Programmable, 19th century, 6,000-piece automaton: the Writer




18th century, not 19th.


This reminds me of The Invention of Hugo Cabret


Holy CRAP! This is almost as amazing as the Antkythera mechanism in its antiquity vs. sophistication, and even still works; amazing.


Very cool. I wonder if the plans for this survived? Too bad about the hokey dramatic soundtrack on the video though.


Tuning a device like that must be the stuff of heroism and madness...

Even in our present day, of advanced metallurgy, computerized machine tools, and similar neat stuff, vendors specify tolerances for parts(and definitions of 'interchangeable' are not always as... rigid... as they might be). With 18th century fabrication techniques, 6,000 mechanical parts must have been a nightmare of slight variations, variable deformation under load, and general 'file to fit, until the madness comes' issues.

I know Babbage ran into trouble for (among other reasons) issues of mechanical precision, and that was on a device dealing with discrete values...


Shuffle the cards and it foretells the future.


That is because the prop on that film was based on Pierre Jaquet-Droz's work.


18th century spambot.


Speaking of which, I bet the mortgage industry would really find this useful!


Maybe -- if the 6000 parts includes all the cams and all the interchangeable letter keys on the main rear disk, it's really not as bad as it sounds. each cam is engaged only now and then. Further, it looks like one part of the stack is purely for the eye movement, and another for the left hand to move the writing plate, so the amount of mech required to run the pen & right arm is not all that frightening.
I mean, take a look inside one of the great early 70's Gottlieb EM pins, and you'll see a crapload of relays, solenoids, and yes, cam wheels. smile


pierre jaquet-droz, his son henri-louis, and their business partner jean-frédéric leschot built another three automata of equal quality and virtuosity. two of them are, like the writer, more specifically categorised as androids, since they are shaped like humans and actually perform an action instead of simulating it. (the more general term, automaton, can come in any shape — mechanical singing birds, for example were automata that only twitched their beaks, simulating the action of singing while the birdsong came from whistles hidden elsewhere in the mechanical structure.)

these two androids are known as the draughtsman, which can draw pre-programmed images, and the musician, who plays pre-programmed tunes on a customised harpsichord. the level of verisimilitude is incredible, the writer taps his quill to get rid of excess ink before writing, the draughtsman periodically blows a puff of air onto the sheet of paper to remove bits of graphite that might smudge, their eyes follow the movement of their hands etc. the last grand automaton of jaquet-droz was a full-on pastoral scene that included mooing cows, frolicking shepherds, farmers on donkeys, barking dogs and the like, but it's been lost since the late 18th century.

bit of trivia — the draughtsman and the writer both used to have very nicely made shoes to match their rich clothes, but they were removed when being exhibited, to show audiences that the mechanism was entirely contained within the androids and that they were not worked through some massive contraption that led up through their legs. with all the taking off and putting back on, the shoes got lost over the years. they're all on display at the museum of art and history in neuchatel, switzerland.

the jaquet-droz trio made these things as show-off pieces for their watch business and used them to impress wealthy and aristocratic audiences all over europe, to whom they would then sell smaller mechanical objects — watches, singing-bird snuffboxes etc. the modern-day swiss watch company jaquet droz continues in this tradition, making mechanical watches with integrated automata. pretty amazing stuff.

(source of information: i write for and edit a magazine on luxury mechanical watches.)


In some ways, analog writing might be easier than digital calculation. If the line of a letter is a little titled to the right, no problem. If your digital "1" comes out a "0", not so good. Granted, it's incredible regardless. smile


This cuts both ways: it is true that a bit flip is not acceptable, while minor analog deviations often are; but the nice thing about digital systems is that, at any step along the way, you can look at the signal(whether it be electrical, mechanical, light, whatever) and ask 'Is this a 1 or a zero?' and then generate a new, perfectly pristine, 1 or zero to pass along to the next step in the process. For practical reasons(eg. CPUs often run with '1' represented by ~1.2-1.5v, because higher voltages consume considerably more power and can destroy very small transistors) digital systems tend to try to keep the swing between 1 and zero relatively small; but you can make it bigger if needed(eg. to support reliability with fairly primitive hardware over long, potentially noisy, signal runs, RS-232 serial is '1'=15v, '0'=-15v, for a 30 volt swing, and a 15 volt swing relative to ground in both directions). Some systems, albeit relatively few today, use more than two values, in order to pack data more densely; but at the cost of having to identify discrete states that are less widely separated. "Multi-level cell" flash memory is probably the most common example. It's cheaper, since you can represent more binary bits in the same number of cells if you are using 3 or 4 discrete states; but it's also less reliable, since the discrete states are not as distinctly separated from one another, and easier to misread. In principle, you can use any number of discrete states greater than one(unary logic is... a trifle limiting... in ways that binary logic is not); but building systems that do binary logic really, really, fast seems to have carried the day in most places.(that, and drawing the truth tables for a logical operator in a logic system of Aleph-0 values takes quite a while...)

With an analog value, you are subject to the same unpredictable perturbations; but, since the 'legitimate' values aren't discrete (there may be a min and a max; but anything in between is fair game) you have no way to know whether what you have is nearly perfect or badly wrong. So errors you make have a nasty tendency to, at best, propagate and, at worst, get amplified in some later stage.

(Purely by way of interest, there are some clever tricks used in analog transmission to provide otherwise unavailable information about what is signal and what is noise; but that's the fundamentally tricky bit about analog design.)


This topic was automatically closed after 5 days. New replies are no longer allowed.