--- title: Introduction — An Engine Without a Governor book: Purposive Accelerationism author: '@jamie247' position: 3 of 24 canonical: https://whatispacc.org/ebook/#introduction version: 1.1 language: en-GB licence: all rights reserved; quote with attribution --- # Introduction — An Engine Without a Governor “Men will set the goals and supply the motivations, of course, at least in the early years.” — J. C. R. Licklider, “Man-Computer Symbiosis”, 1960 The scale, at least, everyone can see. The build-out now underway around artificial intelligence is, in absolute terms, the largest concentration of private capital in the history of enterprise. Trillions of dollars are moving into a single technical programme: the chips, the data centres to house them, and, increasingly the binding constraint, the electricity to run them. Firms at the frontier no longer plan their futures in units of engineers or customers. They plan in gigawatts. Power-purchase agreements have become strategic weapons, and decommissioned nuclear plants are being restarted under exclusive contract to single buyers, Three Mile Island among them, revived for Microsoft on a twenty-year deal signed in September 2024. Asked what stands between them and their ambitions, the men running these companies do not say talent, or capital, or even compute. They say energy, the raw thermodynamic input, as though the economy were being re-founded on the only currency physics accepts. Sam Altman has said that the cost of intelligence will converge over time on the cost of energy. On his account, however cheap the chips become, nobody can discount an electron. Whether that convergence is real remains an open question. But the conviction tells you what the buyers believe. Commentary on all this divides, predictably, into awe and alarm, and both camps share an assumption: that the scale is the remarkable thing. We have built at civilisational scale before — railways, electrification, the post-war highway systems, the internet itself — and the pattern of those build-outs is instructive where today’s departs from it. What is remarkable about the machine being assembled now is not what has been put into it. It is what has been left out. To see what is missing, you have to know what was once there. In 1788 James Watt faced a problem that had nothing to do with making his steam engine more powerful and everything to do with making it usable. An engine’s load and its steam supply both vary. Left alone, the imbalance compounds, so that the engine either slows towards stalling or races towards destruction. Watt’s solution, adapted from a device millers already used on their grindstones, was two iron balls mounted on a spindle driven by the engine’s own output shaft. As the engine speeds up the balls whirl faster and swing outward, and their swing, through a linkage, narrows the throttle valve and admits less steam. As the engine slows the balls drop, the valve opens, and more steam flows. No operator, no schedule of permitted speeds, no inspector: a loop of iron in which a measure of the engine’s own output has been bent back to govern its input. It was called the centrifugal governor, and it is in a strict sense the founding artefact of the modern world. What it added was not power but answerability. From that point a machine could be held, mechanically, to a setting its builders had chosen. Before it, a steam engine was a demonstration. After, an engine could be trusted with a purpose that remained at every moment its owner’s and never its own, which is what turned combustion into industry, because steam could now drive looms, mills and pumps unattended. In 1868 James Clerk Maxwell, taking time away from electromagnetism, published a paper titled simply On Governors, working out the mathematics of when such feedback holds steady and when it oscillates out of control. Control theory, on which every aircraft, nuclear reactor and rocket now depends, descends from that paper. From the beginning the deepest minds understood that the governor was not an accessory to the engine but the condition of the engine being effective. An engine without a governor does not run free. It runs away. Runaway is a term of art rather than a metaphor: the condition in which output feeds input with nothing to bend the loop back, so that the machine accelerates until it tears itself apart, along with the mill and anyone standing in the way. The choice was never between a governed engine and a free one. It was between a governed engine and an explosive with a delay. The term governor descends from the Latin gubernator, which renders the Greek kybernetes, the helmsman at the steering oar. Who holds the ship on course through forces he does not command, the wind, the current and the vessel’s own momentum, and he could not survive by opposing them head-on. He wants the wind, and steering is what turns it into progress. When, in 1948, Norbert Wiener needed a name for his new science of purposive machines, systems that hold a course by comparing what they are doing against what they are supposed to be doing and correcting the difference, he reached past the steam engine all the way back to the steersman and called it cybernetics. Nearly three thousand years runs through helmsman, governor and cybernetics alike, and it is still at work in the industry’s own stack, where Kubernetes, the software that steers fleets of computing containers, is the Greek word verbatim. To govern is to steer. The question underneath everything that follows is who, or what, is at the oar today — and what to make of a school of thought, now commanding the largest capital formation in history, which holds that a hand on the oar is itself the problem: that the current knows where it is going, and steering is drag. *** That school has its temples in the American technology industry, and its leadership is the priesthood met at the start of this book. Its texts are creeds, its mode is liturgy, and its anathemas are pronounced with theological confidence. Two men will stand for the movement here, as its most consequential members and its most explicit. The first is Marc Andreessen: co-author of the first mass-market web browser, co-founder of one of the largest venture firms of the software era, a16z, and author, in October 2023, of The Techno-Optimist Manifesto, the closest thing the movement has to a catechism. It names its enemies in a formal list, which is itself instructive: sustainability, tech ethics, risk management, the precautionary principle. Every entry is an institutional form, imperfect perhaps, by which modern societies bend a technology’s output back towards its input. And it rests its optimism on energy, which it calls life itself, and names among its patron saints the founders of e/acc, whose borrowed thermodynamics has the universe conveniently wanting what they want. In genre it is a hymn rather than a brief, and hymns are not meant to be examined. They are meant to be sung. The second is Peter Thiel, who built PayPal, took from his Stanford teacher René Girard a theology of desire and sacrifice he has never stopped applying, wrote Facebook’s first outside cheque, took Palantir into the security state, and supplies the movement’s temperament and its candour. In 2009, in an essay for the libertarian journal Cato Unbound, he wrote the sentence to which everything that follows is an extended response: “I no longer believe that freedom and democracy are compatible.” The extension of the franchise had, he argued, made democratic majorities hostile to the libertarian project, so freedom would have to be pursued where votes cannot reach — in technologies that “create a new space for freedom” beyond the reach of democratic politics. He was describing his own loss of faith in elections as a route to liberty, not proposing to disenfranchise anyone, and when the essay drew fire he added that no class of people should lose the ballot, only that he held little hope it would make things better. Thiel matters more to this argument than a thousand vaguer enthusiasts because he has written his premises down. Where others gesture, he states: when freedom and collective self-rule come apart, it is democracy that should give way. That is no retreat from politics but the choice of a different kind, in which whoever builds the new spaces writes their rules. The proposition has serious money behind it, and everything turns on the work the word freedom is performing in his creed. The priesthood is formidable. Its argument is that technology has been the great engine of human flourishing and that halting it would cost more than it saves. By the measures that decide what gets built, capital deployed, political money spent and the absence of any rival in speed, it is winning. Answering it requires knowing where its ideas come from, which is somewhere older and stranger than a venture firm’s marketing department. Accelerationism’s roots run back to French philosophy of the 1970s, and behind that to the Italian Futurists, who in 1909 declared a roaring motorcar more beautiful than any statue and went on to write the founding manifesto of fascism. But it took its modern form not in California but in Warwick, England, in the 1990s, inside a renegade academic group called the Cybernetic Culture Research Unit. Founded in 1995 by the cybernetic feminist Sadie Plant, it was inherited and then radicalised by Nick Land, whose feverish theory-fictions broke with the academy and seeded accelerationism as we now know it. Land’s thought is radical in the strict sense, and no summary survives contact with his prose, which is fast, polemical, funny and frequently brilliant. At its core is this: capitalism, fused with computation, is not a human institution at all. Importantly, by capitalism he means the profit-motivated system of private ownership, not the mere existence of markets. For him it is an inhuman process, self-exciting, self-expanding and indifferent, which uses human beings the way a wasp uses a caterpillar, as a living host for its young. Humanity is not the protagonist of history but its larval stage, and the process now accelerating through our economies is the thing we are in service to. Where a humanist recoils, Land’s writing exhibits something closer to worship. This is the Valley’s intellectual inheritance: not Adam Smith, not even Ayn Rand, but a 1990s philosophy of capital as an alien intelligence bootstrapping itself. Land put it at its most extreme when he called capitalism a kind of artificial intelligence, “an invasion from the future by an artificial intelligent space that must assemble itself entirely from its enemy’s resources”. On this reading the machine intelligence now emerging is no novelty at all but capital becoming literal, finding in silicon the substrate it had until now borrowed from us. Land’s claim, however, mistakes a name for an agent. Capitalism is a process built from arrangements — property, contract, price, credit — much as the weather is wind and rain, and there is nobody inside it to want anything. The horror is real but mislocated. It needs no alien mind using us, only a process automated from end to end, each result feeding the next, that runs with no will at all. It was from here the inheritance forked. On the left, Alex Williams and Nick Srnicek argued in their 2013 #Accelerate manifesto, the founding text of what is now called left accelerationism or l/acc, that the machine should be accelerated and captured, that the technologies capital builds could be repurposed by political mastery for human emancipation. This version kept the question what is it for? at the centre. The Valley’s variant, effective accelerationism, whose creed was authored principally by the physicist-founder Guillaume Verdon, performed the opposite surgery on each parent. From Land it kept the sovereign process — capital as the thing that answers to nothing while everything answers to it — and discarded the dark honesty about where that process leaves us. From l/acc it kept the speed and discarded the emancipatory politics. Its authors wanted automation to win back people’s time, and in their 2015 book Inventing the Future they asked for a shorter working week and a universal basic income, demands addressed above all to the state. Its popular cousin, Aaron Bastani’s Fully Automated Luxury Communism (2019), went further and recast the communist ideal for an age of machine abundance. What remains, once you have removed both the destination and the question, is a throttle. E/acc is the doctrine of throttle: pure acceleration, cosmically licensed, with the steering reclassified as drag. The family has produced one serious internal dissent. In late 2023 Vitalik Buterin published the case for d/acc, its d standing at once for defensive, decentralised, democratic and differential acceleration. Buterin is the nearest thing crypto has to a resident philosopher, ironically a Thiel Fellow who left university at 20 on a hundred-thousand-dollar grant to build Ethereum, and one of the few figures the priesthood cannot dismiss as a Luddite. Accelerate, he argued, but accelerate selectively, favouring the technologies that defend and decentralise over those that attack and concentrate. This was an advance, the first accelerationism native to the technology industry to admit that direction is not one variable among others but a presupposition. And it concedes the decisive premise: technology answers to values, and the only question left is whose. But selection alone is not steering. Favouring defensive technologies expresses a preference. It does not supply the reference value, the chosen heading against which a system’s behaviour is measured and corrected. D/acc fits the brakes and the airbags and asks that the passengers be heard, but the vehicle still has no steersman. The advance is real, and so is the vacancy at the oar. *** The territory of all this is usually mapped on a single axis, faster or slower, which is just the picture the priesthood needs, because on one axis every governor looks like a brake. Drawing the real map, as the Three-Axis Map figure does, requires three: speed, running from deceleration to acceleration; the governor, from absent to present; and the shape of power, from a single to plural hands. The decelerationists — degrowth, the precautionary establishment — hold the slow-and-governed region, and their position concedes the decisive ground. Moratoria have worked before, on human cloning and on germline editing, but never against a technology with a trillion-dollar industry lobbying over competing visions and a global arms race behind it with no definition a legislator could write down. The engine will not be stopped by people who do not hold the throttle, and the attempt guarantees that whoever wins on speed faces no rival there. In November 2026 the slow corner found its electorate, when voters in ten American states claimed the right to refuse a data centre. The priesthood holds the fast-and-ungoverned corner with overwhelming force, and drifts along the third axis as it does so, because unchecked concentrations of power do not stay plural. At the time of writing, in late 2026, China occupies fast-and-governed with the power axis closed: an acceleration with a steersman, a party-state whose helm answers inward, to its own hierarchy and to nobody else. On two of three axes China and I agree, and we part on the third, because a governor answering to a single hand is itself a totalisation, and the oldest. Among the democracies, the European Union comes nearest to governed and plural, and pays for it in speed; Britain for now runs fast and lightly governed, and the United States fast at the frontier and contested at the grid. The left accelerationists stand nearest the corner this book argues for, fast and steered, but they fix the ends in advance and lean on the state to keep them. [Figure: the Three-Axis Map. Positions plotted on speed (slow to fast), governor (absent to present) and power (one hand to many). p/acc: fast, governed, plural. e/acc: fast, ungoverned, power gathering to one hand. decel: slow, governed, plural. TAS: fast, fully governed, power in one hand. Interactive at the page.] The map filled in during one week of September 2026. OpenAI called for mandatory federal safeguards. Anthropic’s chief executive, Dario Amodei, published a plan to pace the frontier and asked Washington for an antitrust waiver so that the foremost companies of the democracies could agree limits on the rate of their own progress, and Sam Altman endorsed him within hours. David Sacks, until March of that year the White House’s adviser on artificial intelligence, told them to go ahead, since the two firms already held “a duopoly on frontier intelligence”, and to stop pretending they needed a waiver or an approval process that superseded product liability. A moratorium had its advocates both in Congress and, days earlier, in a letter from more than seventy UK MPs and peers asking for a ban on superintelligence. The leading labs had moved to fast and held in two hands, governed only by a throttle they would set themselves, with the state as guarantor. Their critics held fast and plural, with liability and the customer as the only governor they would admit. The movement’s popular wing, e/acc, went further, meeting the labs’ pacing plan with a call for decentralised AI that anyone could run without permission, an open stack of models, tools and local inference. That hints at a direction, and it is an improvement I endorse, but on its own it is a topology. It says where the weights sit, not what they answer to, and a model on your own machine can carry an objective you cannot inspect as easily as one on a frontier lab’s server. By the end of the month six of the industry’s leaders had signed a voluntary accord at the White House, binding on no one, and just six weeks later voters backed a hostile legislature against a president with a veto, leaving a bill that would stop the work on pain of prison as the only federal alternative. Each offer puts a single party in charge of the speed. The corner that stands empty, and the subject of this book, is fast, governed and plural: full speed, with a measure of the output bent back by design, at machine speed, to hold the machine to humanly chosen and humanly revisable ends. This is the position the book’s title names, purposive accelerationism, or p/acc in the family’s own naming convention. Purposive, because what distinguishes it from e/acc is not less throttle but the presence of a destination. Accelerationism, because unlike the decelerationists it wants the engine run at full speed and the technology built. That corner has never been empty of people. Engineers, cryptographers, constitutional cypherpunks and Buterin himself have all stood in it. However, it has been empty of architecture. Its occupants have had instincts, tools and ways out, the means to leave a system rather than change it. What they have lacked is a constitution, meaning a worked-out account of what the governor is, what reference value it holds, what materials it is built from, and why it binds. Supplying that is the work of this book. Its answer is not a charter asking governments or firms for anything. It is a set of rights built into the internet itself and, importantly, into the emerging AI stack, a cognitive commons no single owner can close, with ways for those living under it to steer and tests to measure its degree. They are gathered on a page in the appendix: the reference value, the criteria any governor must meet, the rights, the tests that check them, the provision and the rulebook, and the one red line never to be crossed, for any reader who wants to keep score as the argument builds. The engineering, which primitives in which arrangements at which parameters, belongs to founders and the market, so that nothing argued here dates with any particular configuration. That matters more than it sounds, because the materials themselves carry a theoretical expiry date. The public-key cryptography beneath every key and ledger we have pointed to in these pages is scheduled for retirement, deprecated by the body that standardises it from 2030 and disallowed from 2035, because the expectation is a large enough quantum computer breaks it. Its successors however are not waiting on that machine; they are ordinary mathematics, chosen in an open competition that ran for eight years, and already published. So the architecture proposed here has to do one thing the cypherpunks never had to: carry a constitution across the replacement of its own electronic foundations, with no party in charge, on a deadline set by physics. That is not an objection to the project but its first test, and a protocol nobody owns is the only kind whose passage everyone can watch. My own position needs stating before going further. Alongside a decade of investing in crypto, I was an early advocate of something called platform cooperativism, a movement that set out to give the people who create a platform’s value a share in owning and governing it. The claim made here for crypto is accordingly a narrow one. Co-operatives, mutuals, unions and commons have held shared goods by construction for centuries and still do. What crypto adds is the ability to hold such things across borders, at the speed machines now work, without anyone’s permission, which none of the older forms can manage. Two sets of instruments are needed for our work here, one technical and one philosophical. The technical set comes from cybernetics: Wiener’s account of purposive systems, widened by Gregory Bateson into an account of minds that extend beyond any one skull, and hardened by Stafford Beer into an architecture for institutions that can hold a course. That tradition named the governor half a century before the priesthood deleted it. The philosophical set reaches further back still. Isaiah Berlin and Amartya Sen, writing when freedom was at its most contested in modern history, supply the anatomy of what it consists of, and catch the priesthood invoking one kind of liberty while its prescriptions serve only the other. Hume set the bar the priesthood must clear, by showing that no quantity of is yields an ought, and Moore devised the test that no natural definition of the good survives. Both blades fall across e/acc’s thermodynamic catechism. And finally Arendt supplies what the reference value must in the end protect, which is plurality. The witness I most look forward to calling is Friedrich Hayek, the priesthood’s own house god. He understood the price system as a feedback device, a governor for allocation though never for direction, and his own writing is a sustained argument that it works only inside constraint. He will testify, against his admirers, that the wall they are demolishing is load-bearing. My method is to take the strongest version of the case against constraint, stronger than its proponents might usually state it, grant everything in it that is true, and then show that the single move that matters, the deletion of the governor, does not follow from any of it. A defeat on home ground, by the movement’s own first principles, shows that an acceleration which keeps its governor is not a timid compromise but the only version of the project that is coherent, and truer to what the priesthood says it wants. Nobody has ever been argued out of a religion by being told it was stupid. They have occasionally been shown, from inside, that their god requires something other than what the priests demand. The reason to do this now, rather than watch another decade of evidence accumulate, is a fact about infrastructure that every engineer knows and every politician learns too late: constraints are cheap to design in and ruinous to retrofit. The internet’s original sins — no native identity, no native money, no native provenance — were each a decade of papers away from being design choices. They are now structural absences that entire surveillance and platform economies have calcified around, and some may never be corrected. The machine being assembled today at gigawatt scale is still in development. Its protocols are still being written, its settlement layers are still contested, its governance, where there is any, is still molten. For a short time, what it optimises for is a design question rather than an archaeological one. The governor can be specified now, by people who understand the substrate, or excavated later from the rubble. So the case must be heard before it is answered. The strongest argument against everything I propose — the case for removing the governor outright, made in earnest by the most capitalised movement on earth — deserves its day at full strength. It begins, as the best arguments do, where everyone notionally stands: with the cause of freedom.