Software is eating the world. … All of the technology required to transform industries through software finally works and can be widely delivered at global scale. … Software programming tools and Internet-based services make it easy to launch new global software-powered start-ups in many industries—without the need to invest in new infrastructure and train new employees. … With lower start-up costs and a vastly expanded market for online services, the result is a global economy that for the first time will be fully digitally wired. (
Surprising to see this from the author of the Age of Em! I think a more straightforward extrapolation of current trends is a world of declining population but increasing software.
In response to worries about economic decline, current governments seem much more likely to double down on software (by making huge investments in software R&D, datacenters, robots, AI) than in subsidizing parents to have more children. We might disagree with their values, but with typical government goals of economic growth and national security I'm not sure they're mistaken!
Even without government intervention, software will increasingly require less human labor to develop and maintain. A single innovation (language models as coding assistants) has already led to a ~2x human labor cost reduction in software development, and a further 10x reduction this decade seems very plausible. There's no plausible fertility decline fast enough to counteract that rate of productivity growth.
Extrapolating today's trends in the far future makes for fun scenarios, always and ad absurdum. Doing it for 3 trends at once - population/hardware/software - seems useless. Population is the most stable trend, yep it will peak at 11 billion and go slowly down. But during this century this is counteracted by: a) the raising amount of people taking more part in the global economy (+science) - even Pakistanis and Nigerians shall have a higher GDP per head in 2100 than now b) the higher birth-rate of the elite (US-households with an income of over 1 million a year are well over a TFR of 2.1. It is not just Elon Musk.) The global market in 2100 will NOT be smaller than today. And innovation? We just had GPT4; in my life, I have seen no innovation slowing (outside Japan, maybe), and we all are expecting a major UP with GPT 5. We may not even need many people to keep hardware and software up and running and ever improving.
As I said here before, I doubt a long deep fall of population over centuries to pre 1900 levels. Those who feel less like having kids will strongly be selected against, by definition. When Japanese have living space as large as in the US (instead of "rabbit houses"), growth may pick up. When energy will be abundant (PV or fusion), and homes robot-built: what to enjoy more than cuddling your next (not last) baby? - In 1800 there were 30 million Japanese and no one considered the island deserted. In 1900 there lived less than 40 million in France. When Einstein had his golden year, there were less than 60 million Germans (let alone Swiss), many just kids, most still poor, a tiny percentage with university-degrees. A time of stagnation, really?
You have had a very different experience with the growth of software than I have. I've mostly experienced it as increasing interoperability failures, increasing waste as systems that are unnecessary have caused the complete destruction of expensive equipment because they didn't handle exceptions properly, and increasing failure on the part of software engineers to appreciate the nature of complexity and how to handle it. It's a standard joke that the first 10 minutes of every meeting is trying to get the sound to work in Zoom or Teams. I've seen $50,000 cars scrapped because a door lock luxury feature malfunctioned, the software bus got flooded with useless messages, and the engine couldn't start, but it was too hard to diagnose the problem and replace the door handle. And the failure to understand software engineering concepts like the Normalized Distance to the Main Sequence, a metric for controlling complexity in systems, have caused vital underlying systems in the Internet to become unstable because, for instance, browser makers have insisted on a rapid, short release cycle for their product, making it impossible for other systems to upgrade their systems fast enough for their products to be tested against their browsers. You couldn't pay me enough, for instance, to approve the use of Firefox in a corporate environment because the browser's LTS period is shorter than the development and purchase cycle of corporate software packages with an HTTP front-end. So we centralize all of our data and secrets into cloud-based systems, and a single data breach at a major provider like Salesforce, or a fire at an AWS facility, risks bringing down the massive interconnected infrastructure of the modern world. The fact is we aren't very good at softwares development, the rush to low cost that has caused us to abandon more physical solutions has made everything more brittle and likely to fail, constantly and in millions of small ways. There was a recent revelation that commercial aircraft are struggling to navigate over the Middle East because the GPS system is vulnerable to interference because the signals aren't authenticated, even with public key authentication that would maintain the openness of the system, and the backup systems for the GPS are calibrated by ... GPS. We are failing to understand the need to design these systems wisely, and it's going to go critical soon.
In the world without hardware progress and with stagnant economy, most causes of software rot will be absent. Outside of some legal-dependent domains like accounting, nothing will prevent using software developed centuries ago.
This is an interesting exploration of one of the impacts of a declining human population. However, I find it difficult to believe your conclusions because AI is lowering the cost of software development, and is doing so much faster than population growth deceleration. Your points are only relevant if we depart substantially from at least one of these trends.
"The world’s innovation rate will shrink a bit faster than does this exponentially shrinking economy, "
Why is that? Currently, the majority of the world population is on low paying jobs that did not require a high quality education. Dont you think we can grow the effective population by providing better education for more people?
> When software suppliers themselves choose what other software suppliers to rely on, they should in principle worry about how long each supplier’s software will be supported, and whether they will raise their prices. But in an expanding software world these are only mild concerns. If some suppliers die or raise their prices, other superior options will probably become available soon.
Great, except that the new supplier's stuff will have interfaces different from the old supplier, so you can't switch without paying a big cost to hook your software up to the new supplier's software and write custom exports for all your legacy data from the old formats to the new (which will never exactly correspond), and bug test everything. You get vendor lock-in.
> But if your product is near state of the art in features, then expanded revenue from your expanded base of customers will usually let you rewrite it all from scratch later.
Who is rewriting their giant software from scratch today? The MBAs practically never approve such a huge expense just to fix technical debt. Technical debt tends to stick around. A rewrite is not even necessarily a good idea from a technical perspective because a lot of "ugly cruft" in your current software is actually just bugfixes that took a lot of time and customer feedback to produce.
Why? It seems counterintuitive maybe, but AI innovation ( re: software ) would not be constrained by anything but energy production, and even then, could find innovation and efficiencies in its own processes. And energy availability may be inversely proportional to declining population.
" When the economy isn’t large enough to support the huge fixed costs of the computer hardware industry, who will want to pay more for new hardware that isn’t better than plentiful durable old hardware?"
This arguably already true when workstation and business class notebook computers from 5 to 8 years ago have more than enough CPU power to even do strenuous tasks like edit 4K video, and compile code. Why would I pay a 1000 dollars for a consumer grade notebook, when I can get an older workstation class laptop for 200 that has a better keyboard and screen, and is way more durable, and will do everything I need to do?
Surprising to see this from the author of the Age of Em! I think a more straightforward extrapolation of current trends is a world of declining population but increasing software.
In response to worries about economic decline, current governments seem much more likely to double down on software (by making huge investments in software R&D, datacenters, robots, AI) than in subsidizing parents to have more children. We might disagree with their values, but with typical government goals of economic growth and national security I'm not sure they're mistaken!
Even without government intervention, software will increasingly require less human labor to develop and maintain. A single innovation (language models as coding assistants) has already led to a ~2x human labor cost reduction in software development, and a further 10x reduction this decade seems very plausible. There's no plausible fertility decline fast enough to counteract that rate of productivity growth.
Extrapolating today's trends in the far future makes for fun scenarios, always and ad absurdum. Doing it for 3 trends at once - population/hardware/software - seems useless. Population is the most stable trend, yep it will peak at 11 billion and go slowly down. But during this century this is counteracted by: a) the raising amount of people taking more part in the global economy (+science) - even Pakistanis and Nigerians shall have a higher GDP per head in 2100 than now b) the higher birth-rate of the elite (US-households with an income of over 1 million a year are well over a TFR of 2.1. It is not just Elon Musk.) The global market in 2100 will NOT be smaller than today. And innovation? We just had GPT4; in my life, I have seen no innovation slowing (outside Japan, maybe), and we all are expecting a major UP with GPT 5. We may not even need many people to keep hardware and software up and running and ever improving.
As I said here before, I doubt a long deep fall of population over centuries to pre 1900 levels. Those who feel less like having kids will strongly be selected against, by definition. When Japanese have living space as large as in the US (instead of "rabbit houses"), growth may pick up. When energy will be abundant (PV or fusion), and homes robot-built: what to enjoy more than cuddling your next (not last) baby? - In 1800 there were 30 million Japanese and no one considered the island deserted. In 1900 there lived less than 40 million in France. When Einstein had his golden year, there were less than 60 million Germans (let alone Swiss), many just kids, most still poor, a tiny percentage with university-degrees. A time of stagnation, really?
Halfway through, this article goes rather absurd... was it written by AI?
You have had a very different experience with the growth of software than I have. I've mostly experienced it as increasing interoperability failures, increasing waste as systems that are unnecessary have caused the complete destruction of expensive equipment because they didn't handle exceptions properly, and increasing failure on the part of software engineers to appreciate the nature of complexity and how to handle it. It's a standard joke that the first 10 minutes of every meeting is trying to get the sound to work in Zoom or Teams. I've seen $50,000 cars scrapped because a door lock luxury feature malfunctioned, the software bus got flooded with useless messages, and the engine couldn't start, but it was too hard to diagnose the problem and replace the door handle. And the failure to understand software engineering concepts like the Normalized Distance to the Main Sequence, a metric for controlling complexity in systems, have caused vital underlying systems in the Internet to become unstable because, for instance, browser makers have insisted on a rapid, short release cycle for their product, making it impossible for other systems to upgrade their systems fast enough for their products to be tested against their browsers. You couldn't pay me enough, for instance, to approve the use of Firefox in a corporate environment because the browser's LTS period is shorter than the development and purchase cycle of corporate software packages with an HTTP front-end. So we centralize all of our data and secrets into cloud-based systems, and a single data breach at a major provider like Salesforce, or a fire at an AWS facility, risks bringing down the massive interconnected infrastructure of the modern world. The fact is we aren't very good at softwares development, the rush to low cost that has caused us to abandon more physical solutions has made everything more brittle and likely to fail, constantly and in millions of small ways. There was a recent revelation that commercial aircraft are struggling to navigate over the Middle East because the GPS system is vulnerable to interference because the signals aren't authenticated, even with public key authentication that would maintain the openness of the system, and the backup systems for the GPS are calibrated by ... GPS. We are failing to understand the need to design these systems wisely, and it's going to go critical soon.
In the world without hardware progress and with stagnant economy, most causes of software rot will be absent. Outside of some legal-dependent domains like accounting, nothing will prevent using software developed centuries ago.
This is an interesting exploration of one of the impacts of a declining human population. However, I find it difficult to believe your conclusions because AI is lowering the cost of software development, and is doing so much faster than population growth deceleration. Your points are only relevant if we depart substantially from at least one of these trends.
What's the production function for innovation?
"The world’s innovation rate will shrink a bit faster than does this exponentially shrinking economy, "
Why is that? Currently, the majority of the world population is on low paying jobs that did not require a high quality education. Dont you think we can grow the effective population by providing better education for more people?
> When software suppliers themselves choose what other software suppliers to rely on, they should in principle worry about how long each supplier’s software will be supported, and whether they will raise their prices. But in an expanding software world these are only mild concerns. If some suppliers die or raise their prices, other superior options will probably become available soon.
Great, except that the new supplier's stuff will have interfaces different from the old supplier, so you can't switch without paying a big cost to hook your software up to the new supplier's software and write custom exports for all your legacy data from the old formats to the new (which will never exactly correspond), and bug test everything. You get vendor lock-in.
> But if your product is near state of the art in features, then expanded revenue from your expanded base of customers will usually let you rewrite it all from scratch later.
Who is rewriting their giant software from scratch today? The MBAs practically never approve such a huge expense just to fix technical debt. Technical debt tends to stick around. A rewrite is not even necessarily a good idea from a technical perspective because a lot of "ugly cruft" in your current software is actually just bugfixes that took a lot of time and customer feedback to produce.
Urbit fixes this
https://urbit.org/
Last week I commented here that:
Software isn't cheese. It doesn't rot. And real innovation is much more than what transpires from this post.
As promised, here is the full explanation: https://mfioretti.substack.com/p/of-software-rot-fertility-cities
It will take a thousand years before fertility increases? Ho ho ho.
Flowers for Algernon
Why? It seems counterintuitive maybe, but AI innovation ( re: software ) would not be constrained by anything but energy production, and even then, could find innovation and efficiencies in its own processes. And energy availability may be inversely proportional to declining population.
" When the economy isn’t large enough to support the huge fixed costs of the computer hardware industry, who will want to pay more for new hardware that isn’t better than plentiful durable old hardware?"
This arguably already true when workstation and business class notebook computers from 5 to 8 years ago have more than enough CPU power to even do strenuous tasks like edit 4K video, and compile code. Why would I pay a 1000 dollars for a consumer grade notebook, when I can get an older workstation class laptop for 200 that has a better keyboard and screen, and is way more durable, and will do everything I need to do?
I dont remember, maybe I even picked it up from Scott Alexander ... thanks for linking the review