> 3) Transformative technologies succeed by redesigning a task, and usually not by simply trying to reproduce the existing forms of human labor. Paige built an amazingly complex mechanical imitation of the human compositor which attempted to repeat a human worker’s actual movements. By contrast, Linotype simplified the problem by doing something humans didn’t already do.
That would in fact be the main argument against humanoid robots in many (most?) applications.
The counterpoint is that what humans already do is exactly what's needed in some cases, making those cases the ideal applications for humanoids.
Actually, no. Replicating the manual process was quite feasible.
There were keyboard-driven typesetters that worked well. Here's the Unitype.[1] This sets type by assembling type slugs into a line. It doesn't cast hot metal. It even has an automatic distribution system; it can sort type slugs by letter, so you can reuse the type. That's the same thing the Paige machine did. The Unitype was much simpler.
The Unitype was late to market. Mergenthaler already had the Linotype out when the Unitype shipped. The Unitype was half the price of a Linotype. Unitype machines sold for about 20 years, so it wasn't a total failure. The Paige machine uses a lot more parts and space to do the same job.
Mechanism design skill is very rare. All the good Teletype machines were designed by just two people - Howard Krum and Edward Kleinschmidt. There were competing machines, and they were all much worse. William Burroughs invented the first adding machine that worked reliably, and it was better than all the competitors. Burroughs dominated banking well into the computer era.
The video claims that the big problem with the Unitype that it required not just a keyboard operator, but someone manually justifying the lines at the output end. Linotype figured out automatic justification, using expanding wedges. No need for a second worker.
Plus someone had to feed used type back into the Unitype for sorting and return to inventory. With a Linotype, the matrices with the letter forms never leave the machine; they are dispensed, set into a line, used for casting, and then immediately recycled back to the magazine for further use. As long as the supply of lead bars ready to melt holds up, you can keep setting type. With ordinary type setting, mechanized or not, you can end up with all your type tied up in pages waiting to be printed.
So the real problem with the Unitype was that it wasn't labor-saving enough.
Despite that, it powered many small country newspapers for years.
They're Mediocre at everything, and worse than a dedicated machine at almost every task, just like a CPU is worse than an ASIC. However, they adapt to new tasks quickly. This decreases ownership costs (you only need one humanoid, not one machine per task), as well as manufacturing costs, as you can make one design and spend a lot of R&D money on scaling that design.
There's no reason why an ideal body needs to have such a high center of gravity though? Clearly we are wasting a lot of energy and capacity to maintain that balance.
I am pretty the ideal body is actually a spider.
The 30 second version, sewing is very labor intensive and people started trying to make a sewing machine from about when manufacturing tolerances got tight enough to attempt the process. It still took about 100 years until machines that actually work well were invented, The genius bit that took so long is to not try and sew like a human does but to develop an inverted process that a machine can do. Honestly, despite being quite common at this point in time sewing machines still blow my mind when you consider the intricate process they are required to do with thread.
And a fun fact for free: Apparently patent cartels(cough MPEG LA) are not a new thing and Singer et al set one up in the 1850's to lock up the market on these marvelous machines.
There's also plenty of examples where technology creates something very different than the thing that humans were doing: photograph vs painting, car vs horse/bike. It would be silly to say that we have all this horse infrastructure and saddle UX that we don't want to break compatibility with.
Maybe the lesson is that humans are adaptable but machines/manufacturing is not? It's easier to make a great new thing that humans will use than it is to make a cost-effective production thing that exactly replaces what humans did.
This whole episode suggests to me that people are misjudging the future of robotics. Many companies are building robots to exactly imitate humans, perhaps with the thought that the "egocentric" training data widely available will be more useful for this form factor. Look at Tesla, Unitree, Boston Dynamics, or Figure... But humans are just barely adapted to bipedalism and hunter-gathering as it is.
The perfect kind of robot is perhaps something closer to the linotype... a AI controlled macine that has the capability to birth or re-assemble to be optimal for the task at hand. Just like the linotype automated the "casting" of lines of text.
Not necessarily a 3d printer or a nanobot universal assembler... but maybe something more akin to a modular assembly line that can rearrange itself intelligently.
If you have a general solution like a robot that can do human tasks, i.e. copying manuscripts with a pen, that is often highly inefficient compared to a specialized solution like a printing press but the R&D effort of applying it is minimal compared to inventing a specialized machine.
It's similar with LLMs. Compared to a bespoke piece of software for the task they are incredibly inefficient, yet they are extremely useful because they can do things where developing that piece of software would require a lot of research.
Well, LLMs are a great example of the argument for the power of doing what "humans already do": language, however imperfect for the task, so maybe humanoid robots turn out to be the most natural way to interact with a human environment (broad strokes here, I know, but hopefully you get the gist)
Another point on this is that as you improve the robot you improve those endless tasks it can do, whereas with specialized robots improvements to that specialization are much more limited in scope. So there's a lot more incentive and value to be gained from improving the tech than there are for specialized solutions. Sike with LLMs.
I wish school curriculum included things like "how printing actually worked through ages". Monks writing out manuscripts is somewhat relatable since we all (at least for now) still learn how to write. Everyone is taught the Gutenberg press was a big deal -- but how did it actually work? Why was it still a pain and how did later inventions fix those problems?
I was able to do something probably few have, which is visit one of those small hole in the wall museums staffed essentially by just a single passionate old guy assembled with his life savings, dedicated to the printing press. Incredible experience.
Anyways, it’s been about 15 years, but the thing that made an impression on me was that it wasn’t just one invention. Gutenberg basically had to assemble like two dozen separate innovations to make it all work.
The type alloy (3 metals!) had to be just soft enough to cast easily, just expandable enough to set well, just hard enough to last. How to cast in quantity and re-melt as needed. The ink had to be just runny enough to apply easily but not so runny it gets everywhere, sticky enough to stick to type, and dry at a good middle speed. The blotter to apply the ink to type. The system of wooden casing to hold the pieces together with the right spacing, that won’t fall apart. How to set up the casing. The screw press needed to have very even pressure. The paper needed to be able to slide and take ink properly without spreading but also not tear. Proper fonts and ligatures for the type. And so on…
IIRC most things stayed pretty similar for a few hundred years, mostly changing stuff like the paper feeding mechanism or tweaking ink application or minor variations in how the lever or screw work. So to be honest, Gutenberg did almost all the hard work himself. It was a big deal because a lot of these smaller inventions had to come together for a narrow purpose all at once (on top of the obvious huge historical effect of cheaper books!)
Gutenberg's press was made out of a wine press. Letters were cast in lead, then lined up on a tray under the press. The tray was inked. A piece of paper was set over the inked letters, and the wine press forced the paper onto the letters. This cut the cost of a book by a factor of 300(!).
The trouble for Gutenberg was it was too easy to duplicate his press, and that combined with not being good at business led to his bankruptcy.
> overestimate how well they learn things without the feedback of formal education
People benefit greatly from external human feedback, but 1:1 tutoring (with a good tutor) is roughtly an order of magnitude more effective and efficient than a classroom, and classrooms themselves vary wildly; in general the more participatory the better.
Self learning without any help can be more effective and efficient than a typical classroom, because as it is self directed, self paced, can use the best materials someone can find and a whole variety of them vs. whatever was assigned, etc., but it takes a lot of skill (which comes from past practice) and effort on the part of the learner. Many researchers/scholars are doing significant amounts of self-directed learning as their career, and can get very good at it with decades of practice.
The speed of learning in a classroom with 15+ students is inevitably unbelievably slow, and the time is used incredibly inefficiently. This isn't any particular problem with specific teachers, but as far as I can tell an inherent problem with aligning a bunch of people's preparation, motivation, and attention, and a problem with giving feedback in a 1:many setting. Teaching/learning speed generally seems worse at a primary school level vs. at a college level, given a very good teacher in both; I'm not sure if this is because of better prepared (because older) students, or just more motivated students.
It’s not only feedback, having arbitrary checkpoints like exams is an important motivator, and it also creates a whole culture around it that makes the learning experience into a social one.
No, no, just the opposite: school curricula are full of lofty highfalutin ideas.
I didn't say anything is hopeless. It's just that adding more good and lofty ideas and topics to the curriculum isn't addressing the actual bottleneck.
Yet the end result appears that people on their own accord go and "learn" things like vaccines are bad, fluoridated water is used for mind control, rectal enemas can cure bowel cancer, and that the moon landings were fake.
At least the education system enforces some baseline level of standards
That's really awesome! Did they have you try and typeset something by hand? I feel like for us it presented as some magical thing that made producing books trivially easy, whereas in reality typesetting a page was still very labor intensive and not that fast!
We learned about how the Gutenberg Press physically worked, how it was based on wine presses and how difficult it supposedly was to produce a single metal letter. Also, how it was a PITA to re-arrange them to print something else, alas it was mostly used to print the Bible.
I wish the article talked more about how the machine actually worked. It seemed pretty simplistic and glossed over most of the details, but now I'm really curious and want answers lol.
I don't like the first photo, historical photos getting colorized does not feel right to me, as all the colors are just assumptions and the whole thing may have looked different in real and if we look at this photo it sure did
When I heard that story, I couldn't help but thing of Steve Jobs "Real Artists Ship." The device was good but the inventor was a perfectionist instead of an entrepreneur.
super interesting article thanks never heard of the entire story. uncanny parallels to todays humanlikewritermachines and the billions being lost and won there.
> The easy moral of the story is that Twain was gullible and Paige a huxster.
It absolutely isn't. Paige was actually the engineer/tinkerer earnestly developing this machine, which really worked. One of his main problems is that he always thought he needed to improve it more before it was ready. He didn't understand "release early and often": that time-to-market matters.
> Twain felt that he had an insider’s perspective and could intuit how others in publishing would respond to the machine. But he was a printer’s devil no longer: he was a famous, millionaire author.
No longer, perhaps, but Twain had been in the printing industry, so he knew a thing or two. From age 13, Twain worked as a typesetter and compositor for newspapers. So even if he overestimated his knowledge of that, it was not simply the conceit of someone cushioned. What he remembered from his past experiences with freely moving type set by hand would have still been relevant at the time he embarked on this thing with Paige. Linotype was not yet in the market.
Even as the Linotype appeared, it was not obvious that it is superior. It is a gross simplification to say that Paige's machine was imitating manual typesetting whereas the Linotype wasn't. Both machines work by moving letter blocks into rows, and the letters have to be sorted.
Evidently, what the Linotype did better was:
- Linotype characters are negatives, made of brass. Brass is hard and durable. These negatives are cast to positives with molten lead. The lead is melted for reuse after a run of the typesetting. Paige used cast lead pieces, which were more or less permanent. The same letter "A" would be used over and over again in different typesetting jobs. Lead is heavy to move around in the machine as well as soft. Brass allows nicks or slots to be cut in it for easy sorting into bins, but if you do the same with lead, it won't be durable; these features will wear out and cause missorting and jams.
- Paige's system used a complex system for justifying words and filling spaces at the ends of paragraphs and whatnot; it had to calculate and insert spacers. Linotype had clever tricks: special wedge-based expander bars were inserted between words, and then pressure against the bars expanded the sliding wedges to justify an entire line at a time. So in that regard, that's where Linotype deviated from trying to imitate manual typesetting.
- A key difference is that Linotype makes one line at a time, out of brass moulds, with spacers between. This is then cast with molten lead: just one line, called a slug. After a line is cast, the brass moulds can be released and sorted back into bins. The lead slugs are assembled into a page. Paige's system, maybe because of his surname, tried to work more traditionally, typesetting a whole page. That creates a problem because of the limitations of the workspace. If you are doing lines separately, you can have machinery/tooling/jigging above and below the line; there is no consideration for interference with previous or subsequent line of text. This freedom is what enabled the mechanisms for spacing to be developed.
But the thing to note is, that the Linotype changed typesetting work into something more heavily industrial, involving continuous hot metal casting.
Paige and Twain might have been betting on such a change being rejected by the industry, in favor of their solution which was doing the same sort of typesetting.
Lineotype they use lead letters. They use an alloy. There was a lot of lead, but there's also other metals in it because lead is too soft to be useful in a press. The alloy had the interest stream property that it has very low shrinkage when it cooled.
Classic letterpress type is lead, antimony and tin, but I don’t know if the so-called hot lead machines used that particular alloy. People usually just call it all “lead type” though.
Linotype machines were revolutionary. The New York Times used them for 127 years until 1978 and this documentary shows them in action for the last time:
The wedge-shaped expander bars you mention are a brilliant bit of engineering: they deliver perfectly equal spacing across the entire line without any calculation or measurement needed.
Reading about a machine from one hundred and fifty years ago and getting progressively more and more upset that the writing doesn’t give sufficient praise to the real topic of every article, my lovely chat bot
My lovely pocket calculator, he says these are lessons for us, for the present. And, since he is talking about technology in general, presumably for the future. The past events were only the inspiration.
That would in fact be the main argument against humanoid robots in many (most?) applications.
The counterpoint is that what humans already do is exactly what's needed in some cases, making those cases the ideal applications for humanoids.
The Unitype was late to market. Mergenthaler already had the Linotype out when the Unitype shipped. The Unitype was half the price of a Linotype. Unitype machines sold for about 20 years, so it wasn't a total failure. The Paige machine uses a lot more parts and space to do the same job.
Mechanism design skill is very rare. All the good Teletype machines were designed by just two people - Howard Krum and Edward Kleinschmidt. There were competing machines, and they were all much worse. William Burroughs invented the first adding machine that worked reliably, and it was better than all the competitors. Burroughs dominated banking well into the computer era.
The video claims that the big problem with the Unitype that it required not just a keyboard operator, but someone manually justifying the lines at the output end. Linotype figured out automatic justification, using expanding wedges. No need for a second worker.
Plus someone had to feed used type back into the Unitype for sorting and return to inventory. With a Linotype, the matrices with the letter forms never leave the machine; they are dispensed, set into a line, used for casting, and then immediately recycled back to the magazine for further use. As long as the supply of lead bars ready to melt holds up, you can keep setting type. With ordinary type setting, mechanized or not, you can end up with all your type tied up in pages waiting to be printed.
So the real problem with the Unitype was that it wasn't labor-saving enough. Despite that, it powered many small country newspapers for years.
[1] https://youtu.be/5n5JQrN8qx4?t=152
They're Mediocre at everything, and worse than a dedicated machine at almost every task, just like a CPU is worse than an ASIC. However, they adapt to new tasks quickly. This decreases ownership costs (you only need one humanoid, not one machine per task), as well as manufacturing costs, as you can make one design and spend a lot of R&D money on scaling that design.
The 30 second version, sewing is very labor intensive and people started trying to make a sewing machine from about when manufacturing tolerances got tight enough to attempt the process. It still took about 100 years until machines that actually work well were invented, The genius bit that took so long is to not try and sew like a human does but to develop an inverted process that a machine can do. Honestly, despite being quite common at this point in time sewing machines still blow my mind when you consider the intricate process they are required to do with thread.
And a fun fact for free: Apparently patent cartels(cough MPEG LA) are not a new thing and Singer et al set one up in the 1850's to lock up the market on these marvelous machines.
There's also plenty of examples where technology creates something very different than the thing that humans were doing: photograph vs painting, car vs horse/bike. It would be silly to say that we have all this horse infrastructure and saddle UX that we don't want to break compatibility with.
Maybe the lesson is that humans are adaptable but machines/manufacturing is not? It's easier to make a great new thing that humans will use than it is to make a cost-effective production thing that exactly replaces what humans did.
https://en.wikipedia.org/wiki/Elias_Howe#Invention_of_sewing...
The perfect kind of robot is perhaps something closer to the linotype... a AI controlled macine that has the capability to birth or re-assemble to be optimal for the task at hand. Just like the linotype automated the "casting" of lines of text.
Not necessarily a 3d printer or a nanobot universal assembler... but maybe something more akin to a modular assembly line that can rearrange itself intelligently.
It's similar with LLMs. Compared to a bespoke piece of software for the task they are incredibly inefficient, yet they are extremely useful because they can do things where developing that piece of software would require a lot of research.
Anyways, it’s been about 15 years, but the thing that made an impression on me was that it wasn’t just one invention. Gutenberg basically had to assemble like two dozen separate innovations to make it all work.
The type alloy (3 metals!) had to be just soft enough to cast easily, just expandable enough to set well, just hard enough to last. How to cast in quantity and re-melt as needed. The ink had to be just runny enough to apply easily but not so runny it gets everywhere, sticky enough to stick to type, and dry at a good middle speed. The blotter to apply the ink to type. The system of wooden casing to hold the pieces together with the right spacing, that won’t fall apart. How to set up the casing. The screw press needed to have very even pressure. The paper needed to be able to slide and take ink properly without spreading but also not tear. Proper fonts and ligatures for the type. And so on…
IIRC most things stayed pretty similar for a few hundred years, mostly changing stuff like the paper feeding mechanism or tweaking ink application or minor variations in how the lever or screw work. So to be honest, Gutenberg did almost all the hard work himself. It was a big deal because a lot of these smaller inventions had to come together for a narrow purpose all at once (on top of the obvious huge historical effect of cheaper books!)
The trouble for Gutenberg was it was too easy to duplicate his press, and that combined with not being good at business led to his bankruptcy.
It's never been easier for you and interested kids to learn these things on their own.
Easier on its own is therefore effectively meaningless when people aren’t actually learning beyond the most superficial layers.
People benefit greatly from external human feedback, but 1:1 tutoring (with a good tutor) is roughtly an order of magnitude more effective and efficient than a classroom, and classrooms themselves vary wildly; in general the more participatory the better.
Self learning without any help can be more effective and efficient than a typical classroom, because as it is self directed, self paced, can use the best materials someone can find and a whole variety of them vs. whatever was assigned, etc., but it takes a lot of skill (which comes from past practice) and effort on the part of the learner. Many researchers/scholars are doing significant amounts of self-directed learning as their career, and can get very good at it with decades of practice.
The speed of learning in a classroom with 15+ students is inevitably unbelievably slow, and the time is used incredibly inefficiently. This isn't any particular problem with specific teachers, but as far as I can tell an inherent problem with aligning a bunch of people's preparation, motivation, and attention, and a problem with giving feedback in a 1:many setting. Teaching/learning speed generally seems worse at a primary school level vs. at a college level, given a very good teacher in both; I'm not sure if this is because of better prepared (because older) students, or just more motivated students.
I agree that school curricula is often bad. I disagree that we should just accept it as hopeless and not hope it could get better.
I didn't say anything is hopeless. It's just that adding more good and lofty ideas and topics to the curriculum isn't addressing the actual bottleneck.
At least the education system enforces some baseline level of standards
To see a typecaster which works find a Linotype or Monotype.
https://circuitousroot.com/artifice/letters/press/noncastcom...
https://readmeastoryink.com/wp-content/uploads/stories/the_c...
It absolutely isn't. Paige was actually the engineer/tinkerer earnestly developing this machine, which really worked. One of his main problems is that he always thought he needed to improve it more before it was ready. He didn't understand "release early and often": that time-to-market matters.
> Twain felt that he had an insider’s perspective and could intuit how others in publishing would respond to the machine. But he was a printer’s devil no longer: he was a famous, millionaire author.
No longer, perhaps, but Twain had been in the printing industry, so he knew a thing or two. From age 13, Twain worked as a typesetter and compositor for newspapers. So even if he overestimated his knowledge of that, it was not simply the conceit of someone cushioned. What he remembered from his past experiences with freely moving type set by hand would have still been relevant at the time he embarked on this thing with Paige. Linotype was not yet in the market.
Even as the Linotype appeared, it was not obvious that it is superior. It is a gross simplification to say that Paige's machine was imitating manual typesetting whereas the Linotype wasn't. Both machines work by moving letter blocks into rows, and the letters have to be sorted.
Evidently, what the Linotype did better was:
- Linotype characters are negatives, made of brass. Brass is hard and durable. These negatives are cast to positives with molten lead. The lead is melted for reuse after a run of the typesetting. Paige used cast lead pieces, which were more or less permanent. The same letter "A" would be used over and over again in different typesetting jobs. Lead is heavy to move around in the machine as well as soft. Brass allows nicks or slots to be cut in it for easy sorting into bins, but if you do the same with lead, it won't be durable; these features will wear out and cause missorting and jams.
- Paige's system used a complex system for justifying words and filling spaces at the ends of paragraphs and whatnot; it had to calculate and insert spacers. Linotype had clever tricks: special wedge-based expander bars were inserted between words, and then pressure against the bars expanded the sliding wedges to justify an entire line at a time. So in that regard, that's where Linotype deviated from trying to imitate manual typesetting.
- A key difference is that Linotype makes one line at a time, out of brass moulds, with spacers between. This is then cast with molten lead: just one line, called a slug. After a line is cast, the brass moulds can be released and sorted back into bins. The lead slugs are assembled into a page. Paige's system, maybe because of his surname, tried to work more traditionally, typesetting a whole page. That creates a problem because of the limitations of the workspace. If you are doing lines separately, you can have machinery/tooling/jigging above and below the line; there is no consideration for interference with previous or subsequent line of text. This freedom is what enabled the mechanisms for spacing to be developed.
But the thing to note is, that the Linotype changed typesetting work into something more heavily industrial, involving continuous hot metal casting.
Paige and Twain might have been betting on such a change being rejected by the industry, in favor of their solution which was doing the same sort of typesetting.
It glosses over it and Twains autobiography goes into more detail.
What impressed Twain about it was the demonstration had an operator typing at high speed.
When he got the device, he discovered the operator was typing the same phase over and over. They had memorised it so as to type faster.
So one of the overlooked lessons is don't believe product demos.
https://www.youtube.com/watch?v=LTiCTFdTd60
> 2) Society uptake of transformative inventions is almost always slower than their proponents expect.
... except for LLMs.
> 3) Transformative technologies succeed by redesigning a task, and usually not by simply trying to reproduce the existing forms of human labor.
... except for LLMs.
Seriously, that’s your best example of that lately? Not, I don’t know, Tesla FSD? Theranos? Bernie Madoff?