Why good innovations mint new elites — and how to tell the dangerous cases from the merely disruptive
Every year something like forty thousand Americans die on the roads. The strongest large-scale data we have — from driverless fleets logging hundreds of millions of miles — suggests that within the limited areas where they operate, robotaxis injure people far less often than human drivers do. The data isn’t fully independent, and the operating areas are cherry-picked for ease. But the direction is not really in dispute. So who lobbies hardest against letting these cars spread?
Among the most persistent opponents, for nearly a decade now, have been trial lawyers. The arithmetic is not subtle. Ordinary auto litigation is a vast industry, and an industry built on collisions does not welcome a technology that removes most of the collisions. This looks like a clean morality tale: lifesaving progress on one side, a rent-seeking guild on the other. It is a clean tale. It is also the weakest possible version of a much older and much more dangerous pattern, and mistaking the two is the first error worth clearing away.
Two things that look like one
There are two distinct phenomena hiding inside “a powerful group fights a good innovation,” and they behave nothing alike.
The first is rent dissipation. An innovation destroys a friction that some group was harvesting. The group’s power derived entirely from the friction, so when the friction goes, the rent goes — and there is nothing underneath it. Trial lawyers on autonomous vehicles are this. So were the scribes facing the printing press, the medallion owners facing ride-hailing, the toll keepers facing the railroad. These groups can delay, extract transition payments, win carve-outs. What they cannot do is reverse the change, because their power has no substrate other than the thing being removed. They also, tellingly, retool. The billboard lawyer asking whether you’ve been injured by an eighteen-wheeler will, without missing a beat, start asking whether you’ve been injured by a self-driving one.
The second is elite reconfiguration, and it is a different animal. Here the innovation relocates the material basis of power itself. The gunpowder weapon is the paradigm. The musket did not make the mounted knight poorer; it dissolved the thing that made a hereditary warrior caste possible — an expensive, training-intensive, decades-long monopoly on effective violence. A conscript with a few weeks’ drill could kill the man who had trained since childhood. The knight’s whole world rested on that no longer being true, and once it wasn’t, the world went with it. Run the same story a thousand years earlier: bronze required tin, a rare traded input a palace could choke, and the elite that controlled the chokepoint controlled the weapons; iron is nearly everywhere, and mass infantry — and the political forms that ride on mass infantry — followed the ore.
The trial lawyer is the first kind wearing the second kind’s costume. The interesting question is never really about him. It is about when rent dissipation turns into elite reconfiguration — and what the older cases can teach about the newer ones we cannot yet see clearly.
The layer below the fight
Trace enough of these episodes and a regularity surfaces that is strong enough to state as a near-law, if only so it can be attacked: good innovation rarely abolishes elite rent. It moves the layer at which rent is extracted.
The musket freed the peasant-soldier from the knight and, in the same motion, built the centralized fiscal-military state — the drill sergeants, the logistics tail, the treasury that could field and feed massed musketeers. Cheap violence at the bottom, expensive coordination at the top. The printing press freed the reader from the scribe and the clerical monopoly on interpretation, and then, over the following centuries, built the propaganda ministry and the mass-media apparatus. Cheap reading at the bottom, expensive persuasion at the top. The pattern is almost mechanical: the democratizing innovation dissolves a chokepoint elite at one layer while a new, often stronger, concentration forms one layer up.
This is what the public debate keeps getting wrong. The public good is real — but it is real at the old layer, where the friction used to sit. The danger, when there is one, forms at the layer above, where the new concentration is quietly assembling while everyone argues about the old one. You can love what the innovation does for the ordinary person and still have missed entirely where the power went.
One correction to the geometry, because “one layer up” is too tidy. The new rent does not always form directly above. It forms at whatever complement the innovation makes newly scarce — and that can sit above, below, or off to the side. Free operating systems did not hand the rent to a higher layer of software; they pushed it sideways and down, into the physical infrastructure that free-and-infinitely-copyable code suddenly made decisive — the data centers, the silicon supply, the handful of firms that own the cloud. Cheap generic drugs move the rent not up from the molecule but out to distribution and the ability to navigate the regulator. The rule is not “rent rises.” It is: rent reallocates to the scarcest complement the innovation induces, wherever that turns out to be. The fight you can see is a pointer to that complement, not to a fixed floor above the old one.
The claim is concrete enough to be wrong, which is the only kind worth making — but it has to be guarded against its own elasticity, because if every new coordinating body counts as a chokepoint, the law quietly becomes unfalsifiable. So the guard: only access-gating concentrations count — a complement someone can stand on and charge for, or shut. A municipal water authority that appears after the sanitation revolution coordinates and provisions; it does not gate access to extract rent, and it does not count. The falsifier, tightened: name a major innovation that abolished an elite’s rent and induced no new gating scarcity anywhere — a pure democratization whose induced complements are all competitive or provisioning, none of them a perch. I have looked and cannot find a clean one. If it exists, the law weakens to a tendency. And a softer form is already visible in the cases: sometimes the new concentration is plural rather than monopolistic — the contraceptive pill moved rent up into pharmaceutical and medical licensing but diffusely, across many competing actors, choking nothing. That the reconcentration has a shape — monopolistic or plural, gating or provisioning — is not a hole in the law. It is the very thing the next section needs to measure.
The same event, two directions
It helps to notice that innovations sort loosely into two families. Some are democratizing at their core — iron, the musket as an individual weapon, print as an act of reading, the five-hundred-dollar drone that can kill a multimillion-dollar tank. Cheap beats expensive; a chokepoint dissolves. Others are concentrating at their core — bronze, nuclear weapons, orbital launch — because the innovation creates a new controllable chokepoint and mints the elite that holds it. Space belongs in this second family, not the first; it is a bronze-pattern technology, rare strategic control accruing to a handful of actors, however much it wears the costume of frontier and adventure.
But the deepest cases do both at once, and the drone is the live example. At the level of the individual weapon it is strongly democratizing: precision strike loses its capital moat, and insurgents acquire a poor man’s air force. At the level of integration it re-concentrates hard — electronic warfare, autonomy stacks, sensor fusion, and a supply chain that runs through a few manufacturers a state can squeeze. The two are not in tension. They are the same event seen at two layers: democratize the trigger, re-concentrate the arsenal.
I first wanted to close this by saying the equilibrium favors whoever can regenerate capital fastest — the industrial state — because the drone converts war into a contest of capital attrition. That is one branch, and I think stating it as the equilibrium was a mistake, because it smuggles in the conventional-war frame and calls the smuggling a result. What the drone actually does is lower the cost of imposing attrition, and that cuts both ways at once. It lets the strong regenerate and grind. It also lets the weak bleed the strong cheaply — and the classic asymmetric move has never been to match industrial output but to refuse the attrition contest and turn the war into a contest of political will, where the cheap harasser is advantaged precisely because it can raise the other side’s costs faster than its own. So the drone does not settle whether the future favors mass or insurgency. It changes which game gets fought, and the winner is whoever can force their game on the other. That is a more honest claim and a more uncomfortable one, because it has no built-in victor.
It is still falsifiable, in both directions. If a non-state actor achieves a genuinely strategic decision against a peer-industrial state primarily through cheap drones — a war-ending result, not costly harassment — the “mass still wins” reading is dead. If every such contest keeps resolving in favor of the side with the deeper factory, the “will can refuse attrition” reading is dead. The current cases — real costs imposed, no wars won — discriminate between neither yet. So the honest status is not a settled equilibrium on probation. It is genuinely open, and I priced it as more settled than it is.
Where the benefit actually gets outweighed
This gives a usable answer to the hard question: not whether an innovation is good, but when its public benefit gets outweighed by the concentration it enables. The answer is separability.
The dangerous cases are precisely the ones where the public good is load-bearing for the concentration — where you cannot deliver the benefit through any decentralized substrate, so getting the good requires accepting the new chokepoint. Nuclear power is only weakly separable: the civilian fuel cycle and the weapons capability share too much machinery to fully pull apart. Autonomous vehicles, by contrast, look highly separable. You can mandate open crash-data reporting, standardize no-fault insurance so victims are paid without proving a defect, and keep the discovery process public — and get almost all of the safety with almost none of the concentration. The good detaches from the danger cleanly.
So the decision rule writes itself. If the good is separable from the concentration, you are in a rent-dissipation fight: back the innovation, and mandate the decoupling as the price of admission. If the good is inseparable, the naive scoring — progress good, incumbents bad — is exactly where the danger hides, because you cannot take the benefit without buying the concentration wholesale.
The quiet virtue of this rule is that it sidesteps a question no one can answer. “Is this net progress?” is unanswerable in its own terms, because the innovation reshapes the very values you would measure it against — the musket did not serve pre-existing preferences, it produced the centralized state that then defined what counted as order. There is no view from nowhere to net it out. Separability is a local question — is this particular good load-bearing for this particular concentration — and local questions have answers. You dodge the unanswerable by asking the answerable.
But separability is not a fixed property of the technology, and treating it as one is the rule’s most dangerous simplification. It is largely manufactured — set by design choices that are still up for grabs while the technology is young. Autonomous vehicles are separable only if the crash data is required to be open; let the manufacturers keep their edge-case corpora and evaluation pipelines as trade secrets and the safety signal goes proprietary, and separability collapses. Nuclear power runs the other way: fuel leasing, multinational enrichment control, and continuous monitoring have partially engineered a separability the raw physics does not grant. So the decision rule has to be read in the active voice. You are not waiting to discover whether the good detaches from the concentration; you are fighting over whether it will be built to detach, through the standards, the liability rules, the data-access mandates. “Move before the fortress sets” is not advice to seize a window that happens to be open. The window is not found; it is poured. The mandate is not a response to separability — the mandate is the separability, and losing the mandate is how a separable technology becomes an inseparable one.
The trap of the legible number
Now the practical problem: whoever writes the rules almost never writes them in terms of the thing they actually care about. The autonomous-vehicle bill carves out heavy trucks. The drone regulation sets a payload weight above which the rules bite. Neither weight tracks the stated goal. A truck’s tonnage does not track road safety; it tracks the number of driving jobs at stake. A drone’s payload does not track lethality; a two-pound charge in the right place does plenty, and the threshold tracks what an agency can actually enforce at a weigh station or a registration desk.
This is the deep insight of James Scott’s Seeing Like a State, and it is worth stating sharply because it does most of the work. A state can only act on variables it can see and enforce at scale — legible variables. Weight is legible: a number on a scale, a line on a spec sheet. Lethality and safety are not legible; they are contextual, emergent, and adversarially mutable. So the regulator does not choose weight because weight is a good proxy. The regulator chooses weight because it is structurally blind to the thing it cares about and can only grasp the legible neighbor.
And here is a fact about which states suffer this blindness, because it is not all of them, and it turns out to matter enormously for what comes later. A state that governs through status orders rather than formal rules does not need a legible proxy at all. It can prohibit a whole class of people from touching a whole class of thing, and enforce that prohibition capillary, house by house, through the standing hierarchy. The legible-proxy trap is the specific tax paid by the liberal administrative state — the one committed to formal equality, to governing by rule rather than by rank. Having renounced status-based prohibition, it is forced to find a measurable proxy for what it wants, and a measurable proxy is always a gameable one. The weight threshold is not evidence that states are dumb. It is evidence that a state which treats its citizens as formally equal has given up the one instrument that could have banned the thing directly. Hold onto that; it is why the oldest move against a democratizing technology is no longer available to us.
Here is the part Scott left implicit. The gap between the legible proxy and the real target is a single opening, and it gets worked from both sides at once. The rent-seeker lobbies to place the carve-out boundary on the legible axis, then sits safely on the protected side — the teamsters live in the weight gap. The threat routes around the legible axis, because the real danger isn’t tracked there — the modified sub-threshold drone lives in the same weight gap. Same gap, two occupants, opposite intentions. Whenever a rule’s dividing line sits on an axis that doesn’t track its stated purpose, it is tracking an unstated one: a group to protect, or an agency’s convenience. The chosen discriminator reveals the true purpose. Read the axis, and you read the intent.
But legibility is not the villain of the story, and this is where the autonomous-vehicle case turns instructive. The insurance-based reform the technologists propose — a single insurer covers the vehicle, pays the victim fast, then recovers from the manufacturer if the software was at fault — is a legibility upgrade, run in the beneficial direction. Litigation adjudicates an illegible question one case at a time: was this specific defect, in this specific jury’s eyes, negligent? The result is a noisy, lagged, lottery-like signal. Experience-rated insurance makes a fleet’s safety record legible as a continuous price — the insurer sees the incident rate across the whole fleet and prices it, becoming a standing safety regulator rather than an after-the-fact one. That is the same axis Scott made famous, running toward accountability instead of toward the crushing of local knowledge. Legibility is the villain when it displaces a functioning form of tacit knowledge or distributed judgment. It is the hero when it prices a harm no one was pricing before. The trick is telling which one you’re doing — and often, as here, it’s both at once.
The fence around the courthouse
Which surfaces the cost the technologists’ proposal quietly pays, and the general reader should see it plainly because it generalizes far beyond cars. Adversarial litigation is not only rent extraction. It is also a decentralized, discovery-forcing accountability mechanism — a machine that drags a manufacturer’s internal documents into daylight, one lawsuit at a time, run by thousands of independent operators no single authority controls. The rent and the accountability function are bundled. You frequently cannot dissolve the rent without dissolving the information-forcing that rode on it. Replacing thousands of juries and plaintiff’s lawyers with quiet settlement between an insurer and a manufacturer — two repeat players who deal with each other for decades — trades a transparent-but-noisy system for a quiet-but-opaque one. That may well be a good trade. But it is a trade, not a free lunch, and it is the layer-shift law operating on accountability rather than on money: the discovery function doesn’t vanish, it relocates behind a closed door.
Push on the word bundled, though, because it flatters the reform. Bundling implies a clean seam — that the rent is a wrapper you can peel off a pristine accountability core. It isn’t. The adversarial system is a particular kind of accountability: public, adversarial sunlight, the internal memo dragged into a courtroom where anyone can read it. That kind does not survive translation into an insurer’s risk model. The experience-rated insurer produces a genuine safety signal, but it is accountable to its own actuarial tables, not to a public — and its competitive advantage is the opacity, the proprietary model it has no reason to publish. So the honest description is not that the discovery function relocates behind a closed door and might be re-opened. It is that one species of accountability is being traded for another species that is better at pricing and worse at disclosure, and there is no version of the swap that keeps both. The question is never whether the accountability is preserved. It is which loss you can live with.
This is a Chesterton’s Fence about power structures — before removing the messy old mechanism, understand what function it was quietly performing. And its sharpest form is not about cars at all. Qualified immunity removes the identical mechanism — the civil suit — from the police, without installing any replacement: no insurer pricing the risk, no experience rating, no continuous safety signal. It is the technologists’ proposal with the accountability stripped and nothing put in its place — not even the actuarial kind. That is the difference between a hard trade and a pure loss, and it is the whole distance between a reform worth arguing about and a giveaway.
Why some elites shatter and others just move
One more distinction earns its place, because it explains why some of these transitions are smooth and some are catastrophic. There are two different things going wrong when an elite faces a substrate shift, and they are usually confused.
One is a legitimacy problem: can the group’s authority survive admitting that the world has changed? The other is a capability problem: can the group still enforce its will once the substrate is gone? These are independent — until they fuse. And they fuse under exactly one condition: when the incumbent’s legitimacy was its enforcement monopoly. The knight’s honor was not separable from his monopoly on effective violence; the two were the same fact. So when the musket dissolved the capability, it dissolved the legitimacy in the same stroke, and the knight could neither admit the change (that surrenders the honor) nor enforce against it (the capability is gone). That double bind is why some elites don’t migrate — they shatter. Sparta is the civilizational-scale case: a warrior identity fused to a battlefield monopoly, unable to acknowledge or adapt, breaking all at once at Leuctra rather than bending.
Trial lawyers and craft guilds migrate smoothly precisely because their legitimacy was never fused to a monopoly on force. They can lose the old rent and reattach to a new one because losing it costs them income, not identity.
The claim needs one relocation to survive its obvious counterexamples, though — and making it precise is what makes it useful rather than merely evocative. What shatters is not the people; it is the institutional vehicle. Individual knights became pistol-armed cavalry and officers of the new armies; individual samurai became prefectural bureaucrats and the founders of industrial houses. Point to those survivors and the shatter/migrate line looks false. But the thing that broke was the feudal host and the samurai estate as organizational forms of elite power — and those broke completely, the samurai class violently, in a rebellion put down by the conscript army that had replaced it. The Meiji state even tried to buy the migration, converting hereditary stipends into bonds to turn a warrior caste into a financial one, and the class shattered anyway while a fraction of its members walked out as individuals. So the test is about the vehicle, not the bloodline: when a group’s legitimacy is fused to an enforcement monopoly, the institution cannot survive the loss of the monopoly, however many of its members climb out into the successor order. It also helps to remember the loss is rarely instantaneous — the musket coexisted with heavy cavalry for well over a century before the balance tipped — so “shatter” names the eventual death of the form, not a single battle.
The test still discriminates, and it can still be falsified: find an institution whose legitimacy genuinely was its enforcement monopoly and which nonetheless carried that same institutional form intact through the loss of the monopoly, rather than dissolving and scattering its members into a successor. The feudal host, the samurai estate, the clerical interpretive monopolies all dissolve. The plaintiff’s bar and the guilds carry their form across. So far the seam holds.
The elite sometimes wins
A caution against the whiff of inevitability in all of this. Suppression is a real terminal outcome, not merely a delay. Tokugawa Japan deliberately restricted the spread of firearms for roughly two and a half centuries, specifically to preserve the samurai order — and it held, until Perry’s ships arrived from outside and ended the arrangement at gunpoint. (The popular claim that Japan “gave up the gun” entirely is overstated; guns remained in use. What is solid is the direction: a deliberate restriction of a democratizing military technology to protect an elite, sustained for a very long time.) The lesson is that elites facing reconfiguration do not always lose. Sometimes they win, durably, by preventing the substrate shift from proliferating — and the thing that eventually breaks the arrangement is usually an exogenous shock, not internal reform.
But look at how the shogunate did it, because it closes the loop with the legible-number trap. Tokugawa did not set a weight threshold. It did not regulate an attribute of the gun at all. It reached for status directly — it disarmed the peasantry as a class and reserved arms to the samurai estate, and it had the capillary reach to enforce a status prohibition across an archipelago. That is precisely the instrument a rule-of-law state has given up. This is why the modern weight thresholds on drones are not just a worse discriminator than Tokugawa’s — they are a different kind of thing, the best a formally-equal state can manage once it has renounced the ability to say “these people, and only these people, may hold this.” The pre-modern state could suppress a democratizing technology because it did not need a legible proxy. The modern one is stuck fighting through a gameable attribute because it does. The feebleness of the weight threshold is not a failure of nerve. It is the shadow cast by rule-of-law onto the problem of suppression.
But it would be too quick to conclude that suppression is therefore off the table, and catching that error is the last turn of the argument. Law gave up status prohibition. Culture never did. Taste, prestige, and shame still enforce rules of the exact forbidden shape — these people may not do this thing — capillary-fine, house by house, with no legislature and no gameable threshold, in the one register where formal equality does not reach. The knights themselves reached for it: unable to legislate the gun away, their culture branded it dishonorable, the coward’s weapon, villainous saltpetre — a status prohibition on a leveling technology, run through contempt instead of statute. It failed there, against gunpowder. It does not always fail. The instrument the modern state renounced did not die; it migrated to the register where it had always worked best. And that migration predicts where the fight goes: when an incumbent class cannot build a legal wall against a leveling technology, it builds a cultural one, and the struggle surfaces as a war over legitimacy and taste rather than as a law — a war whose real subject is who is permitted to produce, wearing the mask of a quarrel about quality. The weight thresholds are only the visible, legal, feeble line. The real prohibition, when there is one, re-forms quietly in the domain of status, and is easy to mistake for mere snobbery precisely because status is exactly what it is enforcing.
What the frame does and doesn’t do
None of this tells you whether a given innovation is, on balance, good. It refuses to. The refusal is the point.
Progress and devolution are, most of the time, the same event seen from two layers. The knight experienced the musket as pure devolution — the collapse of a world in which excellence and training meant survival. The centralizing king experienced it as pure progress — the arrival of an army he could actually afford to command. Both were right. There is no higher layer from which to net their accounts against each other, because the innovation dissolved the shared framework that would have done the netting. Anyone who tells you they have scored the net is standing at one of the two layers and mistaking it for the view from above.
So the honest frame does something narrower and more useful than scoring. It locates who experiences which, and where the new concentration is forming while the old fight absorbs everyone’s attention. Scoring itself is off the table, because there is no view from nowhere to score from. And what the frame leaves standing is not one instrument but two, held by opposite sides. The incumbents keep the cultural prohibition — the status boundary that law can no longer draw but taste still can — and it defends the guild rather than the public, which is why so much of the loudest fighting is a war over legitimacy that only looks like a war over quality. The public keeps the separability mandate — and separability, recall, is not found but poured: while the good can still be built to detach from the concentration, require the detachment — open the data, socialize the insurance, keep the discovery public — and you get the benefit without buying the fortress. Those are the two live moves. Almost everything else, the frame can only diagnose.
Which is exactly why the mandate’s limit has to be stated in the same breath: it serves the public only where a regulator exists with the power and the legibility to impose it. Where the concentrating party has already captured the rule-writers, or where — as in the qualified-immunity case — the state is itself the incumbent whose enforcement monopoly is fused to its legitimacy, the public’s one move is unavailable to the very people who most need it, while the incumbents’ move needs no regulator at all. The frame can then name the danger with total precision and do nothing about it. That is not a flaw in the frame. It is the frame reporting, accurately, that by the time you can see the fortress clearly it is often already built.
Because the separability window does not stay open. It opens when the innovation arrives and the new concentration is still forming, and it shuts once the fortress sets and the people who built it are writing the rules about who may inspect it. The question was never whether to let the friction go. It was whether you moved before the fortress set.
cafebedouin.org 