so this argument seems to be more about how high GDP level can get and not growth rates. I find it extremely easy to imagine a world with advanced technology where a typical person is richer than someone making 500k or even 5M per year today, while consuming the same amount of human labor as people today (because they are the typical person). This is granting that robot massages, childcare workers, etc can't replace the human versions. But if you live in a giant house and robots do all your chores and all your home meals are five-star quality with 0 effort (though missing the human touch of a waiter) and you get around in your hypersonic self-flying car, you can be very rich even if your human service consumption is just as constrained as today.
Yes, I'm saying there's an effective ceiling on average living standards at the point where people's material needs are fully met, and I think we're closer to that point than people appreciate. I'm honestly not sure there's that big of a difference in the material living standard of someone making $500k vs. $5M today. The $5M guy might have a bigger house but often he just has a house in a fancier neighborhood (Upper West Side, Atherton, etc.) where the main draw is proximity to other rich people. His kids go to more exclusive schools, he has a summer home with a better view, he has better art on his walls, his wife has a fancier handbag, etc.
There are a few areas where the $5M guy has a better material standard of living, for example he might ride in a private jet whereas the $500k guy probably doesn't. But in most areas of life a guy with a $500k income lives about the same as someone with 10 times that income.
Bill Gates put this well: "I can understand about having millions of dollars. There's meaningful freedom that comes with that, but once you get much beyond that I have to tell you, it's the same hamburger."
The main thing Bill Gates has more of than someone with $50 million is influence over society, but that the ultimate in goods that AI cannot produce more of.
You've put your finger on the key disagreement, which is partial automation (most jobs) versus full automation (all jobs). We already live in a world of extensive partial automation, where the impacted goods (e.g. pencils) and services (e.g. web searches) are effectively post-scarce, but other things remain expensive (e.g. childcare, healthcare, land).
On the question of whether full automation is possible, in principle it isn't, because the task of "being human" cannot, by definition, be performed except by a human. But, aside from that, I don't see any theoretical obstacle to full automation.
First let's consider the examples of coaches and therapists. If coaching and therapy someday becomes available at higher quality and lower cost from AI, most people will choose that. That choice is comparable to choosing to watch movies rather than attend theater plays.
I don't see "being human" as critical to childcare either. It's okay for children to form emotional connections to non-humans (such as pets today), and the children will still have friends and family for human connections.
Another tricky case you didn't mention is the donation of various biological material (e.g. blood, organs, sperm, eggs). Biotech R&D aided by AI and robotics may someday yield a substitute, but AI and robotics can't directly produce these. Still, I see this as something that delays full automation rather than precludes it.
So, if full automation is possible for everything except the "be human" task, as it appears to be, I'm not sure where that leaves us. Certainly the economy in that state of near-full automation would be very different from the economy today.
Lots of things are available at higher quality and lower cost from machines today where people choose human-provided services anyway. Some examples: making coffee, teaching calculus, making pottery, anything that goes by the label "handmade", etc.
I disagree with "lots." I'm not sure what you mean about making coffee, since it almost always requires a human operating a machine. Teaching calculus is clearly a case where most students still learn better from a teacher than by watching a bunch of youtube videos and googling (or asking ChatGPT) for the parts they're confused about. Stuff that's more expensive, lower quality, and marketed as "handmade" or "authentic" is an exception, but that stuff makes up a very small part of the economy, and demand for it is limited (even ultra-rich people mostly use machine-made goods). If we get to the point where the only jobs available to humans are the ones where the customers demand human labor, the economy will look very different.
Ceilings. Ceilings everywhere. There are only so many people and they can only consume so many shirts and massages... which isn't a problem unless you're making Ponzi-like statements. 30% growth with multiple doublings sprinkled in some years? Where's the math?
It's great to see critical analysis on limits to growth. Minimalism, anti-tech sentiment, and population decline should also be worries of the diligent investor. As always supply and demand will operate even as the numbers of products, services, and desires in which AI is involved explode.
AI is helping to engineer demand for what AI companies are selling, which then creates business for humans who fill in where these systems still struggle. In that sense, AI is making a lot of promises for which only humans can provide the "genuine" article. I look forward to seeing more cost-benefit that takes into account the cost of such hybrid approaches.
Lovely article. One point is alluded to, but not emphasized enough (IMO) is that actual wealth (not measured in economic dollars) goes up. Sure, I can only wear so many shirts. But at some point I'll focus on shirt quality or fashion more than I might have when I could only get two a year.
A real world example is computer software. When I started in the industry, every single tool, every single utility was a $50-$2000 purchase (back when those dollar values meant something). Today I have access to things for free that I could not have dreamed buying at any price (at some loss in privacy, but that's another story). Another example is computer hardware; I carry around in my pocket a device that I use to look at dog pictures and texts from my wife that is an orders of magnitude more powerful than the first computer I used for work. My wealth, and things I can do now that I could never do before at any price, is far higher than before.
Increases in GDP are beside the point. The increase in our collective wealth, much of which cannot be valued in dollars, is what matters.
The potential role of automation and intelligence in care work is one of the most intriguing questions out there. A lot people dismiss child care robots; but in truth a lot of child care, and care work in general, is drudgerous, repetitive and sometimes dangerous (home health aides have higher injury rates than police officers). There is a need for emotional connection, and that may require human labor to accomplish. But the actual time and labor involved might be small - a 10 minute check-in every 2 or 3 hours. Think of the time you spend with your child at the playground: mostly you're looking at your phone or chatting with other parents. I think there's probably a lot of scope for automation in care work, if only to focus care workers' time on the social and emotional aspects of care and leave the boring stuff to robots and computers.
so this argument seems to be more about how high GDP level can get and not growth rates. I find it extremely easy to imagine a world with advanced technology where a typical person is richer than someone making 500k or even 5M per year today, while consuming the same amount of human labor as people today (because they are the typical person). This is granting that robot massages, childcare workers, etc can't replace the human versions. But if you live in a giant house and robots do all your chores and all your home meals are five-star quality with 0 effort (though missing the human touch of a waiter) and you get around in your hypersonic self-flying car, you can be very rich even if your human service consumption is just as constrained as today.
Yes, I'm saying there's an effective ceiling on average living standards at the point where people's material needs are fully met, and I think we're closer to that point than people appreciate. I'm honestly not sure there's that big of a difference in the material living standard of someone making $500k vs. $5M today. The $5M guy might have a bigger house but often he just has a house in a fancier neighborhood (Upper West Side, Atherton, etc.) where the main draw is proximity to other rich people. His kids go to more exclusive schools, he has a summer home with a better view, he has better art on his walls, his wife has a fancier handbag, etc.
There are a few areas where the $5M guy has a better material standard of living, for example he might ride in a private jet whereas the $500k guy probably doesn't. But in most areas of life a guy with a $500k income lives about the same as someone with 10 times that income.
Bill Gates put this well: "I can understand about having millions of dollars. There's meaningful freedom that comes with that, but once you get much beyond that I have to tell you, it's the same hamburger."
The main thing Bill Gates has more of than someone with $50 million is influence over society, but that the ultimate in goods that AI cannot produce more of.
You've put your finger on the key disagreement, which is partial automation (most jobs) versus full automation (all jobs). We already live in a world of extensive partial automation, where the impacted goods (e.g. pencils) and services (e.g. web searches) are effectively post-scarce, but other things remain expensive (e.g. childcare, healthcare, land).
On the question of whether full automation is possible, in principle it isn't, because the task of "being human" cannot, by definition, be performed except by a human. But, aside from that, I don't see any theoretical obstacle to full automation.
First let's consider the examples of coaches and therapists. If coaching and therapy someday becomes available at higher quality and lower cost from AI, most people will choose that. That choice is comparable to choosing to watch movies rather than attend theater plays.
I don't see "being human" as critical to childcare either. It's okay for children to form emotional connections to non-humans (such as pets today), and the children will still have friends and family for human connections.
Another tricky case you didn't mention is the donation of various biological material (e.g. blood, organs, sperm, eggs). Biotech R&D aided by AI and robotics may someday yield a substitute, but AI and robotics can't directly produce these. Still, I see this as something that delays full automation rather than precludes it.
So, if full automation is possible for everything except the "be human" task, as it appears to be, I'm not sure where that leaves us. Certainly the economy in that state of near-full automation would be very different from the economy today.
Lots of things are available at higher quality and lower cost from machines today where people choose human-provided services anyway. Some examples: making coffee, teaching calculus, making pottery, anything that goes by the label "handmade", etc.
I disagree with "lots." I'm not sure what you mean about making coffee, since it almost always requires a human operating a machine. Teaching calculus is clearly a case where most students still learn better from a teacher than by watching a bunch of youtube videos and googling (or asking ChatGPT) for the parts they're confused about. Stuff that's more expensive, lower quality, and marketed as "handmade" or "authentic" is an exception, but that stuff makes up a very small part of the economy, and demand for it is limited (even ultra-rich people mostly use machine-made goods). If we get to the point where the only jobs available to humans are the ones where the customers demand human labor, the economy will look very different.
Ceilings. Ceilings everywhere. There are only so many people and they can only consume so many shirts and massages... which isn't a problem unless you're making Ponzi-like statements. 30% growth with multiple doublings sprinkled in some years? Where's the math?
It's great to see critical analysis on limits to growth. Minimalism, anti-tech sentiment, and population decline should also be worries of the diligent investor. As always supply and demand will operate even as the numbers of products, services, and desires in which AI is involved explode.
AI is helping to engineer demand for what AI companies are selling, which then creates business for humans who fill in where these systems still struggle. In that sense, AI is making a lot of promises for which only humans can provide the "genuine" article. I look forward to seeing more cost-benefit that takes into account the cost of such hybrid approaches.
Lovely article. One point is alluded to, but not emphasized enough (IMO) is that actual wealth (not measured in economic dollars) goes up. Sure, I can only wear so many shirts. But at some point I'll focus on shirt quality or fashion more than I might have when I could only get two a year.
A real world example is computer software. When I started in the industry, every single tool, every single utility was a $50-$2000 purchase (back when those dollar values meant something). Today I have access to things for free that I could not have dreamed buying at any price (at some loss in privacy, but that's another story). Another example is computer hardware; I carry around in my pocket a device that I use to look at dog pictures and texts from my wife that is an orders of magnitude more powerful than the first computer I used for work. My wealth, and things I can do now that I could never do before at any price, is far higher than before.
Increases in GDP are beside the point. The increase in our collective wealth, much of which cannot be valued in dollars, is what matters.
The potential role of automation and intelligence in care work is one of the most intriguing questions out there. A lot people dismiss child care robots; but in truth a lot of child care, and care work in general, is drudgerous, repetitive and sometimes dangerous (home health aides have higher injury rates than police officers). There is a need for emotional connection, and that may require human labor to accomplish. But the actual time and labor involved might be small - a 10 minute check-in every 2 or 3 hours. Think of the time you spend with your child at the playground: mostly you're looking at your phone or chatting with other parents. I think there's probably a lot of scope for automation in care work, if only to focus care workers' time on the social and emotional aspects of care and leave the boring stuff to robots and computers.