Antique Mechanical Computers – Part 2: 18th and 19th Century Mechanical Marvels (Aug, 1978)
Be sure to check out Part 1.
Antique Mechanical Computers Part 2: 18th and 19th Century Mechanical Marvels
Dr James M Williams
58 Trumbull St
New Haven CT 06510
In “Part 1: Early Automata,” page 48, July 1978 BYTE, we traced the development of antique mechanical computers up to the middle of the 18th century, and described such devices as Vaucanson’s mechanical duck. Now we continue with a discussion of talking, writing and music playing automata of the 18th and 19th centuries. (The discussion is not meant to be an exhaustive one, of course, since that would be beyond the scope of this series.) Later Automata.
Vaucanson’s creations blazed across the scene in Europe 240 years ago, casting new light into hitherto dark places by showing what the dedicated mechanician could achieve. But, even after Vaucanson, the way was difficult. 38 years passed before a second flute playing machine was seen, a seated pair of rustics built by Duchamps in 1776 and said to be capable of playing 13 tunes. 109 years after Vaucanson made the original mechanical duck, a mechanician named Rechsteiner, who had restored that original duck, produced and displayed a duck of his own. Rechsteiner’s duck was the product of three years of work. It appeared in 1847 and was the last automaton animal of note.
In the last quarter of the 18th century, first a few, then nearly a flood of automata began to appear, as clockmakers began to realize not only the possibilities of their craft but also the splendid prices their premier work might command. The more standard automata such as ornamented clocks, from snuffbox size to prodigies bigger than steamer trunks, with processions of moving allegorical figures, spirals, pin-wheels, and waterfalls, chimes, bells, dulcimers, whistles, organs, and birdcalls, continued to be made and sold. Every titled person had a score of them and men of substance could own several. The clock-maker of ambition knew where his challenge lay. There were mysteries to be created in machinery, and money and fame to be had. Mechanicians began to devote themselves to duplicating the physical action of parts of the human body. They chose part-behavior because of the immense difficulty of fabricating a mechanism that could imitate even one of the coordinated acts humans orchestrate into the continuous chain of actions; namely, behavior.
It is worth noting that in adults the discrete units of purposeful action which seem so integrated and effortless to most of us are anything but smooth and coordinated in early childhood. Most people can recall their clumsiness and exasperation in learning to tie their shoes or button their garments. The most intense concentration and dedicated repetition is required to cause these action patterns to set in our central computing mechanism (see “The Brains of Men and Machines,” parts 1, 2, 3 and 4, February thru April 1978 BYTE), but once the setting (ie: learning) takes place over time, it becomes possible for us to execute one of these unit actions at will, devoid of effort and concentration. (The mechanism and locus of the setting is obscure: so is other memory storage. Lately, the cerebellar complex is viewed as the best candidate for unitary motor actions.) We can tie shoelaces behind our backs, a thing we never practiced or learned. Even extreme situations, like tying shoelaces while wearing mittens or hanging by the knees from a trapeze, do not begin to strain the capacities of our interior computing mechanism. The required actions have been “frozen” into our brains. Not only are they refractory to disarrangement (they endure as long as we live) but they are also flexible enough to permit our adapting them to novel circumstances. We all possess within us many thousands of such unitary chunks of learned behavior, now fully automatized and playable on command.
This is the part-behavior, smooth, continuous and automatic, that was being imitated by mechanicians. It requires substantial storage of program to duplicate. From our vantage point program storage is the most important feature these machines possessed. Consequently, many very beautiful mechanisms (the display pieces of Carl Faberge, jeweler to the Imperial Russian Court; a wide range of novelties such as soothsayers, magicians and other conjurers, acrobats and ropewalkers, agile harlequins and jugglers, automatic confectioners and wine stewards, and a great many more display mechanisms) are not mentioned here because they had little stored programming.
Walking and Running Machines.
Early walking and running automata were represented only by dolls and toys. They were essentially trivial, programmed devices for they always very ingeniously arranged an apparent walking action (only a simple repetitive motion). The walk lacked directionality, nor was there provision for walking on other than smooth surfaces. It would be difficult to design a machine to walk in the same sense that people do: that is, the weight of the trunk is for a moment supported by one leg alone while the other leg is being drawn forward for a next step. Walking is in fact organized falling, with the mobile extremity brought forward just in time to forestall disaster. When you stop to recall that every known mechanical man actually rolls on wheels, and that at least three wheels are always employed to define the plane, you gain a new respect for human locomotion and a valuable perspective on the limitations of mechanisms that undertake to imitate it.
As far as I can discover, no programmable device uttering words, or their approximations, was ever known before the late 19th century (or even in later periods up to the time of Bell Labs’ A/odor of 1939 World’s Fair fame, which required an operator). Still, some remarkable devices appear to have existed. Leaving aside the brazen talking heads that dot Greek and Byzantine mythology (they were without a doubt all hoaxes), we learn that the Abbe’Mical in 1774 was said to have exhibited two talking heads which he later destroyed. In 1779 Kratzenstein won a prize offered by the Russian Imperial Academy of Science for a device that could pronounce distinguishable vowels. This device was made from a set of five specially shaped pipes. Baron Wolfgang Kempelen, creator of the Great Chess Automaton, worked for many years on talking devices, and one was said by Goethe to be “.. .able to say some childish words very nicely.” The machine was a kind of bellows, soundbox, artificial tongue and mouth contrivance that the Baron manipulated under cover of a cloth; it now resides in a museum in Munich. Farber invented a machine which apparently spoke well enough to induce PT Barnum to purchase it for exhibition, in 1873. The device was operated by a keyboard.
It is a very curious thing that investigation of artificial speaking devices was so neglected by gifted mechanicians, for speech is the unique achievement of man. Moreover, the ear is so adaptable and forgiving of faults in the spoken word that virtually any kind of squawk might pass for a sentence. The mechanical problems would have been very great, but not insuperable.
Between 1753 and 1760 Friedrich von Knaus of Darmstadt devised and constructed four different machines that wrote block letters or cursive script according to programming using a quill pen and ink with programmed pauses to dip the pen. One machine produced three texts from three pens, while the last machine could inscribe up to 107 letters of preset text from its stored program or write individual letters one at a time from dictation under control of the operator. It may accurately be described as the first typewriter or scriptwriter. The mechanism appears to have been a cluster of shaped cams on which rode an array of cam followers, each one directing movements of the pen to form a letter. Text composition was managed by a drum that bore many rows of holes into which studs could be placed to activate the required cam. Thus text was easily altered by changing the pattern of studs. The tablet, bearing paper, moved one step after inscription of each letter. Knaus described his machines in a 1780 book, Selbstschreibene Maschine. His machine number 4 was shown at the Paris Exposition of 1937. It now resides in the Vienna Technical Museum.
The Automata of Jacquet-Droz and Leschot.
How can one describe machines so marvelously devised and “tutored” (ie: programmed) in their tasks that they rival the actions of human beings proficient in the art the machine imitates? One can compare them to humans and the analogy is intriguing, but humans are born with the necessity to learn many advanced action patterns and the automata were able to perform several advanced action patterns directly after construction. And humans age and die while the machines are two centuries old and act as well as the day they were set in place. They are seemingly flawless, ageless, potent and wise. And if you compare them to spirits you will be very nearly right, for they are shaped to resemble otherworldly creatures: cherubs or angels. If the compactness, beauty and simplicity of their mechanism with its nearly perfect functioning leads you to compare them to fine watches, you will be very nearly right again, for their builders were first of all horologists. They were the family of Jacquet-Droz (two brothers and a son) and Leschot, their master mechanician.
Long involved in making elaborate timepieces in Geneva, Jacquet-Droz the younger may well have been influenced by word of Knaus’ writing automaton. The Writer, Draftsman and Musician he designed and constructed, were placed on display simultaneously in 1774, and they have charmed every person who has seen them. They are on display in the Museum of Automata, in Neuchatel, 30 miles east of Geneva in western Switzerland. Consider the fact: here are devices seen and admired today, as well as by the courts of Louis XV, Louis XVI, George III, Napoleon and even by Franklin and Jefferson.
The Writer writes a preset text of 40 letters and spaces in about the same time and with quite a bit more skill than it might be written by an 8 year old child. The Draftsman draws a series of stored pictures, any one you choose, about as well as a gifted child of 12 years might do, while the Musician plays five melodies on her harmonium, as a musical child of 10 years might do. They have been performing these feats for 204 years.
The Writer is 28 inches (71 cm) tall. Carved of wood and painted, this automaton produces “an unusual impression of life” similar to top quality wax figures. He is clothed in a flowing robe and is seated on a Louis XV stool at a mahogany desk. His right hand, poised an inch above the desk and writing tablet, holds a short tube in which a quill pen is fixed. When the mechanism is activated the Writer raises his hand, swings it laterally, dips his pen into the inkwell fixed to the right margin of the desk, shakes the hand twice to clear the pen of excess ink and pauses. Another touch on the mechanism and he begins to write, forming letters with slow, patient care.
After each letter, the pad of paper moves to the left by an amount sufficient to leave space for the next letter, but more for a wide letter or a capital than for is and Is and fs. He can write 40 different letters on two or three lines, and there is programming for several pen dips. Most remarkable is the provision for the unit to vary the pressure of the pen so that the letters produced are weighted, formed of thick and thin strokes.
Except for the few levers controlling movements of the paper tablet, all of the automaton’s mechanism is contained in the torso, accessible from the back. There are two parts of the mechanism, and they interact with each other. The first is a cluster of letterforming cams on a common shaft, the cam follower of which rides on a carriage that slides on rails so it can cover the length of the cluster to settle on the rim of the desired cam. There are actually three cam followers and three cams provided for each letter. Two govern movements of the right arm and the third regulates pen pressure for varying the stroke width.
The second portion of the mechanism is the text selector, a disk 4 inches (10 cm) in diameter at the bottom of the cam cluster shaft. The rim of the text selector disk is divided into 40 sectors, or an angular wedge of 9° per sector. The sectors are not fixed, but rather slide radially when one of their 40 screws is turned. In this way the radius of the disk can be varied sector by sector, giving the appearance of a snaggle toothed gear. Each sector in turn regulates the position of the cam follower carriage (with its three cam followers) according to where that sector is set. Thus the text selector disk selects which set of three cams will be employed, and the letter those three cams control is the letter the right arm inscribes. Changing the text is as easy as turning 40 screws to just the right position. The zero radius (baseline) position of the text disk appears to control the pen dipping mechanism, so you can set up as many pen dips as you wish at the loss of a letter or space for each one.
Control is handed back and forth between the text selector disk and the letter forming cam cluster. Either one or the other operates at a given moment, but the text disk is stationary almost all the time (moving in jumps) whereas the cam cluster that forms the letters is moving most of the time (halting only to permit the text selector to turn to its next position and choose the next letter). An intriguing point, for 1774, is that the surfaces of greatest wear (the three cam follower bearing points) are apparently jewelled with ruby so that the high pressures (probably a 40:1 lever ratio, or more) will cause minimal wear and distortion of the letter shapes over time. All this machinery is said to be quite sensitive to temperature changes.
A point which is obscure to me is that the letter forming cams are alleged to operate on a polar coordinate system. Suppose the letters are formed on X-V coordinates. Photo 2 is a greatly magnified letter superimposed on a grid of 1 mm lines. Now you can appreciate the delicacy of the mechanism, for it is clear that a deviation of ±0.25 mm at any point will make a very different looking letter. (Incidentally, at a 40:1 lever ratio, a 0.25 mm movement at the pen is equivalent to 0.00625 mm on the cam face.) Clearly, the letters as inscribed on paper are well within this deviation (see photo 1 and figure 2).
Look how the es from several different words are exact duplicates: Probably the deviation is within about a tenth of that figure (ie: ±0.025 mm).
The mechanism is analog, of course, but if it were digitalized, the scale applied (resolution) has got to be less than 0.025 mm per bit, or in a letter of 8 mm height and 4.5 mm width:
8/0.025 = 320 bits for height.
4.5/0.025 = 180 bits width.
A grid of 320 by 180 equals 57,600 points, which would be the upper margin of the error. The limit is plus and minus this, so each letter may be digitalized with 57,600/ (2×2) = 14,400 points. But that is the amount for each letter, and we have 26 of them, which is 14,400 x 26 = 374,400. Adding upper case letters, the proper figure is 14,400 x 52 = 748,800 bits to digitalize the entire alphabet within the limits of error the machine consistently displays. You may wish to adjust the figures slightly because not all letters are the size of the y, and hence do not require as much storage of information (see photo 2). However, many letters fall below the line, and the capitals are larger than all the lower case, so it evens out. We have not taken account of the stroke shaping bits, which might require 4 to 6 more increments of information. Altogether, the machine’s “read only memory” has over three quarters of a million 1 bit bytes stored within it!
The Draftsman was constructed to resemble the Writer, and works in practically the same manner except that the tablet of paper is fixed, and the arm holds a pencil instead of a pen. The device moves under guidance of a cam cluster and draws designs in segments with pauses while the mechanism shifts from one cam pair to the next. During these pauses the Draftsman blows a puff of air from his lips to disperse the graphite debris. I would estimate that there might be 20 or more cam pairs for each of the four designs (there are no depth cams) on a slip of paper about 2 by 3 inches (5 by 7.5 cm). The designs were simplified reproductions of popular etchings of the age: cupids in chariots being hauled by butterflies, etc; and the head of Louis XV. The little Draftsman appears to have elicited a good deal more excitement than the Writer, but he was actually easier to construct, since the builders profited from their earlier experience with the Writer and simplified the mechanism.
Assume that the Draftsman’s paper is 50 by 75 mm, that any point on it could play a part in the design, and that it was necessary to provide a mechanism that could discriminate between lines as close together as 0.5 mm (ie: to a tolerance of ± 0.25 mm). You end up with a grid of 50/0.25 by 75/0.25 = 200 x 300 = 60,000 points that may be encoded. These were parcelled out among 20 “read only memory” cams. The total information contained in the machine would be 60 K bits by 4 designs = 240 K bits. The total information storage was much less because the eye can accept more line deviation in a drawing than in the formation of a letter.
The Musician is the triumph of automata that counterfeit life. She is 42 inches (1.07 m) high, seated at her instrument with a pleasant expression on her face. Her clothing is rich satin brocade in the elaborate style of the period, and her coiffure is impressive. She consecutively plays five pieces on her instrument, a curious device rather like a harmonium but called by some accounts a flute-organ, suggesting tuned pipes instead of metal reeds. The keyboard consists of two arcs of keys, 12 on a side. It is double arc shaped because the musician’s arms pivot at the elbows (concealed by lace sleeves on her gown) enabling her to cover all 12 keys with five fingers. The music, or most of it, was composed by Jacquet-Droz the younger, a musician who studied composition with Marchal.
She actually fingers the keys that produce the music! The mechanism to accomplish this feat consists of a connection for each digit, and some extremely clever devices must be employed to enable the arms to swivel while maintaining continuity for the digit controlling mechanism. I leave you to contemplate the delicacy of the arrangements of mechanism that trigger each finger in the tiny hands, but keep in mind that this machine is a workhorse; this musician has been playing music for 200 years.
Her programmed movements are startlingly lifelike in the accounts. All the jacquet-Droz and Leschot automata turn their heads and move their eyes, but this automaton also raises her head to look at the audience, drops her gaze, takes a deep breath, and starts to play. She turns her head as she plays and, swaying from side to side as artists will do, breathes all the while. At the end of a piece she looks up and seems to smile, then shyly lowers her gaze, drops her head, and curtsies.
with Jacquet-Droz (fils) in London, introduced a new and improved version of a lady musician. She played a sort of piano, perhaps actually a harpsichord, and it is known that she had 17 or 18 melodies in her programming. She was lost in 1833 when sent to St Petersburg together with other automata.
The Dulcimer Player of Roentgen and Kintzing first appeared in 1780, and was said by the magician Robert-Houdin (who repaired her in 1866) to have been designed to resemble Marie Antoinette and emulate her skill with the string dulcimer. This figure is famous for her beauty, and much praise has been lavished on her musical skill, for the instrument is clearly a difficult one to play (and is hardly known in this country). The mechanism is a cluster of cams mounted below the figure, concealed by her gown.
J N Maelzel, mechanician to the Austrian court and later the proprietor of the Chess Automaton, personally designed and had built a life-size Automaton Trumpeter, which he exhibited beginning in 1808. It was destroyed in a fire, about 30 years later. At least two other trumpeters have existed. What remarkable mechanism they must have contained, especially in view of In 1784 Maillardet, who was in business the praise their performances evoked. None survives.
Maelzel invented and displayed, beginning in 1804, the Panharmonicon, a compound musical mechanism which produced the sounds of flutes, trumpets, drums, cymbals and triangle, and plucked strings, a menage then called Turkish music and much favored by the public. This machine was followed by his Orchestrion, imitating the sound of the military band (which had become popular during the French Revolution). An improved Panharmonicon, with clarinets, violins, and violas added, was so well received that Maelzel commissioned music from Dussek, Pleyel, Weigl, and even Beethoven, whose “Wellington’s Victory,” opus 91, employing the Automaton Trumpeter as well as the orchestra, had its premiere on December 8 1813, in Vienna. These devices were the first of the programmable multiple instrument machines so popular 75 years later.
A Combination Automaton by Maillardet.
It was known that Maillardet, constant collaborator with the Jacquet-Droz and Leschot organization, had constructed a writing and drawing automaton about 1811, which was exhibited in London in 1815, and was owned by several persons until 1833 when it was sent to St Petersburg where it disappeared.
Long ago a resident of Philadelphia mentioned to a staff member of the Franklin Institute that his family owned an automaton that drew pictures and wrote poems. He supposed it to be Maelzel’s work. When the owner’s house was destroyed by fire, reducing the automaton to a “mass of cams and wheels,” the museum acquired it, but it took immense patience and care on the part of the museum restorer, Charles Roberts, to make the machine completely whole. In the restoration process the sex of the automaton was changed. When the time came to sample the machine’s program, it was found to be Maillardet’s missing automaton (see photo 3 in this article and Charles Penniman’s article, “Philadelphia’s 172 Year Old Android” in this issue, page 90).
The machine is about 30 inches (76 cm) high, and represents a child (originally a little boy, as alluded to in one verse, and in an 1812 encyclopedia article) kneeling before a desk and holding, since restoration, not a brush but a pen. The mechanism is in the base and consists of a common shaft holding about 60 cams, each one 6 inches (15 cm) in diameter. The whole is driven by a pair of powerful spring motors. Three triplets of cams are devoted to each of the seven productions of the automaton, except that the depth cams are minimally employed. The follower arms, one for each dimension of the drawing, are jewelled and move from pair to pair of cams in the course of one machine cycle (one drawing). The automaton executes its seven productions rapidly, completing one in 7 to 8 minutes. This would appear to explain Maillardet’s need to skeletonize the 60 programming cams: they turn rather swiftly (about 3 mm of linear motion per second) and at changeover they must be brought quickly to a halt, then accelerated to working speed again. Storing all information on three pairs of large cams per production would have made grinding the cam faces much easier, and would have minimized the effects of wear compared to a small cam. Shifting to a new program is done by simply sliding the common shaft laterally to set up a new triplet of cams.
Maillardet evidently took it as his task to produce a machine that worked on its productions rapidly and casually, perhaps in the manner of a person inspired. The sketches are marked more by fluency of line than by precision, but they are very sophisticated, as a glance at the ship sketch will show (see page 91). The poetry is interesting and is done more in the manner of a design with scriptwriting than writing in script (see figure 2).
In terms of brute force memory storage, if each of the points 1 mm apart on an 89 by 120 mm paper is to be stored, 10,680 points would be required. But discriminating between points with an error of no more than 1 mm requires ± 0.5 mm precision, resulting in 42,720 points that must be stored on the three triplets of cams. But this is the amount of point storage required for one production. There are seven of them, so the total storage capacity within the machine is 42,720 x 7 = 299,040 points (with ±0.5 mm precision). This figure, the digital equivalent of the analog storage, begins to make the impressive forest of cams seem more useful.
All of the above speaks about the information capacity (in terms of a grid of points) necessary to encode the designs and script that our automata can produce by analog means. The great majority of those digital data would not be employed in a display, just as an automaton will not inscribe marks on, say, more than 2 percent of the area of paper available to it. There is a lot of wasted (unused) space in any character generator. For example, most of the billions of micrograins of silver halide in a sheet of emulsion are not actually developed and play no part in a photographic negative.
The same is true of standard character generator read only memories where the 5 by 7 matrix with its 35 points is the absolute minimum matrix you can employ and still produce recognizable, if not particularly legible, alphanumerics. Even so, 50 percent of these bare minimum 35 points are not utilized for any given character, hence there is 50 percent waste. Premium character generator read only memories are set up to use a great many more points, and their displays are still manifestly coarse in structure (“crude” would not be too strong a word, when you know there is something much better).
Here we are simply making visible the difference between analog and digital modes of storing information. The analog mode is obviously more economical, for there are nowhere near 750,000 jiggles in 20 cams.
Next month, we’ll conclude with a final example in this series about antique automata, the chess playing robot of Torres (circa 1911) and some philosophical comments on automata.”