A column about curious stories from the worlds of science, technology, and film. Today—no film, but plenty of plastic springs. The occasion: this week Habr came across a post about digitizing the 1987 manual for the “Detskaya logicheskaya mashina (Children’s Logic Machine, DLM)”—a Soviet toy with not a single transistor inside. Its author (drlexium) scanned the booklet, built a browser emulator, and, after running automated tests, found factual errors in the instructions—including a game where the machine can be beaten, and where the manual itself suggests the losing move. But the most curious thing here is not the bugs in the manual, but the lineage: this box with “A-332” batteries is a direct descendant of machines that solved syllogisms under George III.
The “Detskaya logicheskaya mashina” (DLM) was produced by the All-Union Scientific Research Institute “Elektronstandart” in Gatchina from 1984 onward. Inside—zero electronics: three sliding bars, which switch A, B, or C moves right or left depending on the input bit; seven columns of paired sockets (14 sockets per bar) with spring contacts; thirty plastic plugs; and four light bulbs with 8-4-2-1 “weights.” A plug in the left socket closes the circuit when the bar moves right; in the right socket, when it moves left. The three contacts in a column are connected in series—that is AND; wire jumpers combine columns into one lamp—that is OR. In essence, the machine simultaneously computes seven products such as A·¬B·C and adds them by OR: a programmable 3×7×4 logic matrix molded from polystyrene. There is no “computer” inside—the entire “thought” takes place in the topology of closed contacts.
A program is the physical placement of plugs according to a diagram in the manual. The manual promised eleven games: from guessing a chosen number through crafty questions about remainders, to the game “Twelve” against the machine itself and a “psychological test” for which the machine was reprogrammed as you went along. The author of the digitization exhaustively ran all 34,650 layouts of the card game “Zoological Cards” and discovered that in two cases out of seven, the program from the instructions gives the wrong prediction; in “Twelve,” there was a position where, with a three on top, the rule “place the nearest lower one” forces you to hand victory to your opponent—even though a six lies beside it, and 6+6 ends the game immediately. A solitaire game in which a 1987 Soviet instruction manual lies was found on a quiet autumn evening with the help of automated tests—the machine did not need a single clock cycle for this. It simply has none.
But dig deeper, and the toy stops being a curiosity. The DLM is the degenerate—in the evolutionarily respectful sense—last member of a dynasty of mechanical logic machines stretching back to the 18th century. And this dynasty has its English branch, its Russian story involving a professor of chemistry, and even its early argument over whether a machine can “think”—the exact argument now playing out around LLMs.
The first true logic device in history is generally considered Charles Stanhope’s “demonstrator,” built by the third Earl Stanhope—a statesman, amateur scientist, and inventor whose name is attached both to the 1800 cast-iron printing press and to the popular little optical magnifier known as a “stanhope.” His first apparatus dates from 1775: a square brass plate around four inches across, mounted on mahogany. At its center was a recess with the poetic name “holon” (“whole”): the complete set of objects under discussion. From the right, a translucent red glass panel slid in; beneath it from the left, a gray wooden one. Each panel represented one of the two affirmative premises of a syllogism—Stanhope himself called them “ho” and “los.” Sliding them over one another, the owner literally saw the area of intersection—and the answer to the syllogism became an area visible to the eye. A scale from 0 to 10 also made it possible to solve quantitative syllogisms such as “8 out of 10 A are B”—what de Morgan would later call the numerically definite syllogism (Formal Logic, 1847), unaware that the lord had done it seventy years earlier.
The device also had probabilistic logic: for the “rule of probability,” the gray panel was inserted from above; for the “rule of certainty,” from the left. At least four copies were made; two were kept by Henry Prevost Babbage—the son of that very Charles Babbage who worked to complete his father’s Analytical Engine. The Stanhope family donated one copy to London’s Science Museum in 1953. Tellingly, the earl barely published the device at all: finished instruments were handed out to acquaintances, and the world learned of it through Robert Harley’s letters of 1879—a century after its creation.
The next leap was made by William Stanley Jevons—economist, logician, and, by a strange turn of circumstance, former assayer at the Sydney Mint during the gold rush. In 1869, he hired a Salford watchmaker—whose name he did not give, and their relationship was tense: “I have to go almost every day to see that he does it right”; every stage of the work was accepted separately—and received a machine he himself christened the logical piano. Outwardly, it was a miniature upright piano roughly three feet tall, with a keyboard of 21 keys, “exactly like the white keys of a piano.” In windows on the front panel, all 16 combinations of four terms and their negations were visible. The logic was this: you enter premises with the keys, and concealed levers and pulleys mechanically eliminate the combinations incompatible with what has been entered. When the premises run out, you look at what remains: those are all the logical consequences.
In 1870, Jevons demonstrated the machine at a meeting of the Royal Society of London, and that same year his substantial paper “On the Mechanical Performance of Logical Inference” appeared in Philosophical Transactions (vol. 160, pp. 497–517). It was the first logic machine to solve a complex problem faster than a person could in their head. Jevons saw in it “a palpable demonstration of the generality and power” of his method—and, at the same time, of the superiority of Boolean algebra over scholastic logic. Skeptics appeared quickly: F. H. Bradley, in Principles of Logic (1883), sneered that the machine merely enumerated the rows of a truth table and could not collapse them into a compact form—a complaint that now sounds like a rebuke of brute-force inference. The argument—“does the machine enumerate, or does it reason?”—was already raging then, in the Victorian press, and is reproduced exactly today around large language models: the dispute is not about what the machine can do, but about what reasoning means for a human being.
Jevons’s piano outlived its critics: it is kept at Oxford’s History of Science Museum, and in 2022 the Computer History Museum devoted a separate essay to it—“Playing AI on Ivory and Wood.” Allan Marquand of Princeton reduced the design in 1881 to a third of the height, with 10 keys and 16 indicators; Annibale Pastore in Genoa assembled a three-wheel apparatus in 1903, where a valid syllogism made all three wheels turn from a single crank—pure metaphor of inference, set upon endless belts. And, according to Gardner, the mechanical era did not end there: in 1947, two Harvard students, Burkhart and Kalin, built the first electrical propositional-calculus machine—already with switches and light bulbs, but essentially “Jevons’s piano in electricity,” sequentially scanning truth-table rows. By 1955, the specialized niche was dead: digital computers had made it unnecessary. In 1960, Hao Wang ran the first 220 theorems of Principia Mathematica on an IBM 704 in under three minutes.
And now—the Russian story that has almost fallen out of the canon. In the mid-1960s, researcher G. N. Povarov “rediscovered” two Russian logic machines built before the Revolution. The first was assembled by Pavel Dmitrievich Khrushchov—not the leader, but a physical chemist from St. Petersburg who worked under Kekulé and Berthelot, translated Mendeleev during his famous discussion of the benzene ring, and owned a laboratory on his Karasevka estate outside Kharkiv. Most likely in 1897–1898, while preparing his course “General Methods of the Physical Sciences,” he built his own logic machine—the first in Russia—from a translation of Jevons’s Principles of Science. Frustratingly little is known about it: Khrushchov himself took no pride in the device, never mentioned it in publications, and died of a stroke in 1909. The sole source with a description and photographs is the 1913 book by his friend, the chemist Aleksandr Nikolaevich Shchukarev, Problems of the Theory of Knowledge (Odesa: Mathesis).
Shchukarev is a character in his own right. A professor of general chemistry at the Kharkiv Technological Institute and Luginin’s coauthor on calorimetry, he first repaired his late friend’s machine and showed it in 1912 at a meeting of the Society of Physical and Chemical Sciences at Kharkiv University—“at the insistence of the public,” the demonstration had to be repeated almost ten times. Then he built his own improved machine: entirely of metal, with the elimination of “constructive defects, of which, it must be admitted, there were rather a great many in Jevons’s instrument” [spelled “Jivons” in the text]. The crucial innovation was a light screen: 16 horizontal shelves with two lamps each, in front of them tracing paper bearing verbal formulations of the combinations—“Silver is a metal, a conductor of electricity, and possesses free electrons.” In the zero position, all 32 lamps are lit; enter premises, the rods rise, the lamps go dark, and only what is compatible with the premises remains illuminated in the darkness. The audience saw the inference in Russian, not in the letters A/B/C. Shchukarev, unlike Jevons, was convinced that the machine was useful not just for teaching: he took it around on public lectures on the theory of knowledge.
And this is where the chronicles begin. A memoirist recounts: at one demonstration, a female student indignantly declared that the machine was “like every machine—an instrument for the exploitation of the masses”; when asked “do we have a constitution,” she answered “not we do not have not a constitution” (the machine honestly produced the square-nest logic of negations, which people read as a political comment; it came to the intervention of the police officer—public lectures were supposed to be kept orderly). In Soviet times, the professor would come once a year with the machine to the Kharkiv House of Scientists, and whenever examples approached “the possibility of the existence of the Creator of the world,” the hall of Marxists flew into a rage—while Shchukarev “laughed merrily, good-naturedly, and somehow childishly.” And there was a precedent that grows more valuable with every decade: after an anti-religious lecture, an elderly university attendant rose “for discussion” and declared: “You say that God does not exist. But Professor Shchukarev’s machine has clearly proved that God exists.” The lecturer nodded along without listening to the end—the Bolshevik cult of technology, obligatory for any Bolshevik, played a joke in which the machine was made to say what it had not said. Never has the drama of “the machine said one thing, the audience heard another” been stated more cleanly—hello to every discussion of language-model hallucinations.
There is no information about the fate of Khrushchov’s prototype after 1913; Shchukarev’s machine was listed as property of Kharkiv University and disappeared along with its era. Shchukarev published an article about his machine only in 1925, in Vestnik znaniya—12 years after its construction—under the title “Mechanization of Thought (Jevons’s Logic Machine).” By the mid-1960s, both machines had faded so completely that Povarov had to literally rediscover them from periodicals.
Against this background, the Soviet DLM looks like Shchukarev’s ideological heir, only translated into the language of children’s industrial toys: the same visible states, mechanical “registers,” logic you can touch. Curiously, the game had a Western cousin: Digi-Comp I—a polystyrene kit from the United States in 1963 for $4.99 (around $52 in 2025 prices), the “first home computer” according to some historians. It too had three mechanical flip-flops, programmable with cylindrical pegs, and clocked by a lever; officially, an “actual working digital computer,” honestly, a finite-state machine. The DLM was more honest: it made no such claim.
The scale of the phenomenon is not trivial: the DLM was produced from 1984; the serial numbers of two 1987 units described on Habr are 005957 and 9924, suggesting roughly ten thousand machines over three years, plus its predecessor (“Detskaya vychislitelnaya mashina [Children’s Computing Machine]” from the mid-’70s)—around a hundred thousand in all, by the authors’ estimate. Each unit substitutes for roughly thirty transistors; put them all together and you get a “processor” with three million transistors, against the legendary Z80’s eight and a half thousand. An absurd tally, full of meaning: one hundred thousand separate logic machines instead of a single silicon chip—an industry that could have happened, but did not, though all the while it was catching up with programming through people’s hands.
drlexium’s emulator follows the original’s limitations: if you insert a plug with the bar shifted, it warns that “you can break the machine that way”—no ignoring the instructions. The automated tests found bugs in the manual: “Zoological Cards” lies in 2 of 7 reachable scenarios; in “Twelve,” with a three on top, a six at the side gives an immediate win, while the manual tells you to place a two. Given that the DLM is incapable of anything but honest Boolean logic, every bug found is not a hardware malfunction but a human mistake in translating logic into mechanics. As always.
The dynasty’s running moral—Stanhope → Jevons → Khrushchov → Shchukarev → DLM—is this: every machine was built the same way—to physically enumerate all combinations and mechanically erase the impossible ones. None could do anything beyond that, and all of them—including the DLM with its “psychological test”—honestly did only one thing: Boolean filtering. The distinction between “reasoning” and “enumeration,” then as now, was drawn not by the machines but by the people looking at them. It is telling, too, how success was measured. Stanhope did not publish his machine at all—personal certainty was enough for him; Jevons measured speed against a human; Shchukarev by the number of scandalous lectures; the DLM by the number of kindergarten games in its accompanying booklet. None survived as a technology, but all outlived their critics as an argument: automated inference is technically trivial, while its social reception is not. The mechanical piano proved that the fate of logic depends on contacts; the police officer who stepped in at Kharkiv proved that the fate of logic depends on the police officer.
The 1987 “Children’s Logic Machine” is the finale of an evolutionary branch that began in 1775 with two panels in a brass frame. Inside it is the same architecture as the Oxford piano: a field of possibilities from which premises erase the impossible. Only Jevons used pulleys to erase them, Shchukarev used rods with lamps, and the DLM used plastic bars with springs. Three generations of different materials, the same semantics, and almost no distance between them in time from logic’s own point of view—it has not changed at all. The bugs in the Soviet booklet, found by automated tests in 2026, are a beautiful epilogue: a machine containing not a single computing device survived into an age when it began to be tested in software. Two worlds met at one point: a device from 1987, unable to calculate, is now debugged by a program that can. There is nothing in it that can break—only the person who puts the plugs into the wrong sockets breaks down.