The Electronic Brain, Revisited

In the 1950s, newspapers called the first computers "giant brains" and worried, in almost equal measure, that they were miracles or monsters. In the 2020s, we are having the same argument about AI, often in nearly the same words. This essay traces the first computer revolution — from the ledger book to the spreadsheet to the smartphone — to see what it actually did to jobs, hours, and incomes, and to ask what that history can and cannot tell us about what comes next.

I. A brewery in Rheinfelden

Sometime in the 1960s, my father, then a young accountant, went to work each day at Feldschlösschen, Switzerland's largest brewery, in the small town of Rheinfelden on the Aargau bank of the Rhine. The company was in the middle of a genuine boom — a new brewhouse in 1959, a fourth brewing line in 1965, bringing the plant up to twelve copper vessels — and its books reflected a business of real complexity: raw-material costing for imported malt and hops, per-batch production accounting tied to the brewing schedule, a running reporting relationship with the Swiss Federal Alcohol Administration over excise duty, and — a detail peculiar to breweries — a whole shadow ledger tracking the kegs and returnable bottles that circulated, as valuable and monitored as any other asset, between the brewery and thousands of restaurants and taverns.

The tools on his desk sat at an odd hinge point in office history. Double-entry bookkeeping was still substantially a hand-posted affair, in bound ledgers, but a firm of Feldschlösschen's size would also have had mechanical or electromechanical bookkeeping machines — the Buchhaltungsmaschinen built by NCR, Burroughs, or Olivetti — and perhaps an early punch-card tabulator for payroll. Full computerization of Swiss corporate accounting was still a decade away. It is a small, human window onto a much larger story: this was the last generation of accountants who did the job essentially as it had been done for a century, right before the machines arrived that would redefine what "doing the job" even meant.

That transition — and the sixty-odd years of anxiety, adaptation, and argument that followed it — is the subject of this essay, because it is currently being re-run, nearly beat for beat, under the banner of artificial intelligence.

II. Four generations of machine

The technology moved through fairly distinct phases. Herman Hollerith's punch-card tabulators (the 1890 census) and, later, electromechanical relay machines like the Zuse Z3 and the Harvard Mark I carried computing into the 1940s. True electronics arrived with wartime codebreaking machines like Colossus (1943) and ENIAC (1945) — seventeen thousand vacuum tubes filling a room — alongside the equally important stored-program concept, realized by EDVAC, EDSAC, and the Manchester Baby by 1948–49. Tubes were fast but fragile and hot, and the transistor — invented at Bell Labs in 1947 — began displacing them commercially around 1953–56, in machines like the IBM 7090, making computers reliable and affordable enough to leave military and scientific labs for ordinary corporations and universities. The integrated circuit, developed independently by Jack Kilby and Robert Noyce in 1958–59, gave the 1960s the IBM System/360 (1964) and the first minicomputers, like the DEC PDP-8 (1965) — machines shrinking from room-sized to desk-sized. Finally, in 1971, Intel put an entire central processing unit on a single chip, the 4004, and Moore's Law began setting the tempo for everything that followed: the Altair 8800, the Apple II, and the whole personal-computer wave of the mid-to-late 1970s.

These phases overlapped rather than cutting cleanly — mainframes stayed tube-based for years after transistors existed commercially, and discrete-transistor minicomputers persisted well into the IC era — but the broad arc is clear enough: computing moved, generation by generation, from something only a government could own to something that sat on an individual desk. That migration from institution to individual turns out to matter enormously for how people felt about it, which is where the story gets interesting.

III. The electronic brain arrives

Long before anyone lost a job to a computer, they lost their peace of mind to one — or, more precisely, to the language used to describe it. Newspapers covering ENIAC and its successors did not call them computers so much as brains. The press dubbed them the "Army's New Wonder Brain" and "Magic Brain"; the New York Times called ENIAC an "electronic speed marvel" that would "revolutionise modern engineering"; the Washington News went with "electronic super-brain"; the Boston Herald offered "Mechanical Einstein"; and the Los Alamos Times, tipping into outright menace, called it "a mathematical Frankenstein." That range, from marvel to monster, captures the era's real ambivalence: nobody was quite sure whether the country had built a genius or a creature, and the coverage borrowed the same rhetorical instinct then being applied to the atomic bomb — a thing simultaneously salvational and apocalyptic.

The hinge moment for public opinion arrived on election night, 1952. CBS let UNIVAC attempt to call the Eisenhower–Stevenson race live on air; after analyzing a sliver of the returns, it predicted an Eisenhower landslide, flatly against the consensus of every professional pollster, who expected a close Stevenson win. The network's own engineers assumed the machine had malfunctioned and sat on the result for two hours before broadcasting it. UNIVAC turned out to be accurate within about four electoral votes — the moment public sentiment tipped from "is this thing dangerous" to the stranger, more unsettling "this thing might be smarter than our own experts." CBS's presenter, Charles Collingwood, deliberately chatted with the machine on air and joked about its "impolite" glitches — a conscious strategy, as later analysis of the broadcast described it, to humanize UNIVAC because the audience "wasn't yet ready for overly rich doses of high technology" presented as cold and alien.

Underneath the spectacle sat a very specific, very concrete fear: that this machine was coming for people's jobs. The columnist Hal Boyle warned that computers were "gradually making man unnecessary in the world by taking over his functions," and the anxiety reached even into the home, with automation cast as a threat to the housewife's domestic role. Pop culture worked more as a pressure valve for this fear than a cause of it — the 1955 Broadway comedy Desk Set, later the Katharine Hepburn/Spencer Tracy film, put a research department's dread of replacement center stage, then resolved it by having the computer outsmarted by a woman with a hairpin. Cartoons of the era followed the same script, showing UNIVAC-like machines failing comically, letting audiences laugh at the thing they were also afraid of.

What made accounting and clerical departments specifically anxious, rather than society in general, was a very particular institutional ritual: before a company installed a computer, it almost always commissioned a feasibility study, in which efficiency consultants documented, task by task, exactly what every clerk and bookkeeper did all day. Workers understood, correctly, that having your job function fully inventoried like this was usually a prelude to something. Management routinely promised no layoffs would follow. As the historian Charles Yood has written of the period, "all the signs suggested otherwise" — so the reassurances rarely landed.

IV. What the machines actually did

Here is where the history becomes genuinely useful, because the fear and the outcome, measured directly against each other, tell a consistent and slightly surprising story: dread ran far ahead of destruction, in both of the computer revolution's major waves.

The best-documented single case is the U.S. Internal Revenue Service's conversion to automatic data processing in Atlanta between 1960 and 1962 — well-documented because the Bureau of Labor Statistics tracked it as it happened. At the start, 1,045 employees worked in the affected units. Over the following two years, 315 transferred to other positions (many moving into enforcement roles rather than clerical ones), 128 left through ordinary attrition — 74 quit, 51 retired, 3 died — and 241 completed after-hours classroom courses in accounting and tax law specifically to qualify for the new roles opening up elsewhere in the agency. Total headcount in the unit fell by 22 percent. The number of involuntary separations was zero. A broader BLS survey of twenty private companies that installed computers in the same period found the same pattern at industry scale: just nine layoffs among 2,800 affected employees, with most of the reduction happening quietly through attrition rather than firing — falling, notably, disproportionately on women in clerical roles, who were simply not rehired rather than formally let go.

A quarter-century later, the second wave arrived, and it hit a narrower, more specific target: the physical labor of recalculation. Before 1979, revising a budget or repricing an order meant an accountant or bookkeeper — as one who lived through the change, Allen Sneider, put it — taking out "that large eraser and start erasing... every column had to be changed," a full day's manual rework for a single set of revised figures. Dan Bricklin and Bob Frankston's VisiCalc, released that year, could redo the same recalculation in seconds; Lotus 1-2-3 and later Excel extended the same trick across the entire office. This time the fear was at least partly justified: bookkeeping and accounting-clerk positions really did decline, by roughly 400,000 jobs, after 1980. But accountant positions overall grew by about 600,000 in the same period, because once recalculation became nearly free, clients started asking far more "what-if" questions than they ever could have afforded to ask a human to answer by hand — a repricing exercise, a staffing scenario, a financing comparison — and the profession's center of gravity shifted from arithmetic accuracy toward judgment and advisory work. One 1984 magazine anecdote captures the transitional awkwardness nicely: an accountant who could now finish a rush job in hours rather than days deliberately sat on the finished work for two days before sending it, so the turnaround wouldn't look suspiciously fast for what he was billing. Bricklin himself put the general lesson simply: "Technology is a double-edged sword... Some will cause people to lose their job. Some makes new jobs possible."

Across both waves, then, the pattern holds: the fear of job loss was genuine and, in the mainframe era especially, actively cultivated by the "electronic brain" narrative and by feasibility studies that made every worker feel individually surveilled. But the actual outcome, each time, was less "the job disappears" than "the mechanical core of the job disappears, and what's left of it moves toward interpretation and judgment" — a shift upward in the value chain rather than a vanishing act.

V. The mood swings

Public sentiment toward computing did not simply improve as the fear proved overblown; it swung, decade by decade, in a genuinely oscillating pattern that says as much about the surrounding culture as about the machines themselves.

The 1960s opened as perhaps the single most confidently optimistic decade about technology of the entire century. The 1964 New York World's Fair drew some fifty million visitors through General Motors' Futurama, a gleaming vision of automated cities, and IBM's Charles Eames–designed pavilion deliberately showed computers helping with recipes and football strategy — a calculated effort to make the "electronic brain" feel domestic rather than alien. That confidence crested with Apollo 11 in 1969, though even that triumph was immediately, publicly contested: Ralph Abernathy led a civil-rights protest at the launch site itself, arguing the money belonged on poverty at home, and public enthusiasm for the space program cooled within a year, culminating in Nixon canceling later missions by 1970. In the same six-month window that produced Apollo 11, the Boeing 747 (February 1969) and the Concorde (March 1969) both made their maiden flights — arguably the highest concentration of "the future has arrived" feeling in modern history. But the American supersonic transport program, Boeing's government-backed 2707, collapsed within two years of that peak, killed by a coalition of environmental protest over sonic booms and ozone risk and fiscal skepticism in Congress — a near-perfect microcosm of the pivot from '60s optimism to '70s reckoning, playing out in real time, within the same industry, in the same few years.

The 1970s absorbed that reckoning fully. Alvin Toffler's Future Shock (1970) sold over six million copies arguing that the real danger wasn't any single technology but the sheer pace of change itself; the oil shocks of 1973 and 1979, alongside the Limits to Growth report and E.F. Schumacher's Small Is Beautiful, pushed public sentiment toward distrust of large-scale industrial technology generally. The mood reversed hard in the early 1980s around a specific, newly personal machine: Time named "The Computer" its 1982 Machine of the Year, reporting that nearly 80 percent of Americans expected computers to become as common as televisions by decade's end, and 67 percent believed the "computer revolution" would raise living standards — a giddiness Time's own publisher likened to "the enduring American love affair with the automobile." The 1990s pushed that optimism further still, with early Wired magazine treating the internet as a historical inevitability on par with "the discovery of fire," a mood that survived even the 2000–2001 dot-com crash largely intact. It broke, finally, over the 2010s "techlash" — platform centralization, the 2016 election's disinformation crisis, and China's social-credit system all demonstrating that networked technology concentrated power at least as readily as it dispersed it.

The AI moment, which by most reckonings began around 2022–23, loops the cycle back to where it started: a genuine split between fascination and dread, playing out in almost the same rhetorical register as the 1950s "electronic brain" coverage, with "AGI" and "superintelligence" standing in for "giant brain" and "Frankenstein." It is a striking historical rhyme — awe and anxiety about a machine's apparent intelligence, giving way eventually to confident optimism once the technology becomes personal and controllable, giving way in turn to disillusionment once its concentration of power becomes visible. History suggests the cycle simply starts again with the next wave.

VI. The economy underneath

It is worth remembering how much of this sentiment was riding on a specific, and specifically temporary, economic foundation. The 1950s and 1960s were a genuine golden age: real U.S. GDP growth averaged around 4.2 percent in the fifties and 4.5 percent in the sixties, with inflation tame at roughly 2.4–2.5 percent and unemployment near 5 percent by the mid-sixties. Against that backdrop, "we will keep getting richer and working less" felt like a reasonable straight-line extrapolation rather than a leap of faith.

The 1970s broke the extrapolation on real economic grounds that had little to do with computers. Two oil shocks — the 1973–74 Arab embargo and the 1979 Iranian revolution — collided with inflationary momentum already building from Vietnam-era spending and Great Society programs, producing true stagflation: rising prices and rising unemployment together, a combination the standard economic theory of the day said should not happen. Inflation, roughly 1 percent a decade earlier, exceeded 12 percent by the mid-seventies and peaked above 14 percent in 1980. Paul Volcker's Federal Reserve broke that inflation deliberately, pushing the federal funds rate toward 20 percent and triggering the 1981–82 recession — the worst downturn since the Great Depression at that point, with unemployment near 11 percent — before a genuine recovery took hold through the mid-to-late 1980s.

It is in the middle of that recovery, in 1987, that the economist Robert Solow made an observation that has outlived almost everything else written about the decade: "You can see the computer age everywhere but in the productivity statistics." Despite the entire personal-computer and spreadsheet revolution, and genuine transformation at the level of the individual desk, aggregate U.S. productivity growth simply did not show it. The gap — since known as the productivity paradox — did not close until the mid-to-late 1990s, roughly fifteen years after the technology that supposedly caused it first arrived.

VII. The vanishing leisure dividend

Every serious mid-century forecast of the computer's effect on work rested on one shared, largely unexamined assumption: that rising productivity would be broadly shared, arriving for the ordinary worker as a choice between more income and more free time. John Maynard Keynes set the template in 1930, projecting that compounding productivity growth would eventually deliver a "fifteen-hour week"; Bertrand Russell made a parallel argument about learning to use leisure well. By the mid-1960s, computers made the question feel urgent. In 1964, a memorandum known as the Triple Revolution — signed by figures including Linus Pauling, Gunnar Myrdal, and Bayard Rustin, and later cited by Martin Luther King Jr. in his final sermon — warned that "cybernation" was building "a system of almost unlimited productive capacity" and called for breaking the link between income and employment altogether through a guaranteed income. The U.S. government took the concern seriously enough to convene the National Commission on Technology, Automation, and Economic Progress, which reported to President Johnson in 1966 with a more measured version of the same worry, summarized in its own phrase: "technology eliminates jobs, not work." Herman Kahn and Anthony Wiener's widely read 1967 book The Year 2000 put numbers on the leisure side of the bet, forecasting a four-day, 7.5-hour-a-day workweek by century's end; the political scientist Sebastian de Grazia, writing the same year, went further, projecting a 21-to-31-hour week by 2000 and as little as 16 hours by 2020. At ETH Zürich, the work psychologist Eberhard Ulich made the same prediction from a different, more explicitly normative angle in 1984, arguing that "redistribution of work through substantial reduction of working hours" was the correct — though not automatic — societal response to computerization.

None of it arrived. Later assessment of Kahn and Wiener's roughly hundred named forecasts found only about 45 percent panned out, despite the authors having claimed 90–95 percent confidence across the board, with their big socioeconomic predictions about leisure and hours faring notably worse than their narrower technical ones. Work hours in advanced economies stayed roughly flat, or crept upward, rather than falling toward anything like a three-day week, and the reasons compound rather than standing alone.

The single largest reason is that the productivity gains simply stopped reaching the typical worker after 1979. Before that year, according to Economic Policy Institute data, worker pay and productivity moved together in something close to lockstep — exactly the world Keynes, Kahn, and de Grazia were extrapolating from. Since 1979, productivity has grown roughly 93 percent while typical hourly pay has grown only about 34 percent. The missing gains did not vanish; they went to executive compensation and capital returns rather than paychecks, which broke the basic mechanism — rising income that ordinary workers could trade for time — that every leisure forecast depended on. Compounding this, employer cost structures actively push in the opposite direction from "spread the remaining work across more people": a large and rising share of labor cost, above all health insurance, is now fixed per employee rather than tied to hours worked, so it is cheaper to pay existing staff overtime than to hire and insure someone new — employer benefit costs have risen from under 3 percent of total compensation in 1929 to nearly 30 percent today. Rising consumption norms absorbed much of the income that did arrive: as Juliet Schor argued in The Overworked American (1992), households benchmark their living standard against a rising social reference point, so additional income tends to be spent maintaining relative position rather than banked as free time. The institutional muscle that had actually won shorter hours in the first place — the 1938 Fair Labor Standards Act, itself the product of decades of organized labor pressure — atrophied as private-sector union density fell from roughly a third of the U.S. workforce at midcentury to under 10 percent today; tellingly, countries that kept that institutional strength, such as France with its statutory 35-hour week, converted a meaningfully larger share of postwar productivity growth into time rather than income, which is itself evidence that shorter hours were never a technological inevitability but a policy and power choice. And there is a specific irony in the technology itself: a large share of the modern workforce is legally exempt from overtime pay, so laptops and then smartphones made work portable in a way a 1960s office job never was, letting the same or a growing volume of work leak into evenings and weekends — computers, in the end, extended work rather than compressing it.

VIII. Who got rich

If the productivity dividend did not arrive as leisure, the natural next question is where it actually went, and this is one of the better-quantified parts of the whole story. U.S. labor's share of national income fell from about 63.3 percent in 2000 to 56.7 percent by 2016 — a 6.6-point move in sixteen years — and McKinsey's analysis notes that three-quarters of the entire decline in labor's share since 1947 happened after 2000, squarely inside the computer-and-internet productivity era. The missing share went to capital: corporate profits and returns to shareholders and asset owners.

Within what remained of labor's share, the gains concentrated further at the very top. Since 1978, CEO compensation at large firms has grown roughly 1,085 percent on a realized basis against just 24 percent growth for a typical worker's pay, pushing the CEO-to-worker pay ratio to about 290-to-1 by 2023, up from something closer to 20-to-1 in the 1960s and '70s. Tellingly, 77.6 percent of average realized CEO compensation in 2023 was stock-related — exercised options and vested awards — meaning executives were increasingly enriched by the same equity appreciation flowing to shareholders generally, not by higher salaries drawn from operating revenue. And the shareholder concentration is stark in its own right: the top 1 percent of U.S. households now own roughly 54 percent of all publicly traded stock, up from 40 percent in 2002; the top 10 percent own about 93 percent of household stock market wealth; the bottom half of American households own less than 1 percent of it. The total value of the U.S. stock market roughly tripled between 2003 and today, from about $14.2 trillion to $46.2 trillion, and essentially all of that additional wealth landed inside an already-wealthy minority, because that is who owned the underlying shares to begin with. At the very top, this converges into a small number of familiar names: Bill Gates, Microsoft's founder, was the world's richest person for most of the 1990s and 2000s, the clearest personal signature of the PC-and-spreadsheet era; today's list has, if anything, tightened further around technology, with seven of the world's ten richest people tech-linked.

One honest complication is worth stating plainly, because it cuts against a simple morality tale: technology itself is a smaller direct cause of the labor-share decline than commonly assumed. McKinsey's decomposition attributes only about 12 percent of the shift to machines and software directly substituting for labor; larger contributors were real-estate and commodity boom-bust cycles (33 percent), the accounting shift toward depreciation and intangible capital like software and intellectual property (26 percent), and the consolidation of markets around a small number of "superstar firms" (18 percent). That last category is arguably technology's real mechanism of action here — not workers being mechanically replaced and their wages redirected to owners, but digital, network-effect-driven markets letting a handful of platform companies capture disproportionate market share, with a correspondingly disproportionate share of the economy's returns to capital following the same handful of owners.

Some further share of this is simple financial engineering rather than productivity at all. Stock buybacks were treated as market manipulation and effectively illegal until the SEC's Rule 10b-18 created a legal safe harbor for them in 1982 — buyback spending tripled within a year of that single regulatory change. From roughly $6.6 billion in 1980, buybacks grew to about $200 billion by 2000, over $1 trillion announced in 2018, and $942.5 billion actually executed in 2024, an all-time record; combined with dividends, S&P 500 companies returned nearly $1.6 trillion to shareholders in 2024 alone. Buybacks mechanically raise per-share price and earnings by shrinking the share count, entirely independent of any change in a company's actual output — a channel for converting corporate cash into shareholder wealth that simply did not exist, at this scale, before 1982, at almost exactly the moment the tech-productivity story begins. Layered on top of that is valuation inflation pure and simple: the Shiller CAPE ratio — a smoothed measure of price relative to a decade of earnings — sits around 40.6 today against a long-run median of 16.6, a level exceeded on only about twenty months since 1881, essentially all of them either 1999–2000 or now; the "Buffett Indicator," total market value divided by GDP, has climbed from roughly 50 percent in 1960 to about 153 percent at the 2000 peak to around 148 percent in 2021. Put simply: the correlation between the stock market and the real, productivity-driven economy is real but loose — a rolling ten-year correlation between the S&P 500 and U.S. real GDP fell from about 0.7 toward zero between 1958 and 1993 before climbing back to about 0.8 after 1993, and decoupled altogether in 2020, when the real economy cratered and stocks surged. A meaningful share of "the market's" growth, in other words, is not productivity finding its way to a price — it is investors paying ever more for the same stream of earnings, and companies choosing, since 1982, to spend a growing share of their cash buying their own stock rather than raising wages or reinvesting.

IX. The second coming

Set against this history, the current discourse around artificial intelligence reads less like a new phenomenon than a rerun with the volume turned up. The fear-versus-fascination framing is nearly identical, with "AGI" and "superintelligence" occupying the same rhetorical space "giant brain" and "mathematical Frankenstein" once did. The institutional mechanism that generated 1960s anxiety — the feasibility study cataloguing exactly what a worker does all day — has a direct descendant in Goldman Sachs' widely cited estimate that AI could automate tasks representing 25 percent of all U.S. work hours: an authoritative institution publishing precisely how exposed a given job is, which is exactly what made workers anxious sixty years ago before a single position was actually eliminated.

Where the pattern differs is speed. The gap between fear and actual job loss, which took years to close in the 1960s, is closing almost instantly this time: AI or automation was cited in 54 percent of tracked U.S. layoff events in 2026, up from under 8 percent the year before — a roughly sevenfold jump in a single year, touching over 170,000 workers and spreading well beyond technology into finance, logistics, media, and manufacturing. The caveat is the modern equivalent of hollow 1960s management reassurances: analysts at Deutsche Bank have identified what they call "AI redundancy washing," in which ordinary cost-cutting gets rebranded as AI-driven for narrative convenience — a pattern OpenAI's own Sam Altman has acknowledged. Meanwhile, the official "transformation, not destruction" framing that the 1966 National Commission offered ("technology eliminates jobs, not work") is being repeated almost verbatim by the institution with the most to gain from the optimistic reading: Goldman Sachs expects roughly 6–7 percent of workers displaced over a ten-year AI adoption period, offset by new infrastructure, technical, and service-sector jobs, netting to a projected 500,000 new U.S. jobs through 2030 — the same shape as bookkeeping clerks declining while accountants grew, forty years earlier.

The guaranteed-income debate, remarkably, has now actually been tested rather than merely argued about. Sam Altman-backed research through OpenResearch paid $1,000 a month, unconditionally, to 3,000 people in Illinois and Texas, and found recipients worked 1.3 to 1.4 fewer hours a week — a real but modest shift toward leisure, spent maintaining ordinary essentials rather than luxuries, not the wholesale abandonment of work that either side of the historical debate might have expected. It is the closest thing available to an empirical anchor for how people actually behave when income is decoupled from labor, and it looks far closer to the mild version of the Keynes and de Grazia thesis than to the dramatic one.

The specific, confident forecasts of a coming leisure society are also being repeated, nearly word for word, by essentially the same kind of person who made them the first time. JPMorgan's Jamie Dimon expects AI to deliver a three-and-a-half-day week; Bill Gates has mused about a two-or-three-day week and suggested humans may become "unnecessary for most things"; Nvidia's Jensen Huang calls a four-day week "probable"; and Elon Musk predicts work becomes fully optional within ten to twenty years.

Musk's own July 2026 interview with The Economist is worth pausing on as a case study, because nearly every claim in it has a direct sixty-year-old antecedent. "I've said many times, I think work is going to be optional," Musk told the paper, reaching for an analogy about people who still garden by hand even though machines do it better — an argument structurally identical to Kahn and Wiener's "leisure-oriented postindustrial society" and, further back, to Bertrand Russell's 1930 case that civilization's true achievement would be learning to use leisure well once machines had freed people from necessity. His underlying logic — "if that is so abundance that there's more that the robots are providing... than any human could possibly consume, what do you need money for?" — restates the Triple Revolution's 1964 argument that "cybernation" was building "almost unlimited productive capacity," with robots substituted for the word cybernation; his proposed remedy, "I think that the Treasury should just simply issue people checks," is the same guaranteed-income prescription that memo and the 1966 National Commission both reached six decades ago. His specific dates — AI exceeding "the sum of human intelligence" in about five years, "money won't matter" by 2036 — carry the same rhetorical confidence Kahn and Wiener brought to The Year 2000, a body of forecasting later found accurate only 45 percent of the time despite claimed near-certainty. And when The Economist's editor, Zanny Minton Beddoes, pressed him on the transition — what happens to workers in the years between now and the abundant end-state, before any distribution mechanism exists — Musk's answer was a slogan rather than a mechanism, considerably thinner than the operational policy the 1966 Commission actually proposed: income floors, national job-matching systems, relocation assistance, retraining institutes.

There is, finally, one dimension to this round that the mid-century forecasters did not have to contend with quite so nakedly. Musk is not a disinterested observer describing a future that will happen to him the way it happens to everyone else; as the controlling shareholder of the companies building both the robotics (Tesla's Optimus) and the AI (xAI) he says will make work optional, he is positioned to be one of the primary owners of the abundance he describes, in exactly the mechanism this essay has just quantified — equity ownership concentrated among a narrow, already-wealthy group, and executive pay increasingly paid in the very shares whose appreciation is the story. Even the economists responding to the interview have noted this directly: tech billionaires who benefit most from AI-driven wealth creation may resist funding the universal-income programs their own technology would make necessary. It is the same question the Triple Revolution and the productivity-pay gap have been circling for sixty years, arriving now with an unusually candid interested party attached to the argument: not whether machines can produce abundance, but who ends up holding title to the machines that produce it.

X. What history says will — and won't — happen

Treating this record as a base rate rather than a curiosity yields a reasonably specific set of expectations, several of which are already visible in real time.

The productivity paradox is very likely recurring on schedule. Solow's 1987 line — "you can see the computer age everywhere but in the productivity statistics" — is already being echoed almost word for word: the economist Torsten Slok has said "AI is everywhere except in the incoming macroeconomic data," and a survey of six thousand executives found nearly 90 percent report no measurable impact from AI on employment or productivity over the past three years, despite more than $250 billion invested in AI in 2024 alone. Historically, this gap took roughly a decade of organizational restructuring to close after 1980s IT investment; there is no strong reason to expect AI's version to resolve faster.

The gap between feared and actual job destruction will likely keep narrowing in appearance more than in substance, with a meaningful share of headline "AI layoffs" continuing to be ordinary cost-cutting relabeled for a more compelling narrative — a dynamic already documented and already named. A market correction of some kind is a reasonable bet on both historical pattern and current valuation data: every prior cycle of tech-driven exuberance this essay has traced — 1929, 1987, 2000 — eventually saw a valuation correction once pricing outran fundamentals, and today's leading AI companies are, by some measures, more richly valued on forward earnings than the dot-com era's leaders were, notwithstanding that they are actually profitable. And absent some real change in labor-market institutions, value capture will very likely keep concentrating rather than broadening — the "superstar firm" dynamic that characterized the platform era shows every sign of intensifying with AI, given how few firms can afford to build frontier models at all.

Several claims now prominent in the discourse look, by contrast, unlikely to hold up on anything like the timelines being floated. "Money won't matter," or work becoming broadly optional, by any near date is functionally the same claim Kahn and de Grazia made in 1967, a genre of forecast with a documented 45-percent hit rate against claimed near-certainty; there is no structural reason to expect Musk's 2036, Gates's "two or three days," or Dimon's "three and a half days" to fare better, and several well-documented reasons — fixed benefit costs, status-driven consumption, weakened labor institutions — to expect the same failure mode as before. A comprehensive, nationally enacted guaranteed income also looks unlikely on any near timeline: the identical proposal has now been made, in nearly identical language, by the Triple Revolution in 1964, the National Commission in 1966, Nixon's failed 1969 Family Assistance Plan, and today's AI-era commentators, and sixty years of the same idea recurring without being enacted at scale is itself strong evidence about how hard the underlying political-economy problem — who pays, and whether the technology's own beneficiaries will fund it — actually is. Full, category-wide job elimination, as opposed to the uneven, task-level transformation every past wave actually produced, also looks like an overreach: Musk's "everything, everything, everything" is a considerably stronger and more totalizing claim than mainframes or spreadsheets ever delivered, and there is not yet strong evidence this wave is different in kind rather than merely in scale. And broadly, evenly shared abundance, arriving without any deliberate redistributive policy, looks like the least likely outcome of all — the entire arc from the 1979 productivity-pay decoupling to today's equity-concentration data is a demonstration that scarcity was never really the constraint on shared prosperity; ownership and distribution were, and those are political choices, not automatic consequences of what a machine can do.

The honest caveat belongs at the end, not buried in the middle: it remains genuinely possible that this wave is different in kind. The breadth of tasks AI touches is larger than anything in this history, and the pace of capability improvement appears faster than the transition from mainframes to personal computers ever was. But "this time it's structurally different" is precisely what every generation in this account believed about its own automation wave, and a base rate built from seventy years of that belief being only partially right is the most useful thing this history actually has to offer — not a prediction of what AI will do, but a well-earned skepticism about anyone, however sincere, who claims to already know.

Sources and further reading

On the accounting profession and early automation: The History of Computerized Accounting (Fitek); How the Electronic Spreadsheet Revolutionized Business and The First Supercomputer vs. 'The Desk Set' (NPR); History of Adaptation Sets Stage for Future of Accounting (Bloomberg Tax); Episode 606: Spreadsheets! (NPR Planet Money transcript); Impact of Office Automation in the Internal Revenue Service (Bureau of Labor Statistics, via FRASER, Federal Reserve Bank of St. Louis).

On public sentiment and the "electronic brain": The mechanical monster and discourses of fear and fascination in the early history of the computer (Humanities and Social Sciences Communications / Nature); The Public's Fear of "Giant Brain" Computers (Penn State); The 1950s: Science and Technology (Encyclopedia.com); Future Shock (Wikipedia); "In the Presence of a New Force": Time Magazine's 1982 Machine of the Year (CAFE); On WIRED Magazine's Startup Phase (1993–1997) (Dave Karpf); The End of Techno-Utopianism (German Marshall Fund).

On the leisure-society forecasts: The Triple Revolution (Wikipedia); People Will Only Work Fifteen Hours a Week (Quote Investigator); 16-Hour Work Week by Year 2020 (Paleofuture); Automation and Job Loss: The Fears of 1964 (Conversable Economist); Evaluation of Some Technology Forecasts from "The Year 2000" (Coefficient Giving); Eberhard Ulich, biographical entries (Wikipedia; ETH Zürich memorial notice).

On the postwar economy and the productivity paradox: The Great Inflation and Recession of 1981–82 (Federal Reserve History); The Recession of the Early 1990s (Study.com); Solow's "Computer Age" Quote: A Definitive Citation (The Standup Economist).

On wages, hours, and who captured the gains: The Productivity–Pay Gap and CEO Pay in 2023 (Economic Policy Institute); Longer Overtime Hours: The Effect of the Rise in Benefit Costs (EveryCRSReport); The Overworked American (Wikipedia); A New Look at the Declining Labor Share of Income in the United States (McKinsey); The Richest 1 Percent Own a Greater Share of the Stock Market Than Ever Before (Inequality.org); Tech Talent Tops the 2026 Forbes Billionaires List (Yahoo Finance); Myth-Busting: The Economy Drives the Stock Market (CFA Institute); The Buffett Indicator (FullRatio); Shiller PE (CAPE) Ratio Today (thetrading.tools); Stock Buybacks Were Illegal Until 1982 (Boomers Broke America).

On the AI parallel: More Than Half of Layoff Events Tracked in 2026 Cited AI or Automation (IBTimes UK); How Will AI Affect the US Labor Market? (Goldman Sachs); Here's What a Sam Altman-Backed Basic Income Experiment Found (CBS News); Four-Day Workweek Possible in 2026? (Fortune); Elon Musk, interview with The Economist, July 2026 (transcript via SozAI; commentary via ai.joaoqueiros.com and Fortune); Thousands of CEOs Admit AI Had No Impact on Employment or Productivity (Fortune); AI Bubble vs. Dot-com Bubble: Torsten Slok Analysis (Fortune).