 5 years to a decade.
Historically I’ve been a strong believer that technological disruption creates net new jobs. I’m taking the under on that this time.
I’m curious what kind of upskilling the average person could do in order to compete with AI 5 years out. I really don’t see anything. I would prefer to be wrong.
I also think integration into business processes will ultimately be much faster than the internet was. We have gotten way way better at software integration in the last 30 years.
Thanks Brett. The way that I see it is that, it is completely possible that AI will have a competitive advantage over human beings over the coming years at a wide range of tasks. However, given that we will always have a finite amount of "compute", although compute is growing every year, what the future of work may look like will perhaps be perhaps deciding which tasks to leverage AI to maximize output, given finite compute vs. tasks that human beings have a comparative advantage at.
This is largely the McKinsey view of things. The argument is certain tasks just aren’t worth automating. Personally I don’t see compute really being finite. We’ve always managed to grow the level of compute to satisfy all use cases with a positive absolute value. I see that as different than your typical comparative advantage argument which has always been based on the assumption of finite resources.
I’ll argue for practical purposes that compute resources are infinite. Sure if may take time to ramp capacity, but if there is a need for 1000 times the compute we will absolutely get there.
If you are right and compute is finite as a practical matter then I agree that comparative advantage keeps people employed. If there is no comparative advantage I highly doubt there will be absolute advantage for very many tasks, and that list will narrow over time.
One of the arguments is that it’s expensive to train AI for relatively inexpensive tasks. Looking at what figure ai is doing right now I suspect that will not be true even 5 years out.
I’d bet you a muffin, but I’m not sure how we would even measure things.
Well I don't know about McKinsey, but yes, the idea of comparative advantage is based on the world view that we live in a world of finite resources. How do you think we will reach a point where compute will become infinite? My thinking is that should we reach such a point, human beings would be long obsolete, as energy prices will be bid up exponentially.
As far as infinite compute I didn’t mean literally. I mean for practical purposes meaning I don’t think the quantity of compute available will be a limiting factor in replacing cognitive human tasks with machines.
We can already see AI is going to take an astounding amount of energy, no argument there. I think what this probably means is nuclear power becomes necessary, but consider that you could roughly power every single server in googles global footprint on a single good sized reactor not counting cooling and other data center energy needs. That’s about a million servers from what I gather.
How many people can you replace with a million servers? I haven’t looked at the specific details of how many requests an LLM can run, but the training of models in AI are way more computationally intensive than running them. Just as a thought experiment If a single machine could replicate the work of 10 people those million servers could account for the cognitive labor of 5% of the US population give or take.
We aren’t there yet, and this assumes AI lives up to the current hype. We’ve had false dawns in the past e.g. expert systems, but if things work out as they appear I don’t think comparative advantage will be a limiting factor in swapping humans out for machines.
Great piece. Yes, productivity matters but is notoriously difficult to measure. Off the top of my head, I can think of two things we can do to dramatically improve productivity.
1) Stop taxing production. Land Value Taxes can replace income and corporate income taxes.
Fully agree with you J.K. Productivity is insanely hard to measure, especially when trying to account for exponential speed of technological innovation disrupting and creating. Also, great suggestions, something I hadn't thought of in that direction, but it completely makes sense. Glad you enjoyed the post.
Great read. Provides a good context why we are so focused on limiting China"s access to leading edge AI technology. If the AI revolution plays oot as we expect, US will continue dominate the world stage for the next few decades.
Thanks Sonny. Glad you enjoyed the post. You are right about the choice US (and some its allies) are making to limit China's access to leading edge chips and other technology. After all, while the US economy has weakened over the last few decades, it still maintains its stronghold in innovation, financial markets and military.
As to the US dominating the world stage on AI, currently and ironically so, most of the AI innovation from OpenAI to Anthropic are taking place under the umbrella of big tech giants/incumbents who are all US based. So, as long as the economics around the current state of network effects and ownership does not change, the US will continue to have an edge.
Great article Amrita. The big question is, how will China develop. A quick comparison. When I look at the automotive space established car makers develop a car somewhere between 5-7 years (a face lift in 2-3 years). China develops a new car in 2 years (with a lot of "devil" in the detail stuff still not fixed, which makes a car "good/better" in the long run).
Thanks Michael, glad you enjoyed the post. Definitely one of my lengthiest ever. You have a valid point with China. Though we have seen China's growth story decline over the last few years with its property deleveraging and the "common prosperity" agenda of Xi, along with an ageing population, I believe it can still engineer a comeback (albeit, that requires fixing a lot of internal issues). I believe that by limiting China access to high edge tech products, the US is trying to gain a competitive edge, though I will admit, I know very little about the automotive space and how AI may impact manufacturing times in both these nations and ultimately what the competitive landscape will look like. Would love to hear your thoughts on it.
Yeah, I know, we are long due for a call. Let's chat next week on Friday on 5th April, if that works for you? Anytime between 9-11 am PT is good for me.
This was an enlightening read. I genuinely appreciate the efforts of the writers who things together to make an amazing article such as this. I also came across this article but more on the bearish side.
Thanks Rommel for your kind words. Glad you enjoyed the post. Thanks for sharing the post, I fully agree with all the points that the article has talked about. In fact, in a lot of core macro posts, like the ones I have linked below, I highlight some of the same issues. Having said that, I take a more pragmatic approach to the possible long-term productivity boom that can be unleashed by AI and how it will affect productivity and shape the workforce (which I will talk about in my next Monday post).
In a perfect world, a boom in productivity should increase the overall welfare in the society, create jobs, improve sentiment and therefore help with reelection prospects.
However, what I am noticing (and this is several years in the making), is that the boom in productivity is disproportionately benefiting parts of society, sectors, states, cities, etc. And while this is common side effect of innovation for certain industries to benefit disproportionately, the gap is becoming wider and we can see that today in the growing divergence between the haves and the have-nots, and this is pulling down overall sentiment, which can be used against him by his opponent.
If AI starts writing NYT bestsellers and hit songs, we are lost. Elevator music and ad copy, fine, but if we can be fooled so thoroughly that we start weeping over AI generated novels and love songs, we might as well turn in our human cards.
I think it ultimately plays down to competitive vs. comparative advantage. While it is possible that AI can do all of these things, and may even get as good or even than human beings, there is always going to be a finite amount of "compute" power, which is AI's constraint. And as a result, over the coming years, I believe we have to make decisions as to where to leverage AI to maximize output given the finite amount of compute, vs. human beings doing the rest of the work, where we have a comparative advantage.
I often wonder that same thing. How good will it get? We may learn more about ourselves than we want to know as we push AI toward whatever its limits are. If I attended a performance of an unfamiliar symphony and cried my way though it the way I did last time I heard Tchaikovsky’s 6th live, then discovered it had been written by AI, what would that tell me about myself and my emotions? In this scenario it would be performed by humans, so there would be that blade of grass to cling to. Otherwise I might fall off the face of the earth.
So much concern about with the Baltimore bridge accident is if Maersk stock will be affected. Almost no concern about the total incompetence in our “Physical” world
It’s not all virtual. We still live in a physical reality.
That was a fun read. My only criticism is that that emoji was clearly a cupcake and not a muffin.
Lol...you are right Ryan. I tried hard to find an emoji for a muffin, and then I gave up and settled for the cupcake. It was getting late at night, so I made the hard creative choice. Glad you enjoyed the post.
I usually don't repeat "Never bet against America" but in the current macroeconomic environment, which one country is a big winner? The US, by far.
You are right. Even with fiscal recklessness and growing populism, the US is still stronger than the rest of the world when it comes to strength in innovation, financial markets and military. Partly why, it continues to stay resilient, even when we see parts of Europe in recession and a mixed picture in Asia. I don't necessarily believe that US will remain at the center of the world order forever, but it still has a good shot in maintaining its stronghold over the next 5 years to a decade.
Historically I’ve been a strong believer that technological disruption creates net new jobs. I’m taking the under on that this time.
I’m curious what kind of upskilling the average person could do in order to compete with AI 5 years out. I really don’t see anything. I would prefer to be wrong.
I also think integration into business processes will ultimately be much faster than the internet was. We have gotten way way better at software integration in the last 30 years.
Thanks Brett. The way that I see it is that, it is completely possible that AI will have a competitive advantage over human beings over the coming years at a wide range of tasks. However, given that we will always have a finite amount of "compute", although compute is growing every year, what the future of work may look like will perhaps be perhaps deciding which tasks to leverage AI to maximize output, given finite compute vs. tasks that human beings have a comparative advantage at.
This is largely the McKinsey view of things. The argument is certain tasks just aren’t worth automating. Personally I don’t see compute really being finite. We’ve always managed to grow the level of compute to satisfy all use cases with a positive absolute value. I see that as different than your typical comparative advantage argument which has always been based on the assumption of finite resources.
I’ll argue for practical purposes that compute resources are infinite. Sure if may take time to ramp capacity, but if there is a need for 1000 times the compute we will absolutely get there.
If you are right and compute is finite as a practical matter then I agree that comparative advantage keeps people employed. If there is no comparative advantage I highly doubt there will be absolute advantage for very many tasks, and that list will narrow over time.
One of the arguments is that it’s expensive to train AI for relatively inexpensive tasks. Looking at what figure ai is doing right now I suspect that will not be true even 5 years out.
I’d bet you a muffin, but I’m not sure how we would even measure things.
Well I don't know about McKinsey, but yes, the idea of comparative advantage is based on the world view that we live in a world of finite resources. How do you think we will reach a point where compute will become infinite? My thinking is that should we reach such a point, human beings would be long obsolete, as energy prices will be bid up exponentially.
As far as infinite compute I didn’t mean literally. I mean for practical purposes meaning I don’t think the quantity of compute available will be a limiting factor in replacing cognitive human tasks with machines.
We can already see AI is going to take an astounding amount of energy, no argument there. I think what this probably means is nuclear power becomes necessary, but consider that you could roughly power every single server in googles global footprint on a single good sized reactor not counting cooling and other data center energy needs. That’s about a million servers from what I gather.
How many people can you replace with a million servers? I haven’t looked at the specific details of how many requests an LLM can run, but the training of models in AI are way more computationally intensive than running them. Just as a thought experiment If a single machine could replicate the work of 10 people those million servers could account for the cognitive labor of 5% of the US population give or take.
We aren’t there yet, and this assumes AI lives up to the current hype. We’ve had false dawns in the past e.g. expert systems, but if things work out as they appear I don’t think comparative advantage will be a limiting factor in swapping humans out for machines.
And I could be wrong of course.
Great piece. Yes, productivity matters but is notoriously difficult to measure. Off the top of my head, I can think of two things we can do to dramatically improve productivity.
1) Stop taxing production. Land Value Taxes can replace income and corporate income taxes.
2) Ease zoning regulations
Fully agree with you J.K. Productivity is insanely hard to measure, especially when trying to account for exponential speed of technological innovation disrupting and creating. Also, great suggestions, something I hadn't thought of in that direction, but it completely makes sense. Glad you enjoyed the post.
Great read. Provides a good context why we are so focused on limiting China"s access to leading edge AI technology. If the AI revolution plays oot as we expect, US will continue dominate the world stage for the next few decades.
Thanks Sonny. Glad you enjoyed the post. You are right about the choice US (and some its allies) are making to limit China's access to leading edge chips and other technology. After all, while the US economy has weakened over the last few decades, it still maintains its stronghold in innovation, financial markets and military.
As to the US dominating the world stage on AI, currently and ironically so, most of the AI innovation from OpenAI to Anthropic are taking place under the umbrella of big tech giants/incumbents who are all US based. So, as long as the economics around the current state of network effects and ownership does not change, the US will continue to have an edge.
Also, I subscribe to to Eugene O, who had brilliantly on this topic a couple of weeks/months ago.
https://eugeneo.substack.com/p/generative-ai-who-is-all-this-for?
Interesting analysis, thanks!
Thank you!!!
Great article Amrita. The big question is, how will China develop. A quick comparison. When I look at the automotive space established car makers develop a car somewhere between 5-7 years (a face lift in 2-3 years). China develops a new car in 2 years (with a lot of "devil" in the detail stuff still not fixed, which makes a car "good/better" in the long run).
Thanks Michael, glad you enjoyed the post. Definitely one of my lengthiest ever. You have a valid point with China. Though we have seen China's growth story decline over the last few years with its property deleveraging and the "common prosperity" agenda of Xi, along with an ageing population, I believe it can still engineer a comeback (albeit, that requires fixing a lot of internal issues). I believe that by limiting China access to high edge tech products, the US is trying to gain a competitive edge, though I will admit, I know very little about the automotive space and how AI may impact manufacturing times in both these nations and ultimately what the competitive landscape will look like. Would love to hear your thoughts on it.
Sure. We can have a chat about it, since it will take me a while to explain the differences :-D Just shoot me an email and we will schedule a call.
Yeah, I know, we are long due for a call. Let's chat next week on Friday on 5th April, if that works for you? Anytime between 9-11 am PT is good for me.
I've just send you an invitation. Cu soon!
Michael, I haven't seen an invite from you yet. You can send it to me on Direct Messages.
Absolutely, we are with you!
Thank you Jean. Glad you enjoyed the post.
great read!
Thank you Wrigh, glad you enjoyed it.
This was an enlightening read. I genuinely appreciate the efforts of the writers who things together to make an amazing article such as this. I also came across this article but more on the bearish side.
https://hengecat.substack.com/p/monthly-market-rollup-march-to-april
What do you think?
Thanks Rommel for your kind words. Glad you enjoyed the post. Thanks for sharing the post, I fully agree with all the points that the article has talked about. In fact, in a lot of core macro posts, like the ones I have linked below, I highlight some of the same issues. Having said that, I take a more pragmatic approach to the possible long-term productivity boom that can be unleashed by AI and how it will affect productivity and shape the workforce (which I will talk about in my next Monday post).
https://amritaroy.substack.com/p/heres-my-2024-s-and-p-500-target
Do you think this will help Biden's presidential reelection prospects? I would seem so.
In a perfect world, a boom in productivity should increase the overall welfare in the society, create jobs, improve sentiment and therefore help with reelection prospects.
However, what I am noticing (and this is several years in the making), is that the boom in productivity is disproportionately benefiting parts of society, sectors, states, cities, etc. And while this is common side effect of innovation for certain industries to benefit disproportionately, the gap is becoming wider and we can see that today in the growing divergence between the haves and the have-nots, and this is pulling down overall sentiment, which can be used against him by his opponent.
If AI starts writing NYT bestsellers and hit songs, we are lost. Elevator music and ad copy, fine, but if we can be fooled so thoroughly that we start weeping over AI generated novels and love songs, we might as well turn in our human cards.
I think it ultimately plays down to competitive vs. comparative advantage. While it is possible that AI can do all of these things, and may even get as good or even than human beings, there is always going to be a finite amount of "compute" power, which is AI's constraint. And as a result, over the coming years, I believe we have to make decisions as to where to leverage AI to maximize output given the finite amount of compute, vs. human beings doing the rest of the work, where we have a comparative advantage.
I often wonder that same thing. How good will it get? We may learn more about ourselves than we want to know as we push AI toward whatever its limits are. If I attended a performance of an unfamiliar symphony and cried my way though it the way I did last time I heard Tchaikovsky’s 6th live, then discovered it had been written by AI, what would that tell me about myself and my emotions? In this scenario it would be performed by humans, so there would be that blade of grass to cling to. Otherwise I might fall off the face of the earth.
Another issue is what is economic output and hence productivity
USA economy in terms of dollar measures is mostly virtual
Heating cooling shelter quality food and quality education are crazy expensive
Drivel like facebook Twitter lesser so Google Tesla
Finance industry credit cards
If quality of life is going down for most people as their bodies are filled with microplastics
Uggh
Property taxes property insurance
Government at all level almost 50 percent of economy
Exponential rise in housing prices
Exponential rise in debts
Exponential rise in cost of education
Exponential rise in stocks
This usually ends badly historically speaking
lol
Maryland Governor Says No Evidence of Terrorist Attack on Bridge
Francis Scott Key Bridge collapsed after being struck by ship
So much concern about with the Baltimore bridge accident is if Maersk stock will be affected. Almost no concern about the total incompetence in our “Physical” world
It’s not all virtual. We still live in a physical reality.
Not some Meta crap or Twitter universe
That’s why I have doubts this kind of negligence incompetence whatever one calls it