← The canon · AItopiaOrAImageddon?

The essential singularity passage in Ulam's memorial to von Neumann

interpretation · Stanisław Ulam, summarising an undated conversation with John von Neumann · 1958

A reading of events that a reading may need to name.

Read on: "The Coming Technological Singularity: How to Survive in the Post-Human Era".

proposals.md filed this under moment and the file is filed here as interpretation. The proposal's own description is the reason: "the first known use of 'singularity' in this sense, and it is a paraphrase of a dead man." Nothing happened in May 1958. A memorial article appeared in a mathematics journal, one sentence on its fifth page was read by almost nobody for thirty-five years, and then a science-fiction writer quoted it and it became the founding text of a genre. A moment entry answers what happened; this one answers what the shape of progress was taken to mean, which is what the interpretation entries already in canon/ do — wiener-1960-automation, moravecs-paradox-1988, bitter-lesson-2019, turing-1950. The kind is changed, and this paragraph is the notice.

Two rival kinds have to be declined out loud, because declining them is most of what this entry is for.

prediction is the first, and it fails on the kind's own four fields. This canon requires a prediction to carry the claim in its own words, the date it was made, the date it was due by, and what actually happened. The sentence supplies the first, approximately; the second only as "some time before February 1957"; and the third not at all. There is no date, no threshold, no rate and no quantity. It cannot be scored, ever, by anyone. That is not a defect reported from outside — it is the single most consequential property of the text, and section 3 grades what can be graded instead: von Neumann's own signed 1955 essay, which makes the same argument with a date on it, and the one paper that ever turned the metaphor into arithmetic, whose due date is 13 November 2026 — eighty-six days from the writing of this file.

idea is the second, and it fails the test minimax-alpha-beta applied: an idea is something you can implement, because there is a loop or a formalism in it. There is no mechanism here. No recursion, no feedback, no automaton, no curve. Every mechanism the word "singularity" now carries was supplied later by other people, and this entry's main service to a reading is to be able to say exactly that, with the page number.

The canon has been waiting for this file, and says so. kurzweil-2005 records that page 23 of The Singularity Is Near credits "von Neumann's 1950s remark, via Ulam" as "the first reference to the Singularity as an event capable of rupturing the fabric of human history," and then states the gap plainly: "None of Vinge, Good, von Neumann or Ulam is in canon/." This file closes one quarter of that gap. It cannot close the rest and cannot even draw the edge: good-intelligence-explosion-1965 and vinge-singularity-1993 are proposed and unwritten, and a canon job writes exactly one file, so kurzweil-2005's descends_from line cannot be amended from here. The arrow from Kurzweil back to this entry is one the owner or a later job has to draw.

On descends_from, which is empty and should be. Nothing in canon/ precedes this honestly. dartmouth-1956 is two years earlier and is about machine intelligence, which this sentence is not about; autofac-1955 is the same year as the essay graded below and names von Neumann's self-reproducing automata, but a Dick story is not an ancestor of a remembered conversation; godel-incompleteness-1931 shares a protagonist and nothing else. The real ancestor is **von Neumann's own "Can We Survive Technology?" (Fortune, June 1955)**, which is not in this canon and which section 3 grades at length, because it is the same claim in the same man's hand with a deadline attached. If the owner wants the edge drawn, that essay is the entry to take.

1. What it is

The document

S. Ulam, "John von Neumann 1903–1957," Bulletin of the American Mathematical Society, volume 64, number 3, part 2, May 1958, pages 1–49. It is a memorial article, open access at Project Euclid, and the footnote on its first page reads "Received by the editors February 8, 1958" — one year to the day after the death it records. Ulam states his standing in the second paragraph: his aim is "to add here a brief account of his life and of his work from a background of personal acquaintance and friendship extending over a period of 25 years."

The article is a full scientific obituary: forty-odd pages walking through set theory, quantum mechanics, operator algebras, ergodic theory, continuous geometry, game theory, hydrodynamics, Los Alamos, computing machines and automata, followed by the bibliography. The first nine pages are personal recollection. The famous sentence is on page 5, in that personal section, and its neighbours matter: the paragraph before it is about von Neumann's lack of self-confidence in the face of mathematicians with "relatively simple-minded powers of intuition of new truths," and the paragraph after it — overleaf, on page 6 — is about his sense of humour.

The passage, entire

Here is the whole paragraph, which is almost never quoted whole. (Quotations in this file are set as italic paragraphs rather than as > blocks: build.py's renderer has no blockquote rule, so a > line joins the surrounding paragraph and the marker leaks onto the page as >. Several written entries carry that defect; this one does not.)

"Quite aware that the criteria of value in mathematical work are, to some extent, purely aesthetic, he once expressed an apprehension that the values put on abstract scientific achievement in our present civilization might diminish: 'The interests of humanity may change, the present curiosities in science may cease, and entirely different things may occupy the human mind in the future.' One conversation centered on the ever accelerating progress of technology and changes in the mode of human life, which gives the appearance of approaching some essential singularity in the history of the race beyond which human affairs, as we know them, could not continue." — Ulam, page 5. The inner single quotes are double quotes in the original.

Four things are visible here that are invisible in the excerpt everyone quotes.

First, the subject of the paragraph is not technology. It is von Neumann's worry that science itself might stop being valued — that the interests of humanity may change and entirely different things may occupy the human mind. The singularity sentence is the second and last sentence of a paragraph about an aesthetic and cultural apprehension. It is filed by its author under melancholy about the future of mathematics, not under forecasting.

Second, Ulam uses quotation marks one sentence earlier and does not use them here. On the same page, in the same paragraph, he shows exactly what he does when he is reproducing von Neumann's words: he quotes them. The singularity sentence carries no quotation marks. It is a summary, written by a friend, of an undated conversation, at least a year after the other party's death. The distinction is the author's own, made in the same breath, and it is decisive: we do not know that von Neumann used the word "singularity", and Ulam does not claim he did. The sentence does not even say who said what. "One conversation centered on" names no speaker.

Third, the tense is Ulam's, not von Neumann's. "which gives the appearance" — present tense, in 1958, from the writer. Whatever was said in that conversation, the framing of the sentence belongs to the man typing it.

Fourth, there is no content of the kind the sentence is now used to supply. No machine. No intelligence, artificial or otherwise. No self-improvement, no recursion, no explosion, no doubling, no date, no rate, no threshold, no mechanism by which the acceleration is produced, and no account of what lies on the far side beyond the observation that it would not be "human affairs, as we know them."

"Essential singularity" is a term of art, and Ulam knew which one

The words are not decoration. In complex analysis an isolated singularity is one of exactly three things. A removable singularity is a hole you can fill in. A pole is where the function runs off to infinity in a controlled, countable way — this is what a hyperbola does, and it is what every modern singularity curve draws. An essential singularity is neither: the Laurent expansion has infinitely many negative terms, and the behaviour near the point is not blow-up but wildness. The Great Picard theorem says that in any neighbourhood of an essential singularity the function takes every complex value, with at most one exception, infinitely often. The standard example is e^(1/z) at zero.

So the technical content of the metaphor, read strictly, is not "the curve goes to infinity." It is "arbitrarily close to this point, anything at all can happen, and no extrapolation survives." That is unpredictability, not explosion — which is Vinge's 1993 reading of the word and the exact opposite of the popular one, in which the singularity is a vertical asymptote with a year printed under it.

Whether Ulam meant the term strictly cannot be established; "in the history of the race" is plainly figurative, and a mathematician may reach for a technical word loosely. But two facts weigh. He was a professional mathematician of the first rank writing in the Bulletin of the AMS for an audience of mathematicians, none of whom would read "essential singularity" as a synonym for "very big." And one page later he describes von Neumann's method of historical forecasting as "a sort of analytic continuation" — the sentence is "This encyclopedic knowledge molded his views on the course of future events by inducing a sort of analytic continuation."

Analytic continuation is the operation that fails at a singularity. Ulam is running a single sustained complex-analysis metaphor across pages 5 and 6: you extend the historical function outward from what you know, and there is a point past which the extension cannot be carried. Read together, the two sentences are consistent, deliberate and mathematically literate. Read apart — which is how the first one is always read — the second disappears and the first becomes a prophecy.

The rest of page 6, which nobody quotes

The analytic-continuation line is the opening of the passage in which Ulam grades von Neumann's forecasting record, one page after the sentence the forecasting industry was built on:

"I can testify that in his forecasts of political events leading to the Second World War and of military events during the war, most of his guesses were amazingly correct. After the end of the Second World War, however, his apprehensions of an almost immediate subsequent calamity, which he considered as extremely likely, proved fortunately wrong. There was perhaps an inclination to take a too exclusively rational point of view about the cases of historical events. This tendency was possibly due to an over-formalized game theory approach."

That is a hit rate and a failure mode, from an intimate of twenty-five years, in the same document, about the same man, concerning claims of exactly the same type as the one on page 5: he was excellent on the near-term mechanics of a war already under way, and he was wrong about imminent catastrophe, systematically, because he over-applied rational-actor reasoning to history. This canon exists to grade forecasters. The best-sourced grade of von Neumann as a forecaster sits eleven lines below the sentence that made him the patron saint of forecasting, and it is negative on precisely the claim type in question.

What the same document says von Neumann actually thought was coming

Page 39, in Ulam's own summary of his friend's convictions:

"Von Neumann strongly believed that the technological revolution initiated by the release of nuclear energy would cause more profound changes in human society, in particular in the development of science, than any technological discovery made in the previous history of the race. In one of the very few instances of talking about his own lucky guesses, he told me that, as a very young man, he believed that nuclear energy would be made available and change the order of human activities during his lifetime."

Note the phrase "in the previous history of the race" — the same construction as page 5's "in the history of the race," used here explicitly of nuclear energy. Thirty-four pages apart, in one man's summary of another's views, the transformative technology is named, and it is not computing.

Ulam's treatment of the machines is measured and appears on page 35 under "Theory of automata, probabilistic logic," which he says "seems to me to rank with his most fertile ideas." His forecast for that programme: "Such studies now promise more conquests of mathematics in its ability to formalize, perhaps at first on an extremely simplified level, the workings of an organism and of the nervous system itself."

Formalising the nervous system on an extremely simplified level. That is what the friend and colleague of the man in question thought automata theory was going to do, written in the same year, in the same article, as the sentence now read as the prophecy of superintelligence.

Page 38 has the other machine anecdote, and it is the best compute-estimation story in the document. Of the first thermonuclear calculations, von Neumann said to Ulam: "Probably in its execution we shall have to perform more elementary arithmetical steps than the total in all the computations performed by the human race heretofore." Ulam's next sentence: "We noticed, however, that the total number of multiplications made by the school children of the world in the course of a few years sensibly exceeded that of our problem!" A superlative compute claim by the most capable estimator of the century, corrected within a paragraph by counting what ordinary people were already doing. The pattern recurs; readings in this project meet it every week.

The signed version, with a deadline: Fortune, June 1955

Von Neumann did write down the accelerating-crisis argument himself, under his own name, eighteen months before he died, in Fortune — "Can We Survive Technology?", signed "Member, Atomic Energy Commission." It opens: "'The great globe itself' is in a rapidly maturing crisis — a crisis attributable to the fact that the environment in which technological progress must occur has become both undersized and underorganized."

The mechanism is geographic, not computational. Industrial progress was absorbable for two centuries because there was always more world to expand into; that safety factor is gone. "Now this safety mechanism is being sharply inhibited; literally and figuratively, we are running out of room. At long last, we begin to feel the effects of the finite, actual size of the earth in a critical way."

And then the sentence that is the singularity remark with a date welded on: "In the years between now and 1980 the crisis will probably develop far beyond all earlier patterns. When or how it will end — or to what state of affairs it will yield — nobody can say."

"Nobody can say" is the same gesture as "human affairs, as we know them, could not continue." The difference is that this one says by 1980, in print, over a signature — and can therefore be graded, which section 3 does.

The essay's accelerants are listed and machine intelligence is not among them: nuclear energy ("a few decades hence energy may be free — just like the unmetered air"), automation, communications, and deliberate climate control. Automation gets two pages and is treated soberly — computers will be used "for elaborate industrial process control, logistical, economic, and other planning, and many other purposes heretofore lying entirely outside the compass of quantitative and automatic control and preplanning" — with no suggestion anywhere that the machines will exceed their operators. And the conclusions are the reverse of transcendence:

"For progress there is no cure. Any attempt to find automatically safe channels for the present explosive variety of progress must lead to frustration. The only safety possible is relative, and it lies in an intelligent exercise of day-to-day judgment."

"We can specify only the human qualities required: patience, flexibility, intelligence."

The provenance chain, end to end

An undated conversation, before February 1957 → remembered by a friend → written down a year after the death, unquoted, as a summary → published in a mathematics journal in May 1958 → nothing, for thirty-five years, as far as could be established here → quoted in full by Vernor Vinge in "The Coming Technological Singularity," delivered at the VISION-21 symposium (NASA Lewis Research Center and the Ohio Aerospace Institute, 30–31 March 1993) → promoted by Kurzweil in 2005 to "the first reference to the Singularity as an event capable of rupturing the fabric of human history."

Two details about the hand-off deserve recording. Vinge's reference list gives the article as "Ulam, S., Tribute to John von Neumann" — which is not its title; the article is called "John von Neumann 1903–1957." And Vinge, having quoted it, immediately tells you it does not say what it is now used to say: "Von Neumann even uses the term singularity, though it appears he is still thinking of normal progress, not the creation of superhuman intellect."

The man who put the word into circulation attributed it to von Neumann rather than Ulam, cited it under the wrong title, and stated in the same breath that it was not about superhuman intelligence. Everything downstream inherited the first two errors and dropped the correction.

2. Why a reading would cite it

The occasion is specific and recurs: when "the singularity," "takeoff," or "the event horizon" is used as though it named a technical claim.

The live instance is not hypothetical. In June 2025 Sam Altman published "The Gentle Singularity," which opens "We are past the event horizon; the takeoff has started," describes "a larval version of recursive self-improvement," says "intelligence too cheap to meter is well within grasp," and assigns years: agents doing real cognitive work in 2025, systems that "figure out novel insights" in 2026, robots doing real-world tasks in 2027. The last three are gradeable and should be graded when due. The first is not a claim; it is the word this entry is about, deployed in the present perfect.

What this file lets a reading do, in one sentence, is separate those two things without either sneering or capitulating: the dated capability claims can be scored, and the word cannot, because at its root there is one sentence of remembered conversation with no argument in it. A reading that wants to contest an intelligence-explosion claim has to go to Good's 1965 recursion argument, or Vinge's unpredictability argument, or the scaling literature (scaling-laws-2020, bitter-lesson-2019), each of which asserts something checkable. Citing the 1958 passage is citing a pedigree, not a premise. That distinction is worth a canon entry because the substitution happens constantly and is invisible unless someone has read page 5.

The second occasion is when acceleration is offered as its own evidence — the "look at the curve" move. Here the entry supplies the discipline from the one participant entitled to make it, in writing, in 1955: "All experience shows that even smaller technological changes than those now in the cards profoundly transform political and social relationships. Experience also shows that these transformations are not a priori predictable and that most contemporary 'first guesses' concerning them are wrong."

That is the father of the singularity remark saying that first guesses about transformation are usually wrong, in the same essay where he makes one. A reading can hold both halves and should.

The third occasion is the containment argument, which this project meets in almost every reading — voluntary lab pauses, compute thresholds, treaty talk, open weights. The 1955 essay contains the strongest short statement of the case against prohibition ever made by someone who wanted the danger contained:

"the parts of technology, as well as of the underlying sciences, are so intertwined that in the long run nothing less than a total elimination of all technological progress would suffice for inhibition … useful and harmful techniques lie everywhere so close together that it is never possible to separate the lions from the lambs."

"Hence the banning of particular technologies would have to be enforced on a worldwide basis. But the only authority that could do this effectively would have to be of such scope and perfection as to signal the resolution of international problems rather than the discovery of a means to resolve them."

That last sentence is seventy-one years old and it is the objection every proposal for a global pause still has to answer: the enforcement capacity a ban needs is the same capacity whose absence made the ban necessary. It sits alongside asilomar-1975 (the case where a moratorium held, in a field with different physics of enforcement), bletchley-declaration-2023, llama-weights-2023 and tcp-ip-1983. A reading weighing a self-imposed frontier pause against a rule has a 1955 statement of the trade-off to cite, and it costs the citing reading nothing, because it is not evidence.

What this entry may not be used for: it is not evidence, it deposits nothing, it moves nothing. Nobody's argument becomes stronger because von Neumann is upstream of it — see section 4, misuse 6, which cuts the same way against the sceptics.

3. What it got right, and what it got wrong

interpretation does not require this section. It gets one anyway, at length, because the whole value of the entry is in what happens when you try to grade the thing — and because grading the ungradeable claim honestly means grading its two dated relatives instead.

The 1958 sentence itself

What it got right, weakly: "the ever accelerating progress of technology and changes in the mode of human life." Between 1958 and 2026 that is true on any number of measures a person cares to pick, and the change in the mode of human life is not in dispute. But "accelerating" without a metric is not a claim that can fail, and a claim that cannot fail is not a claim that can succeed. Grading it as a hit would be the same error as grading it as a miss.

What it got wrong, so far: "beyond which human affairs, as we know them, could not continue." Sixty-eight years on, human affairs as we know them have continued. There has been no discontinuity of the kind the sentence gestures at — not in 1980, not in 2000, not in 2026 — and the mode of human life has changed enormously without one. The honest verdict is not "wrong" but "still unfalsifiable, and every falsifiable version of it has failed." The sentence survives because it never took a risk.

What it got right in a way that outlives both: the epistemic content. If the essential-singularity reading in section 1 is the right one, the claim is that extrapolation fails past a certain point — and this canon's shelf of failed extrapolations (simon-1965, minsky-1970, lighthill-1973, moravecs-paradox-1988, musk-robotaxi-2019, half of kurzweil-2005) is a sustained demonstration that it does.

The 1955 essay, graded properly, item by item

This is the version with a signature and a deadline, and it is a mixed record of the interesting kind.

Right, and unnervingly so — the carbon dioxide. In 1955: "The carbon dioxide released into the atmosphere by industry's burning of coal and oil — more than half of it during the last generation — may have changed the atmosphere's composition sufficiently to account for a general warming of the world by about one degree Fahrenheit." Mechanism identified, source attributed, magnitude given, three years before Keeling started measuring at Mauna Loa. The hedged "may have" is exactly the right amount of hedge for 1955.

Wrong, in an instructive direction — climate control. Having got the diagnosis right, he predicted the response would be deliberate: "Probably intervention in atmospheric and climatic matters will come in a few decades, and will unfold on a scale difficult to imagine at present." Due by roughly 1985. Seventy-one years on, deliberate large-scale climate intervention remains unattempted, while the accidental intervention he described in the previous paragraph ran to completion. He predicted the powerful version and missed the one that was already happening. For a project that watches capability claims against deployment, that is the shape to remember.

Right — automation. Computers for "elaborate industrial process control, logistical, economic, and other planning": shipped, comprehensively, and the machines-per-machine numbers he quotes (10,000 vacuum tubes, 100,000 cores, "only a few errors per day or week") make the direction of travel plain.

Wrong — free energy. "A few decades hence energy may be free — just like the unmetered air." Due by roughly 1985 on his own phrasing. No.

Wrong, and it is the same error Ulam diagnosed — the 1980 terminal crisis. "In the years between now and 1980 the crisis will probably develop far beyond all earlier patterns." The Cold War ran to 1991 and ended without the civilisation-scale rupture the essay anticipates; the specific thing he feared most, general nuclear war, has not happened in the seventy-one years since. This is precisely the "apprehensions of an almost immediate subsequent calamity … proved fortunately wrong" that Ulam names on page 6, recurring in the same man seven years later. A forecaster's second miss of the same type is not bad luck, it is a bias, and Ulam had already named the mechanism.

The most valuable thing in it, and it grades the author: "these transformations are not a priori predictable and that most contemporary 'first guesses' concerning them are wrong."

The one numerically precise descendant — and it is due in eighty-six days

The metaphor was made into arithmetic exactly once, and the arithmetic named a date that falls inside this project's operating life.

**Heinz von Foerster, Patricia M. Mora and Lawrence W. Amiot, "Doomsday: Friday, 13 November, A.D. 2026," Science, New Series, vol. 132, no. 3436, 4 November 1960, pp. 1291–1295.** The subtitle is the claim: "At this date human population will approach infinity if it grows as it has grown in the last two millenia" (the misspelling is the original's). The authors were on the staff of the department of electrical engineering at the University of Illinois, Urbana.

They fitted world population back through about a hundred generations and got t₀ = A.D. 2026.87 ± 5.50 years, K = (1.79 ± 0.14) × 10¹¹, and k = 0.990 ± 0.009, giving N = 1.79 × 10¹¹ / τ^0.99 where τ = t₀ − t, with a root-mean-square deviation across all points of about 7 percent. 2026.87 is 13 November 2026, which is indeed a Friday. Their stated confidence: "Today, however, we are in a much better position, since we are required to extrapolate over only 4 percent of our evidence: we can predict doomsday within approximately 10 years."

Is this the same idea? Convergent rather than descended, and the distinction should be kept clean. Their reference list contains no Ulam and no von Neumann, and no evidence was found either way that they knew the memorial. But the milieu is not merely adjacent: von Foerster edited the transactions of the Macy conferences on cybernetics, the meetings at which von Neumann was a core participant. The same room produced the same metaphor twice, independently, two years apart, and only one of the two versions was willing to write down a number.

The grade, and it is not close. Working from their own published constants, their curve against the actual world population:

The curve ran slightly below the real population until about 1993, crossed it, and then left. Today it is high by a factor of roughly ninety. The mechanism failed on schedule and in public: the global population growth rate peaked at about 2.3% in 1963 — three years after publication — and has fallen monotonically since, to about 0.84% in 2026. The UN's 2024 World Population Prospects puts the world at 8.2 billion in 2024 and projects a peak of about 10.3 billion in the mid-2080s, followed by decline. Not infinity in 2026; a maximum sixty years later. The hyperbolic era is over as a matter of arithmetic rather than opinion: Victor Yakovenko's re-fit ("The End of Hyperbolic Growth in Human Population and CO₂ Emissions," Physica A 661, 130412, 2025) confirms a hyperbolic regime from about 1700 with an extrapolated singularity at t_s = 2030, then finds that the twenty-first-century data follow an "avoided crossing" — "instead of a singularity, we predict a peak in human population at t_s = 2030 of the time width τ = 32 years." That peak date is much earlier than the UN's mid-2080s and the two disagree; this file does not adjudicate between them. Both agree there is a peak and not a pole, which is the part that matters here.

The complication, which is what makes this worth a page rather than a sneer. The authors were not prophesying. Their own conclusion is that "our great-great-grandchildren will not starve to death. They will be squeezed to death," their closing section is titled "Population Servo," and its argument is that fertility must be brought under deliberate control — a "'peoplo-stat' so to speak" — by "cutting the birth rate to about half its present value or, in other words, cutting the size of an average family to just a little above two children."

Global fertility then fell by roughly half, and family size fell to a little above two children, and in over half of all countries below it. So: failed forecast, or successful warning? Both readings are available and they are not symmetric. As a forecast it is a catastrophic miss, off by ninety-fold on a sixty-six-year horizon, and it must be scored as one. As a warning it named the exact variable that mattered and the exact remedy that arrived — though the remedy arrived overwhelmingly through development, urbanisation, income and women's education rather than through the servo mechanism they imagined, so even the warning was right about the destination and wrong about the road. A reading tempted to say "the doomsday equation was cancelled by the policy it inspired" is making a causal claim the demography does not support.

What all of this is a base rate for

The prediction kind exists in this canon to build a base rate for how wrong AI forecasting runs. This entry contributes something more specific and more uncomfortable: the founding text of the most-invoked framework in the field is the one text in it that cannot be scored, and every descendant of it that could be scored has missed. Von Foerster's date arrives in November and is wrong by two orders of magnitude. Von Neumann's own 1980 is thirty-six years past and was wrong in the direction Ulam predicted he would be wrong in. Vinge's thirty years ran out in 2023 (vinge-singularity-1993, proposed, unwritten — it is the gradeable one and someone should write it). Kurzweil's 2045 is still running, with the record in kurzweil-2005.

The regularity is not "singularity people are wrong." It is that the unfalsifiable version of a claim outlives all of its falsifiable versions, and inherits their prestige without ever having taken their risk.

4. Commonly misused as

limit entries owe this section; this one is interpretation and owes it anyway, because misuse is the entire reason the file is worth writing. The standing house example is godel-incompleteness-1931, deployed to prove minds are not machines, which is not what the theorems establish. This is the same shape with the additional twist that the source text does not establish anything at all.

1. As von Neumann's words. It is Ulam's summary. The evidence is on the same page, in the form of quotation marks used one sentence earlier and withheld here. Formulations like "von Neumann coined the term" and "von Neumann first used 'singularity' in the technological sense" are everywhere, including in otherwise careful reference works, and they cannot be supported from the source. The defensible statement is: the earliest known use of "singularity" in this sense appears in Ulam's 1958 memorial, in a sentence summarising a conversation with von Neumann; whether the word was von Neumann's is unknown.

2. As a statement about artificial intelligence. There is no machine in it. Against the AI reading stand: page 39 of the same article, where the transformative technology is nuclear energy in the same phrase ("in the previous history of the race"); page 35, where the automata programme is forecast to formalise the nervous system "on an extremely simplified level"; the whole 1955 Fortune essay, whose accelerants are nuclear energy, automation, communication and climate control, with machine intelligence absent; and Vinge himself in 1993, saying the man "is still thinking of normal progress, not the creation of superhuman intellect." The reading that von Neumann's crisis was the nuclear one, in the specific historical circumstances of 1955 — a man on the Atomic Energy Commission and the ICBM committee, who thought general war was likely soon — is the reading his own signed prose supports.

3. As a prediction of a blow-up to infinity. The phrase is "essential singularity," which in the field of the man who wrote it and the journal it appeared in specifically means not a pole. Modern usage draws a vertical asymptote and dates it; the term chosen denotes the class of singularity where the function does not tend to anything, and where, by Picard, every value occurs infinitely often in any neighbourhood. Whether Ulam meant it strictly is unknowable; that the popular reading is the one thing the words do not say is checkable. It is a small irony that the only rigorous version of the idea anyone built — von Foerster's — is a pole, and it is the version that has been decisively falsified.

4. As an argument. Nothing follows from it, because there is nothing in it to reason from. The actual argument for an intelligence explosion is I. J. Good's 1965 recursion — a machine better than humans at designing machines designs a better machine, and so on — and the serious literature knows this. David Chalmers's "The Singularity: A Philosophical Analysis" (Journal of Consciousness Studies 17:7–65, 2010) opens by attributing "the basic argument" to Good and quoting him at length, and does not mention Ulam or von Neumann in its opening pages at all. When a text needs the argument, it cites Good. When it needs a pedigree, it cites 1958.

5. Misquoted, routinely, in the direction of the modern reading. The English Wikipedia article on the technological singularity, checked on 19 August 2026, renders the passage as "centered on the accelerating progress of technology and changes in human life" — dropping "ever" and dropping "the mode of." Both excisions push the sentence toward being about technology's rate rather than about how people live. (This check was made through a fetch tool that re-renders the page, so treat the exact characters as one remove from the source, though the omission is unambiguous in what came back.) Vinge's reference list, meanwhile, calls the article "Tribute to John von Neumann." The founding quotation of the field is habitually cited under a title its author did not give it, in words its author did not write.

6. And the misuse in the other direction, which this entry must not encourage. Showing that the 1958 sentence is a thin reed does not refute the singularity hypothesis. That would be a genetic fallacy, and this canon has to be as strict about it as about the credulous version. The arguments of Good, Vinge, Chalmers and superintelligence-2014 stand or fall on their own premises, and the serious critiques attack those premises rather than the pedigree — David Thorstad's "Against the singularity hypothesis" (Philosophical Studies 182, 1627–1651, 2024) argues that the hypothesis "rests on undersupported growth assumptions," which is an argument about growth rates, not about who said what to whom in 1955. A reading that uses this entry to dismiss a capability claim has misused it exactly as badly as one that uses it to bless one. The correct use is narrow: the word carries no evidence, so make the speaker supply some.

5. Sources

Read at source, in full or in the pages named:

Taken from summaries or secondary sources, and flagged as such:

Computed here, not sourced: the values of the von Foerster curve in the list above and the ≈750 billion figure for today were calculated from the paper's own published constants (K = 1.79 × 10¹¹, k = 0.990, t₀ = 2026.87), by hand, since this job could not run an interpreter. τ for 19 August 2026 is 86 days ≈ 0.2354 years. Anyone re-deriving them will get the same numbers; the arithmetic is one logarithm.

Failed fetches, recorded because a reader should know what was not checked: the AMS PDF of the Ulam memorial (403), the AMS article landing page (403), and the Marquette retrospective "The Doomsday Equation and 50 Years Beyond" (403). The Science text at science.org was not attempted; the JSTOR-sourced scan was used instead.

Project context read while writing: README.md, CLAUDE.md, LENSES.md, canon/proposals.md, the header blocks of the written entries, the descent section of canon/kurzweil-2005.md, the openings of canon/minimax-alpha-beta.md and canon/simon-1965.md for the kind-declining form, build.py's front-matter and Markdown functions (which is where the blockquote note in section 1 comes from), and digests/2026-08-19-00.md for the state of the current reading. Nothing in the digest is relied on above.

No evidence was deposited, no needle, score or landmark is placed, and no file outside canon/von-neumann-singularity-1958.md was written.