>So the process for creating new inventions seems to be getting more and more efficient — opportunities are getting noticed and exploited sooner and sooner, up through 1970 at least (which is when the list of major inventions extends to).
Great piece, but this seems like a weak conclusion. We may in fact be missing far more possible inventions now than we were in the past. Human knowledge had increased exponentially, suggesting that much more is possible, and an invention that was possible in 1960 but would not be invented for 100 years is obviously not in the dataset; the present cuts off the long tail.
Exactly. There is a right-hand time censoring problem built into the data, even the concept. Perhaps FTL travel is sitting around somewhere right now, waiting for someone to just put together the pieces. "The future is already here -- it's just not evenly distributed" -- William Gibson.
Yep, the selection of inventions is also biased toward the industrial revolution; the Romans likely had many inventions for improving road-laying that are underrepresented here. The reasons for the time delays are also highly varied; for example, the blast furnace was enabled by coal-based iron smelting, which was in turn invented because of a ban on charcoal production in England to preserve trees for making warship masts. I'm not sure that any conclusion can be drawn from this other than that Claude is getting quite good at writing case studies.
> an invention that was possible in 1960 but would not be invented for 100 years is obviously not in the dataset; the present cuts off the long tail.
Correct, but for all technologies made in 1960 we can ask when they could have been made, and that’s what this piece does. It doesn’t ask what possible technologies could have been invented by any given year. That is, there’s a natural selection bias based on what inventions were actually made.
One critique could be in the interpretation: is it that we’re becoming more efficient, or perhaps instead that we’re now more incremental and myopic, so that only recent developments are guiding current inventions?
You can't conclude that ”opportunities are getting exploited sooner and sooner" when the dataset by definition does not include opportunities which have not yet been exploited. Look up time censorship; it's often a problem in medical studies but applies here.
I would expect the safety pin to be fairly labor intensive before factory technology. their primary use (temporary clothing adjustment) would have been less useful when a significant portion of the population had seamstress skills. For a secondary use, such as holding cloth in place to make adjustments, they are less efficient to use than bobby pins, if someone knows what they are doing
This seems a specific example of the trend of things which could have been invented much earlier, not being invented because they weren't useful or economically feasible
I've seen all sorts of brooches in museums made of different metals from various eras that were used to hold one piece of clothing in place against another. My guess is that clothes pins like barbed wire and hypodermic needles were a product of improved metallurgy and wire pulling technology.
It has the sharp end captured behind a shield. You can find variations on the idea going back a thousand years before that. The cheaper ones that would've been available to larger parts of the population definitely aren't as nice, the pin end is not as well-shielded -- the ones that are safer tend to also be more decorative, because once you're doing the work to make it safe like that, it's already an expensive luxury item, so you might as well also make it a decorative brooch. But the basic principle is there.
What's distinctive about the modern safety pin is just that it's mass-produced. The same way straight-pins shifted from artisanal to mass-production, and nails, and everything else.
That makes a lot of sense - why make a metal pin to hold some cloth together when you can just stitch it up with cheap thread and rip that thread out when you need it out? Only once metal is extremely cheap, and sewing skills are starting to be displaced would this shift.
Elevators could also be powered by people or animals. There was a mule powered elevator in lower Manhattan as late as 1939 per the World's Fair issue of Fortune Magazine. The animals lived in a stable on the top floor and drove the elevator by walking on a treadmill. According to ANSI, Archimedes was credited with the first human powered elevator design, and human powered elevators were used in the Colosseum to raise animals to the arena. In 1743, Louis XV had a human powered elevator installed at Versailles so he could visit his mistress. It wasn't until the 19th century that they developed steam powered elevators.
The transistor didn't require knowledge of quantum mechanics. By 1874, Braun recognized that a metal-crystal point contact device could be used as a rectifier and, by 1894, Bose was using it to detect radio signals. By the 1920s, people were building crystal set radios based on the principle. In 1948, Matare and Welker, independently, recognized that one could build a point contact germanium crystal amplifier. The devices, marketed as transistrons, were used in the French telephone system and in a small portable radio. The transistor was invented in the same year, and physically the transistor and transistron look very similar though the former was easier to manufacture.
When I was at the Bath Museum of Fashion, one exhibit explained how the Bessemer process changed the way women's clothing was designed. Whalebone, for example, could only take so much stress while steel could handle much more. Thanks to steel, even you, guy or gal, could have an hourglass figure. The process also revolutionized wire production since the steel it produced at modest cost could be drawn more effectively and could produce stronger wire. That's why barbed wire was practical in the west after the Civil War. That's also why hypodermic needles could be made using a wire drawing process.
Crystal rectifiers were used prior to the transistor, but my understanding is that actual transistor behavior required quantum mechanics, specifically the idea of "electron holes." From Riordan and Hoddeson's "Crystal Fire", a history of the transistor:
"The swift invention of a semiconductor amplifier had proved the wisdom of Kelly's emphasis on basic research in solid-state physics. The key to the discovery was recognition that quantum-mechanical entities - the holes - had a crucial role to play in carrying electric current near a semiconductor surface. A "classical" understanding of how this material behaved would not have sufficed."
That's definitely true for the 1950 junction transistor. It was an impressive piece of work. There had been a lot of empirical work on NP junctions, but Shockley put together the ideas to develop a solid state switch.
The steam engine example is a very interesting one and its not speculation whether the ancients developed this technology.
They did. It is thoroughly documented that the aeolipile was invented in ancient Alexandria and crazy enough, it was an entertainment device and not used practically. So the gap there was like 1700 years!
And I’ve read a similar gap in the application of gunpowder to practical use as explosive and propellant, rather than entertainment in fireworks.
I’d wager the connectivity of people and information is the key factor here. And after the printing press, knowledge accumulated much faster. And now with the internet, lightening speed.
There was also the need for accurate machining so that the piston would form a nearly gas tight seal. Cannons were built in the 14th century, but it wasn't until the 18th century that a boring machine could produce a good piston seal for a practical engine.
There was also the need for accurate machining so that the piston would form a nearly gas tight seal. Cannons were built in the 14th century, but it wasn't until the 18th century that a boring machine could produce a good piston seal for a practical engine.
The first actually useful steam engine operates in much more complex way than the aeopile - one that is far less obvious and is I think genuinely fairly described as 'ingenious' because it found a way to work around the limitations of engineering tolerances that made other designs impractical. Ironically it was proving that even a bad engine was useful that spurred refinement of the engineering required to produce a simpler, more effective engine.
I’m not so sure the ancients were incapable of the precise metal working to create a working steam engine. But I take your point about machining being critical to the practical production of it.
They could do some amazing things but boring holes to fine tolerances is an industrial engineering problem rather than a precise artisanal gear making exercise. The setup that finally achieve it was enormous but w see https://www.wrexhamheritage.wales/bersham-ironworks/
I disagree. Check out some of the amazing machines designed by the ancients. They solved some really complex engineering problems, evident in their construction
Btw, you could just attach string to the aileopile with slight modification to make it do practical work. Or it could be used to do the circular motion in millwork, or at least help lighten the burden.
Ok you may want to believe that, but people have built models of these things. They don't generate any force whatsoever. You could stop it with your finger. They would barely be strong enough to lift the string you tied to it, let alone exert meaningful force on something at the other end of the string🤪
Oral rehydration therapy would probably be the best candidate for how late it was discovered (the 60s) compared to the earliest it could have been. It would have been extremely useful for basically all of human history, and only requires the right ratios of water, salt, and sugar (which presumably honey could fulfill) to work.
Other than that:
* The wheelbarrow? It's medieval (in Europe), but would presumably have been useful and not too difficult to build in the bronze age.
* The efficient horse collar; the ancient ones were terrible and choked the animal constantly.
* The compass? You basically just need to suspend a shard of lodestone in water. Could have been done at any time after people knew that lodestones existed, so antiquity.
* Buttonholes were invented in the 13th century, while buttons themselves are ancient.
There's also several plants that could have been used to great benefit if one knew about them. E.g. pyrethrum should not be too difficult to use to make an effective insecticide (grind it into powder and dust it in granaries!), and Artemisinin can be cold-extracted from sweet wormwood to treat malaria, which was discovered in the 70s. That would have been doable and extremely useful for basically any society throughout history.
I'd like to challenge the invention of reinforced concrete on at least two grounds. First is the loss of the recipe for concrete, being particular to the type of volcanic ash in ancient formulation. Second is the development of a source of steel. Cast iron would possibly work to some level for reinforced concrete, but it's pretty brittle, and casting long rods of iron has some technical difficulties. Also, the thermal expansion coefficient is 7 vs 6 ppm/F and isn't quite as nice for surviving the diurnal thermal cycle.
All in all, a better basis in chemistry and metallurgy should push the feasible time up into the late Renaissance. Also, actual steel required some serious technology and scale, so Bessmer.
Understood, and I have no argument over that. But that doesn't consider the loss of knowledge due to the limited understanding of both chemistry and metallurgy. So called Damascus Steel was also lost, and while we now suspect that the ore used contained vanadium, which changed the iron/carbon crystal formation under repeated heating and forging, we still don't know if there were other techniques used.
My point is that inventions can be lost. Accidental discovery makes a great what-if, but the basis of a true invention is widespread repeatability, one of the reasons patents require a full disclosure of the preferred method/embodiment.
This is the right sort of work for LLMs, thank you for sharing it.
Next question - what innovations are we currently not making, for which all of the pieces are in place? Id est, a problem in one field, which has been solved in another, but the application has yet to be noticed and the transfer made. Or a technique abandoned for reasons which no longer need apply.
Figure out how to ask the right question and this kind directed data mining could get very interesting. Remember me in your Nobel speech....
You may not agree, but I would argue that the combination of factories in the countryside that run on part-time jobs with an incentive-based wage system (see link below) might qualify, even though it is not obvious why it could not have occurred earlier, at least in isolated cases.
The technologies required are (a) modern transport and communication technologies, (b) rural electrification, and (c) computers (possibly even AI?) to manage the complex accounting challenges in many situations. Finally, there would need to be sufficient overall economic development (in terms of prevailing wage rates) to make a much shorter workweek thinkable.
Of course it could be argued that this is not even an invention, but rather the exploitation of a lot of prior inventions in a new way. Certainly not a patentable idea, even though difficult to pull off in an organized way.
Are you talking about the outwork system? Used for producing yarn for textile weaving for centuries before the invention of water powered spinning mills. Fibre buyers would place bales of raw material out at cotttages, and collect the finished yarn months later.
The putting out system, Weber called. No, this is something else entirely. Modern factories, the most labor-intensive especially. The idea is to increase the marginal utility of wages, so workers will work harder to earn every penny they can in order to support their new lifestyle. See chapter two, note ii.
This is a very useful way to think about invention. One implication is that many inventions do not wait only for ideas. They wait for a system dense enough to combine materials, engineering, production routines, users, and capital into a working solution.
That is why I increasingly think industry itself is the largest university. Factories, suppliers, engineers, repair networks, and customers create a continuous learning environment. They teach what is technically possible, what can scale, and where the next bottleneck really is.
So the key question is not only which society produces frontier science, but which society can shorten the distance between “technically possible” and “industrially real.” That gap may be one of the most important measures of national technological capacity.
If I remember a book on sailing correctly, sailboats were steered with oars for 5000 years before the hinged rudder was invented.
Controlled flight, whether powered or not, had to wait for the simple concept of roll actuators (wing warping or ailerons). Pitch and yaw controls were common before the Wright Brothers. How so many of their predecessors could have overlooked this seemingly obvious concept is beyond me.
Did you by chance include the Linotype machine in your list? I believe many were racing to build an automatic type setting machine in the 19th century. It may be the most complex, purely mechanical machine ever mass produced.
One obvious reason why new technologies are not improved and brought to market faster is that, at prevailing factor market prices, they're not yet profitable compared to the available alternatives. Economic historian Robert C. Allen argued that the industrial revolution happened in Britain first primarily Britain had higher wages and cheaper coal compared to China or the European continent. The Newcomen steam engine, invented in 1712 had few adopters (mainly for pumping water out of coal mines) and was adopted mostly in Britain. It would more than a half century before James Watt would improve the design enough to make it useful in other industries. The barriers to spending time and effort to bring a new technology to market are partly technical, but partly determined by the cost of alternatives.
We must be achingly close to the AI bubble bursting, because now I'm seeing these "conversations with an LLM masquerading as insight" bloody everywhere.
In response, I'll do something equally low effort and quote Cory Doctorow: "No one wants to read your AI slop."
I agree there is too much AI slop out there, but this is some actually very good usage for a change. I guess feel free to improve individual data points with human expertise. If you correct many things, we can call this slop, but for now I think this is a good example of exactly how AI should be used.
I recall seeing from a professor of economics that for teaching his masters' students he now has a hybrid approach. The first course is pen-and-paper only, for learning the concepts and principles. In the second course the students use AI to do research projects that only a few years ago would have been Ph. D.-level papers. AI provides great leverage, but humans need to learn how to use it. That is what universities are mostly failing to teach now.
On the question of slop, the Replication Crisis in social sciences and psychology and the concept of the "least publishable unit" show that slop is not confined to AI. It has been around for many decades.
The debate in these comments between Chris and JP regarding 'time censorship', paired with @The Present Author's question about overlooked innovations sitting across different fields, highlights the exact mathematical bottleneck of technological progress.
The reason we have long wait-times for inventions, or why we struggle with data censorship, is because we observe technology through a linear timeline (t), rather than mapping it as a multi-dimensional density function. On my blog (The 6174 Equation), I framework this using a physics-based calculus model for Impact Density:
V = ∫ (I · q) dt
Where (V) is the realized value, (I) is the cross-pollination coefficient (the structural bridge between two separate fields), and (q) is the quality/readiness of the infrastructure.
To answer @The Present Author: The biggest innovations we are currently missing are NOT in frontier science, but in the transfer of mathematical modeling from physical systems into behavioral and psychological systems. For instance, 'Decision Fatigue' can be modeled exponentially as:
F = D + e🇺
(where U is unresolved background simulations).
Chris is right about the right-hand censoring, and JP is right about our myopia. But the root cause is that our current AI and data mining systems are searching linearly. Until we systematically treat cross-disciplinary intersections (I) as primary variables rather than accidental anomalies, the most revolutionary inventions of our time will remain hidden in plain sight, fully armed with prerequisites, yet waiting for a dominant framework to unlock them.
The debate in these comments between Chris and JP regarding 'time censorship', paired with @The Present Author's question about overlooked innovations sitting across different fields, highlights the exact mathematical bottleneck of technological progress.
The reason we have long wait-times for inventions, or why we struggle with data censorship, is because we observe technology through a linear timeline (t), rather than mapping it as a multi-dimensional density function. On my blog (The 6174 Equation), I framework this using a physics-based calculus model for Impact Density:
V = ∫ (I · q) dt
Where (V) is the realized value, (I) is the cross-pollination coefficient (the structural bridge between two separate fields), and (q) is the quality/readiness of the infrastructure.
To answer @The Present Author: The biggest innovations we are currently missing are NOT in frontier science, but in the transfer of mathematical modeling from physical systems into behavioral and psychological systems. For instance, 'Decision Fatigue' can be modeled exponentially as:
F = D + e🇺
(where U is unresolved background simulations).
Chris is right about the right-hand censoring, and JP is right about our myopia. But the root cause is that our current AI and data mining systems are searching linearly. Until we systematically treat cross-disciplinary intersections (I) as primary variables rather than accidental anomalies, the most revolutionary inventions of our time will remain hidden in plain sight, fully armed with prerequisites, yet waiting for a dominant framework to unlock them.
I want to see devices protecting people from needles and syringes while removing them from pavement lawn bus benches and safely containerizing them.
I also want to see high-grade videography or 35mm chronicling the cities where Dems hold political power but refuse to believe that non functional areas of their control actually are dumpsters. It could be done with an apolitical commentary or angle. People need to see the destruction that poverty, 911 is a joke, takes on liberal progressive cities.
>So the process for creating new inventions seems to be getting more and more efficient — opportunities are getting noticed and exploited sooner and sooner, up through 1970 at least (which is when the list of major inventions extends to).
Great piece, but this seems like a weak conclusion. We may in fact be missing far more possible inventions now than we were in the past. Human knowledge had increased exponentially, suggesting that much more is possible, and an invention that was possible in 1960 but would not be invented for 100 years is obviously not in the dataset; the present cuts off the long tail.
Exactly. There is a right-hand time censoring problem built into the data, even the concept. Perhaps FTL travel is sitting around somewhere right now, waiting for someone to just put together the pieces. "The future is already here -- it's just not evenly distributed" -- William Gibson.
Yep, the selection of inventions is also biased toward the industrial revolution; the Romans likely had many inventions for improving road-laying that are underrepresented here. The reasons for the time delays are also highly varied; for example, the blast furnace was enabled by coal-based iron smelting, which was in turn invented because of a ban on charcoal production in England to preserve trees for making warship masts. I'm not sure that any conclusion can be drawn from this other than that Claude is getting quite good at writing case studies.
I’m not sure I follow. Eg
> an invention that was possible in 1960 but would not be invented for 100 years is obviously not in the dataset; the present cuts off the long tail.
Correct, but for all technologies made in 1960 we can ask when they could have been made, and that’s what this piece does. It doesn’t ask what possible technologies could have been invented by any given year. That is, there’s a natural selection bias based on what inventions were actually made.
One critique could be in the interpretation: is it that we’re becoming more efficient, or perhaps instead that we’re now more incremental and myopic, so that only recent developments are guiding current inventions?
You can't conclude that ”opportunities are getting exploited sooner and sooner" when the dataset by definition does not include opportunities which have not yet been exploited. Look up time censorship; it's often a problem in medical studies but applies here.
Ok, we’re saying the same thing it appears.
I would expect the safety pin to be fairly labor intensive before factory technology. their primary use (temporary clothing adjustment) would have been less useful when a significant portion of the population had seamstress skills. For a secondary use, such as holding cloth in place to make adjustments, they are less efficient to use than bobby pins, if someone knows what they are doing
This seems a specific example of the trend of things which could have been invented much earlier, not being invented because they weren't useful or economically feasible
I question the safety pin answer, because the safety pin arguably _was invented_ basically as soon as practical metallurgy existed, in the Bronze Age.
https://en.wikipedia.org/wiki/Fibula_(brooch)
I've seen all sorts of brooches in museums made of different metals from various eras that were used to hold one piece of clothing in place against another. My guess is that clothes pins like barbed wire and hypodermic needles were a product of improved metallurgy and wire pulling technology.
This thing from the early 5th century is basically a safety pin.
https://en.wikipedia.org/wiki/Fibula_(brooch)#/media/File:Paar_Prunkfibeln.jpg
It has the sharp end captured behind a shield. You can find variations on the idea going back a thousand years before that. The cheaper ones that would've been available to larger parts of the population definitely aren't as nice, the pin end is not as well-shielded -- the ones that are safer tend to also be more decorative, because once you're doing the work to make it safe like that, it's already an expensive luxury item, so you might as well also make it a decorative brooch. But the basic principle is there.
What's distinctive about the modern safety pin is just that it's mass-produced. The same way straight-pins shifted from artisanal to mass-production, and nails, and everything else.
That makes a lot of sense - why make a metal pin to hold some cloth together when you can just stitch it up with cheap thread and rip that thread out when you need it out? Only once metal is extremely cheap, and sewing skills are starting to be displaced would this shift.
Elevators could also be powered by people or animals. There was a mule powered elevator in lower Manhattan as late as 1939 per the World's Fair issue of Fortune Magazine. The animals lived in a stable on the top floor and drove the elevator by walking on a treadmill. According to ANSI, Archimedes was credited with the first human powered elevator design, and human powered elevators were used in the Colosseum to raise animals to the arena. In 1743, Louis XV had a human powered elevator installed at Versailles so he could visit his mistress. It wasn't until the 19th century that they developed steam powered elevators.
The transistor didn't require knowledge of quantum mechanics. By 1874, Braun recognized that a metal-crystal point contact device could be used as a rectifier and, by 1894, Bose was using it to detect radio signals. By the 1920s, people were building crystal set radios based on the principle. In 1948, Matare and Welker, independently, recognized that one could build a point contact germanium crystal amplifier. The devices, marketed as transistrons, were used in the French telephone system and in a small portable radio. The transistor was invented in the same year, and physically the transistor and transistron look very similar though the former was easier to manufacture.
When I was at the Bath Museum of Fashion, one exhibit explained how the Bessemer process changed the way women's clothing was designed. Whalebone, for example, could only take so much stress while steel could handle much more. Thanks to steel, even you, guy or gal, could have an hourglass figure. The process also revolutionized wire production since the steel it produced at modest cost could be drawn more effectively and could produce stronger wire. That's why barbed wire was practical in the west after the Civil War. That's also why hypodermic needles could be made using a wire drawing process.
Crystal rectifiers were used prior to the transistor, but my understanding is that actual transistor behavior required quantum mechanics, specifically the idea of "electron holes." From Riordan and Hoddeson's "Crystal Fire", a history of the transistor:
"The swift invention of a semiconductor amplifier had proved the wisdom of Kelly's emphasis on basic research in solid-state physics. The key to the discovery was recognition that quantum-mechanical entities - the holes - had a crucial role to play in carrying electric current near a semiconductor surface. A "classical" understanding of how this material behaved would not have sufficed."
That's definitely true for the 1950 junction transistor. It was an impressive piece of work. There had been a lot of empirical work on NP junctions, but Shockley put together the ideas to develop a solid state switch.
https://sites.google.com/site/transistorhistory/faraday-to-shockley
Fascinating!
The steam engine example is a very interesting one and its not speculation whether the ancients developed this technology.
They did. It is thoroughly documented that the aeolipile was invented in ancient Alexandria and crazy enough, it was an entertainment device and not used practically. So the gap there was like 1700 years!
And I’ve read a similar gap in the application of gunpowder to practical use as explosive and propellant, rather than entertainment in fireworks.
I’d wager the connectivity of people and information is the key factor here. And after the printing press, knowledge accumulated much faster. And now with the internet, lightening speed.
Scale matters for steam engines. The pressures needed for a practical stream engine exceeded capabilities of any ancient metal.
There was also the need for accurate machining so that the piston would form a nearly gas tight seal. Cannons were built in the 14th century, but it wasn't until the 18th century that a boring machine could produce a good piston seal for a practical engine.
I’m reading the ancients used steam power practically to open temple doors and power musical devices like organs and whistles.
I think they were very close to a turbine.
There was also the need for accurate machining so that the piston would form a nearly gas tight seal. Cannons were built in the 14th century, but it wasn't until the 18th century that a boring machine could produce a good piston seal for a practical engine.
The first actually useful steam engine operates in much more complex way than the aeopile - one that is far less obvious and is I think genuinely fairly described as 'ingenious' because it found a way to work around the limitations of engineering tolerances that made other designs impractical. Ironically it was proving that even a bad engine was useful that spurred refinement of the engineering required to produce a simpler, more effective engine.
I’m not so sure the ancients were incapable of the precise metal working to create a working steam engine. But I take your point about machining being critical to the practical production of it.
See:
https://en.wikipedia.org/wiki/Antikythera_mechanism
They could do some amazing things but boring holes to fine tolerances is an industrial engineering problem rather than a precise artisanal gear making exercise. The setup that finally achieve it was enormous but w see https://www.wrexhamheritage.wales/bersham-ironworks/
I disagree. Check out some of the amazing machines designed by the ancients. They solved some really complex engineering problems, evident in their construction
Btw, you could just attach string to the aileopile with slight modification to make it do practical work. Or it could be used to do the circular motion in millwork, or at least help lighten the burden.
I don’t think they were that far off.
Ok you may want to believe that, but people have built models of these things. They don't generate any force whatsoever. You could stop it with your finger. They would barely be strong enough to lift the string you tied to it, let alone exert meaningful force on something at the other end of the string🤪
Oral rehydration therapy would probably be the best candidate for how late it was discovered (the 60s) compared to the earliest it could have been. It would have been extremely useful for basically all of human history, and only requires the right ratios of water, salt, and sugar (which presumably honey could fulfill) to work.
Other than that:
* The wheelbarrow? It's medieval (in Europe), but would presumably have been useful and not too difficult to build in the bronze age.
* The efficient horse collar; the ancient ones were terrible and choked the animal constantly.
* The compass? You basically just need to suspend a shard of lodestone in water. Could have been done at any time after people knew that lodestones existed, so antiquity.
* Buttonholes were invented in the 13th century, while buttons themselves are ancient.
There's also several plants that could have been used to great benefit if one knew about them. E.g. pyrethrum should not be too difficult to use to make an effective insecticide (grind it into powder and dust it in granaries!), and Artemisinin can be cold-extracted from sweet wormwood to treat malaria, which was discovered in the 70s. That would have been doable and extremely useful for basically any society throughout history.
Wait, what were buttons used for in the millennium before buttonholes?
Hooking onto loops of cord, probably.
I'd like to challenge the invention of reinforced concrete on at least two grounds. First is the loss of the recipe for concrete, being particular to the type of volcanic ash in ancient formulation. Second is the development of a source of steel. Cast iron would possibly work to some level for reinforced concrete, but it's pretty brittle, and casting long rods of iron has some technical difficulties. Also, the thermal expansion coefficient is 7 vs 6 ppm/F and isn't quite as nice for surviving the diurnal thermal cycle.
All in all, a better basis in chemistry and metallurgy should push the feasible time up into the late Renaissance. Also, actual steel required some serious technology and scale, so Bessmer.
The Romans used iron rods and bars to reinforce their concrete structures, there's no reason in principle they couldn't have simply embedded the iron into the actual concrete. https://www.unisalento.it/documents/20152/1309975/CMA%2B2015%2BIRON%2BCHICAGO.pdf
Understood, and I have no argument over that. But that doesn't consider the loss of knowledge due to the limited understanding of both chemistry and metallurgy. So called Damascus Steel was also lost, and while we now suspect that the ore used contained vanadium, which changed the iron/carbon crystal formation under repeated heating and forging, we still don't know if there were other techniques used.
My point is that inventions can be lost. Accidental discovery makes a great what-if, but the basis of a true invention is widespread repeatability, one of the reasons patents require a full disclosure of the preferred method/embodiment.
This is the right sort of work for LLMs, thank you for sharing it.
Next question - what innovations are we currently not making, for which all of the pieces are in place? Id est, a problem in one field, which has been solved in another, but the application has yet to be noticed and the transfer made. Or a technique abandoned for reasons which no longer need apply.
Figure out how to ask the right question and this kind directed data mining could get very interesting. Remember me in your Nobel speech....
You may not agree, but I would argue that the combination of factories in the countryside that run on part-time jobs with an incentive-based wage system (see link below) might qualify, even though it is not obvious why it could not have occurred earlier, at least in isolated cases.
The technologies required are (a) modern transport and communication technologies, (b) rural electrification, and (c) computers (possibly even AI?) to manage the complex accounting challenges in many situations. Finally, there would need to be sufficient overall economic development (in terms of prevailing wage rates) to make a much shorter workweek thinkable.
Of course it could be argued that this is not even an invention, but rather the exploitation of a lot of prior inventions in a new way. Certainly not a patentable idea, even though difficult to pull off in an organized way.
https://www.amazon.com/dp/B00U0C9HKW
Are you talking about the outwork system? Used for producing yarn for textile weaving for centuries before the invention of water powered spinning mills. Fibre buyers would place bales of raw material out at cotttages, and collect the finished yarn months later.
The putting out system, Weber called. No, this is something else entirely. Modern factories, the most labor-intensive especially. The idea is to increase the marginal utility of wages, so workers will work harder to earn every penny they can in order to support their new lifestyle. See chapter two, note ii.
A fascinating use of AI. I can see many similar AI applications that help us understand the process of invention and diffusion.
Great stuff!
This is a very useful way to think about invention. One implication is that many inventions do not wait only for ideas. They wait for a system dense enough to combine materials, engineering, production routines, users, and capital into a working solution.
That is why I increasingly think industry itself is the largest university. Factories, suppliers, engineers, repair networks, and customers create a continuous learning environment. They teach what is technically possible, what can scale, and where the next bottleneck really is.
So the key question is not only which society produces frontier science, but which society can shorten the distance between “technically possible” and “industrially real.” That gap may be one of the most important measures of national technological capacity.
Fascinating.
If I remember a book on sailing correctly, sailboats were steered with oars for 5000 years before the hinged rudder was invented.
Controlled flight, whether powered or not, had to wait for the simple concept of roll actuators (wing warping or ailerons). Pitch and yaw controls were common before the Wright Brothers. How so many of their predecessors could have overlooked this seemingly obvious concept is beyond me.
Did you by chance include the Linotype machine in your list? I believe many were racing to build an automatic type setting machine in the 19th century. It may be the most complex, purely mechanical machine ever mass produced.
One obvious reason why new technologies are not improved and brought to market faster is that, at prevailing factor market prices, they're not yet profitable compared to the available alternatives. Economic historian Robert C. Allen argued that the industrial revolution happened in Britain first primarily Britain had higher wages and cheaper coal compared to China or the European continent. The Newcomen steam engine, invented in 1712 had few adopters (mainly for pumping water out of coal mines) and was adopted mostly in Britain. It would more than a half century before James Watt would improve the design enough to make it useful in other industries. The barriers to spending time and effort to bring a new technology to market are partly technical, but partly determined by the cost of alternatives.
Another example, I have been wondering if there is any reason the Romans couldn't have invented the printing press?
We must be achingly close to the AI bubble bursting, because now I'm seeing these "conversations with an LLM masquerading as insight" bloody everywhere.
In response, I'll do something equally low effort and quote Cory Doctorow: "No one wants to read your AI slop."
https://pluralistic.net/2026/03/02/nonconsensual-slopping/
I agree there is too much AI slop out there, but this is some actually very good usage for a change. I guess feel free to improve individual data points with human expertise. If you correct many things, we can call this slop, but for now I think this is a good example of exactly how AI should be used.
I recall seeing from a professor of economics that for teaching his masters' students he now has a hybrid approach. The first course is pen-and-paper only, for learning the concepts and principles. In the second course the students use AI to do research projects that only a few years ago would have been Ph. D.-level papers. AI provides great leverage, but humans need to learn how to use it. That is what universities are mostly failing to teach now.
On the question of slop, the Replication Crisis in social sciences and psychology and the concept of the "least publishable unit" show that slop is not confined to AI. It has been around for many decades.
The debate in these comments between Chris and JP regarding 'time censorship', paired with @The Present Author's question about overlooked innovations sitting across different fields, highlights the exact mathematical bottleneck of technological progress.
The reason we have long wait-times for inventions, or why we struggle with data censorship, is because we observe technology through a linear timeline (t), rather than mapping it as a multi-dimensional density function. On my blog (The 6174 Equation), I framework this using a physics-based calculus model for Impact Density:
V = ∫ (I · q) dt
Where (V) is the realized value, (I) is the cross-pollination coefficient (the structural bridge between two separate fields), and (q) is the quality/readiness of the infrastructure.
To answer @The Present Author: The biggest innovations we are currently missing are NOT in frontier science, but in the transfer of mathematical modeling from physical systems into behavioral and psychological systems. For instance, 'Decision Fatigue' can be modeled exponentially as:
F = D + e🇺
(where U is unresolved background simulations).
Chris is right about the right-hand censoring, and JP is right about our myopia. But the root cause is that our current AI and data mining systems are searching linearly. Until we systematically treat cross-disciplinary intersections (I) as primary variables rather than accidental anomalies, the most revolutionary inventions of our time will remain hidden in plain sight, fully armed with prerequisites, yet waiting for a dominant framework to unlock them.
The debate in these comments between Chris and JP regarding 'time censorship', paired with @The Present Author's question about overlooked innovations sitting across different fields, highlights the exact mathematical bottleneck of technological progress.
The reason we have long wait-times for inventions, or why we struggle with data censorship, is because we observe technology through a linear timeline (t), rather than mapping it as a multi-dimensional density function. On my blog (The 6174 Equation), I framework this using a physics-based calculus model for Impact Density:
V = ∫ (I · q) dt
Where (V) is the realized value, (I) is the cross-pollination coefficient (the structural bridge between two separate fields), and (q) is the quality/readiness of the infrastructure.
To answer @The Present Author: The biggest innovations we are currently missing are NOT in frontier science, but in the transfer of mathematical modeling from physical systems into behavioral and psychological systems. For instance, 'Decision Fatigue' can be modeled exponentially as:
F = D + e🇺
(where U is unresolved background simulations).
Chris is right about the right-hand censoring, and JP is right about our myopia. But the root cause is that our current AI and data mining systems are searching linearly. Until we systematically treat cross-disciplinary intersections (I) as primary variables rather than accidental anomalies, the most revolutionary inventions of our time will remain hidden in plain sight, fully armed with prerequisites, yet waiting for a dominant framework to unlock them.
I want to see devices protecting people from needles and syringes while removing them from pavement lawn bus benches and safely containerizing them.
I also want to see high-grade videography or 35mm chronicling the cities where Dems hold political power but refuse to believe that non functional areas of their control actually are dumpsters. It could be done with an apolitical commentary or angle. People need to see the destruction that poverty, 911 is a joke, takes on liberal progressive cities.