SLOW VARIABLES // EDITION 002 // 25 SEP 2026 THERE ARE NO LUDDITES LEFT Technology has never been stopped by the people who resisted it. It has always been shaped by them. A.I. is arriving into the weakest resistance environment in two hundred years. Status: exploratory Web: https://slowvariables.net/issues/002/ Technological change isn't always easy to accept. History is full of examples where a major innovation — the kind that upends the old order — shifts the way we understand each other. The debut of the printing press in the 1450s was unlike any innovation in Western history before it. Printing changed everything for European life; it spurred on mass exchanges of ideas, common vernaculars for different countries, and eventually two seismic shifts that forever changed the world: the Protestant Reformation and the Scientific Revolution. So as you can see, change is not good or bad — it just is. The consequences are impossible to know until they sneak up on us. For some people, change is a boon. The world becomes easier. Yet for others, especially those that thrived in the Before Times — it's a direct threat. We have a word for those people. We've had it for two hundred years, and we use it the way it was designed to be used — an insult. Luddite. Someone who fears the new, who can't keep up, who'd rather smash the machine than learn it. We're seeing tremendous resistance to the potentially most transformative technology in human history: A.I. The anti-A.I. backlash is real. And every concern anti-A.I. activists have — the environmental impact, the economics of automation and employment to name a few — have become front and centre. Yet many of these people aren't Luddites. Far from it. Because of that, it's worth knowing where the word comes from, because the people it was coined for weren't any of those things. Nottinghamshire, 1811. A group of men break into a hosier's workshop at night and take hammers to the wide stocking frames — the machines that let an untrained hand turn out cheap, shoddy "cut-up" goods that were flooding the market and driving skilled knitters' wages into the floor. They smash those frames. They leave the others.[1] That selectivity is the whole story. They weren't against the machine, because they worked machines their entire lives. They were against what this particular machine was being used to do, and to whom. It was a negotiation, conducted with the only instrument available to people who had no vote, no union and no one in Parliament who'd take their letter. Parliament's answer came the following year: an Act making the destruction of a frame a capital offence. In January 1813, at York, seventeen men were hanged.[2] So the question of who sets a technology's terms was answered early. That's the origin of the insult. Keep it in mind, because this edition is about what the people we call Luddites actually achieved, and what it means that there's nobody left to do it. WHAT RESISTANCE ACTUALLY DID ============================ The standard story goes like this: people resist new technology, the technology wins anyway, and everyone's better off in the end. Resistance is futile, foolish, and so forth. The first half's true. The technology almost always wins. The frames kept running. The railways got built. The cars filled the roads. I honestly can't think of a major technology in two centuries that organized resistance actually stopped. But look at what resistance _did_ do, in every case where it existed. The early factories ran children sixteen hours a day, and nobody inspected anything. The Factory Act of 1833 created the first paid inspectors and put limits on children's hours; the Ten Hours Act followed in 1847.[3] Neither was a gift from the mill-owners. Far from it. They came after two decades of agitation, strikes, pamphlets and a reformist press that wouldn't drop the subject. The eight-hour day was a demand long before it was a law. The rally at Haymarket in Chicago in 1886 was for it. It took another fifty-two years and a lot more trouble before the United States made the forty-hour week general law, in 1938.[4] No productivity model produced that outcome. People did, over and over, until it stuck. Britain's railways killed passengers at a steady rate for decades, and the companies resisted every safety measure as an unaffordable expense. Then, in June 1889, a train's carriages rolled back down a hill at Armagh and eighty people died, many of them children on a Sunday-school outing. Within months, Parliament made continuous automatic brakes and block signalling compulsory.[5] The railway didn't get safe on its own timetable. It got safe when the public refused to keep dying. In 1906, the U.S. passed the Pure Food and Drug Act, months after Upton Sinclair published _The Jungle_ and the country discovered what was in its sausages. And take cars. Mass adoption by the 1920s, Ralph Nader's _Unsafe at Any Speed_ in 1965, federal safety standards from 1966, seatbelts required in new cars from 1968, the first state law requiring you to actually wear one in 1984.[6] It took roughly sixty years from the technology to the obvious safeguard, and the industry fought every single step. Here's the through-line: in every one of these cases, the technology arrived anyway. What resistance decided was the _terms_: who paid, who was protected, how fast it spread, and on whose say-so. Resistance never stops technology. But it can set the terms of how that technology is used. That's the part the standard story leaves out. Every protection you take for granted at work — the weekend, the safety rail, the inspector, the right to know what's in the thing you're buying — was extracted from a technology's owners by people who broke, struck, sued, reported or refused. None of it was in the original design. RESISTANCE IS INFORMATION ========================= There's a second thing the standard story gets wrong, and it's about who resists and why. People resist when the costs of a technology land on them and the gains land somewhere else. That's it. That's the pattern, in 1811 and 2026. And they're usually right about the distribution, even when they're wrong about the machine. The Luddites' economics were sound. The frames really were being used to replace skilled work with cheap work, wages really did collapse, and the gains really did go entirely to the hosiers. Sound familiar? Their tactics were what they had. Their analysis of who was about to pay for progress was correct. Now look at who's anxious about A.I. today. Illustrators. Translators. Writers. Voice actors. Junior lawyers and analysts. Copywriters. Teachers. Call-centre staff. Anyone whose work produces a plausible artifact that a model can now produce for almost nothing. That list isn't a catalogue of people who fear the future. Most of these folks have been quite willing to take on new tech when it helped them do their jobs. What this does show is an early map of where the costs are hitting first. Dismissing it as technophobia throws away the best information we have about the terms currently being set. It's also worth noticing who does the dismissing. In two hundred years, the people calling resistance irrational have been, without exception, the people the costs weren't landing on. THE CAPACITY TO RESIST HAS BEEN DISMANTLED ========================================== So far this is history, and history is cheap. Here's the part that isn't. Call it a society's _capacity to resist_: its organized ability to refuse a technology's terms. It's not one thing. It's unions and professional bodies with real bargaining power. It's a local press that notices harm before it becomes national. It's courts that move at a useful speed. It's legislatures that fear voters more than donors. And underneath all of those, it's the plain habit of collective refusal — the assumption that when something is done to you, you find the others it's being done to and you push back together. That capacity is the slow variable in this edition. And it has been collapsing for over 40 years. Here's a crazy statistic: in the mid-1950s, about a third of American workers belonged to a union. Today it's around one in ten, and in the private sector closer to one in sixteen.[7] Britain went from roughly half of workers in 1979 to just over a fifth. Local newspapers have closed by the thousand; whole counties now have nobody paid to sit in the council meeting. The professions still gatekeep, but slowly, and mostly on behalf of their own members rather than the public. Compare that to what earlier technologies walked into. The factory met a labour movement that was small and then grew enormous _because of_ the factory — the resistance was built by the thing it resisted. The car met a strong press, consumer advocates who could get a hearing, and a Congress that still legislated. Even the internet met a functioning antitrust tradition, for a while. A.I. is meeting none of those at strength. It's arriving into the weakest resistance environment any comparable technology has faced since the Industrial Revolution began — and it's arriving faster than any of them. For many, this feels like surrender to something they can't stop. There's one exception, and it's worth slowing down for. In May 2023, the Writers Guild of America went on strike. It lasted 148 days. The contract that ended it contained something that, at the time, existed nowhere else on earth: negotiated terms for how a generative A.I. model may and may not be used in an industry's work. The studios can't use A.I. to write or rewrite literary material. A.I.-generated text can't be used to cut a writer's credit or pay. Writers have to be told if what they've been handed came out of a model.[8] In one industry, in one country, the terms of A.I. were _bargained_ rather than announced. The mechanism — the same one that produced the Factory Act and the seatbelt — still works. And why did it work there? Because screenwriters are one of the last groups of workers anywhere who have all four things at once: a union, a contract, leverage over something the owners need, and enough solidarity to sit out a summer. The writers proved it can still be done. Now count who else has all four. It's a short list, and it's getting shorter by the month. TERMS SET BY DEFAULT ==================== What does a low-resistance environment actually produce? You don't have to guess. You're living in it every day. Understand: A.I. frontier models are trained on literally everything that isn't nailed down. Assume anything you write that goes public will end up in a training run. Liability is always pushed to the user in the Terms of Service. Deployment first, evaluation later, if at all. Products that reshape a profession are released without so much as a conversation with the profession. This isn't the A.I. companies being evil. It's just the mechanism of the marketplace. There's a substitute on offer: regulation from above. The EU model, basically. It's real and it's better than nothing, but it isn't the same thing as resistance, and it's worth being honest about why. Policies from government are always being written by people who don't bear the costs. And it's negotiated with the owners' lobbyists, not with the people whose work is being replaced. Regulation without resistance underneath it. The terms being set now will decide something bigger than any single job, which is who captures the gains once human labour stops being the scarce input to production. And it's coming sooner than you realize. That's the next edition. For this one, it's enough to say: the property question is going to be answered either at a bargaining table or by default, and right now most of the tables are empty. Edition 001 argued that the A.I. priesthood dissolves and what survives is accountability — a named human who answers for the output. That's true. But accountability is a _term_. Somebody has to insist on it. Nothing in the technology puts it there. WHAT WOULD HAVE TO BE TRUE ========================== If this edition is right, the next few years should look like this: A.I. protections stay confined to the one worker in ten who has a union to bargain them; the other nine get terms of service. A.I.-specific law arrives late and mostly on industry terms. Union membership stays flat or keeps falling through the transition. And the writers' contract stays a curiosity rather than a template. If it's wrong, you'll see the opposite. A.I. protections start reaching workers who don't have a union to win them. New forms of collective refusal that don't look like unions — data strikes, mass opt-outs, professional bodies setting binding rules for their members. And union density rising for the first time in half a century _because of_ A.I., the way it rose because of the factory. I'd like to be wrong. If, by 2029, A.I. has built a labour movement the way the mills did, then this edition misjudged the capacity, not the pattern — and I'll say so on this page, under Revisions. THE SELECTIVE HAMMER ==================== Go back to that workshop in 1811. The men with the hammers knew which frames to break and which to leave. They could tell a machine that made the trade better from a machine that was being used to gut it. That discrimination — not fear, not nostalgia, but a working judgement about which uses of a technology were acceptable and which weren't — is exactly what a society needs to be able to exercise right now. We have almost no way to exercise it. The instruments that used to do it have been taken apart, mostly on purpose, over four decades, and the one recent example of them working is treated as a Hollywood story rather than a warning. The frames kept running. They always do. The only question that was ever open was the terms, and the terms are being written now, by the people who own the frames, in a room where the other chairs are empty. The claims that follow from this are on the ledger below, with dates. Come back and check. [1] The best account of the selectivity is still E. P. Thompson, _The Making of the English Working Class_ (1963), ch. 14. The Luddites' own letters and proclamations are collected in Kevin Binfield (ed.), _Writings of the Luddites_ (2004); they read as labour negotiators, not vandals. [2] The Frame-Breaking Act (Destruction of Stocking Frames, etc. Act 1812) made machine-breaking a capital felony. Byron's maiden speech in the Lords was against it. The York Special Commission of January 1813 hanged seventeen men in connection with the West Riding disturbances. [3] Factory Act 1833: four paid inspectors, no children under nine in textile mills, limits on hours for those under eighteen. Factories Act 1847 (the Ten Hours Act) limited women and young persons to ten hours a day. [4] Haymarket, Chicago, 4 May 1886 — a rally during a general strike for the eight-hour day. The Fair Labor Standards Act of 1938 established the forty-hour week for most US workers. [5] Armagh rail disaster, 12 June 1889; around eighty dead. The Regulation of Railways Act 1889 received royal assent that August and made continuous automatic brakes, block signalling and interlocking compulsory on passenger lines. [6] Ralph Nader, _Unsafe at Any Speed_ (1965); National Traffic and Motor Vehicle Safety Act (1966); Federal Motor Vehicle Safety Standard 208 required seatbelts in new passenger cars from 1968; New York passed the first mandatory-use law in 1984. [7] US Bureau of Labor Statistics union-membership series: around 35 per cent of non-agricultural workers at the 1954 peak, 10.0 per cent of wage and salary workers in 2025 (14.7 million members) and 5.9 per cent of the private sector; 20.1 per cent when the current series began in 1983. UK figures from the Department for Business and Trade's trade-union statistics: about half of employees in 1979, a little over a fifth now. On local news, Northwestern's Medill School counts well over two thousand US newspaper closures since 2005. [8] WGA strike, 2 May to 27 September 2023. The A.I. terms are in Article 72 of the 2023 Minimum Basic Agreement: A.I. can't write or rewrite literary material and can't be treated as source material; a writer may choose to use it with the company's consent; companies must disclose when material given to a writer was A.I.-generated. The Guild's own summary of the agreement is the clearest source. FAST VARIABLES ============== - A Berkeley researcher counts more than 175 union contracts with A.I. language (Labour) The mechanism, still working: Oregon's state workers banned A.I. from monitoring and discipline, McClatchy's newsrooms banned it from replacing staff, and a Washington, D.C. bookshop wrote A.I. protections into its contract. Then remember the denominator. Those 175 contracts sit inside the one worker in ten who has a union at all. - Colorado's A.I. law, enacted in 2024, is delayed again and narrowed on the way (Regulation) Resistance from above, on schedule: passed May 2024, pushed from February to June 2026 after industry objected, with a draft rewrite on the table that would cover less and land in 2027 — while Washington talks about pre-empting it altogether. Late, lobbied, and not the same thing. - GitHub trains on your code unless you say no; Apple trains on your Siri chats only if you say yes (Terms) Two defaults, same year. Opt-out where the users can grumble but not leave; opt-in where the brand depends on being asked. The default is the term, and who sets it depends on who can walk. - U.S. union membership holds at 10.0 per cent for 2025 — 5.9 per cent in the private sector (The variable) Fourteen point seven million members; 20.1 per cent when the series began in 1983. 'Stable,' the headlines said. Stable at the bottom is still the bottom, and it is the number this whole edition rests on. LONG MEMORY =========== 1812 Parliament passes the Frame-Breaking Act, making the destruction of a machine punishable by death. Seventeen men are hanged at York the following January. The question of who sets a technology's terms was answered early, and with a rope. 1833 The Factory Act creates the first paid inspectors and limits the hours of children in textile mills. Not a gift from the mill-owners; the product of two decades of agitation, and the beginning of the idea that a technology's terms could be legislated at all. 1886 Haymarket, Chicago: a rally for the eight-hour day ends in a bomb and a show trial. The demand it was for took another fifty years to become general law. Resistance sets the terms, but not on its own timetable. 1889 Eighty people die at Armagh when a train's carriages roll back down a hill. Within months, Britain makes continuous brakes and block signalling compulsory. The railway did not get safe by itself; it got safe when the public refused to keep dying. 1965 Ralph Nader publishes Unsafe at Any Speed. Federal car-safety standards follow within three years, mandatory seatbelt use within twenty. Roughly six decades from mass adoption to the obvious safeguard, every year of it fought by the industry. 2023 The Writers Guild strikes for 148 days and wins the first negotiated A.I. terms in any industry contract. Proof the mechanism still works, and a measure of how few groups still have everything it requires. LEDGER ====== [75%, by 2028-12-31, open] By the end of 2028, fewer than one in twenty U.S. private-sector workers will be covered by a collective agreement containing enforceable limits on an employer's use of generative A.I. [70%, by 2029-12-31, open] By the end of 2029, no US federal law will impose liability on A.I. developers for harms caused by model outputs in ordinary commercial use. [60%, by 2030-01-31, open] US union membership as a share of employed workers (BLS annual figure) will be lower for 2029 than for 2024. REVISIONS ========= 2026-09-25 First published. -- Written by a person, under a pseudonym. https://slowvariables.net/colophon/ Corrections: editors@slowvariables.net