When It's All Bathwater

It started, as these sorts of discussions do, with a conversation about cameras around our respective towns. Flock Safety had been pitching automated license plate readers to local governments up and down the Bay Area as well as across the country. I, living somewhat south of the Bay, expressed concern about the proliferation of this sort of tech, having seen at least a dozen or so cameras on my way to my friend’s house that night. He suggested that the technology itself was neither good nor bad, that the question was simply whether our elected officials and government departments would use it responsibly. After all, technologies seem to be capable of both help and harm depending on how they’re used. I found myself in the uncomfortable position of having to explain that this framing, while perhaps intuitive to him, was already a concession underneath a truism, that the argument had been lost the moment we accepted the premise that a networked, mass-surveillance infrastructure capable of tracking every vehicle’s movements across multiple jurisdictions was a neutral tool just awaiting proper application by authorities.

This essay is an attempt to work through why that framing is wrong, how the history of technology keeps proving it wrong, and why the people who profit from the framing are deeply invested in keeping us blind to the sleight of hand.

Can Technology Embody Values?

There is a distinction that gets overlooked, if not outright bulldozed, in many public conversations about technology, ethics, and politics, and erasing that confuses the debate and, ultimately, serves the interests of the powerful. The distinction is between two claims:

1. Technologies cannot have or embody values (moral, political, or otherwise). They are brute objects, like rocks or hurricanes, and talking about their values is a category mistake. Sure, technologies might be used in service to one set of interests or another, but talking about the “values of a technology” is, at best, a sloppy reference to its application or circumstance.

2. Technology has a neutral value. It has no particular valence of its own but takes on the valences of its situation. A technology is good when it is applied as intended for a good effect. A technology is bad when it is used harmfully or made poorly.¹

Some folks who defend a more optimistic technological progressivism against historically-informed critique slide between these two claims as if they were the same. They are not. These claims are both false, but they are false for different reasons, and equivocating between them hides the gap between the question and the response: a defense of the first claim gets passed off as an answer to an objection aimed at the second. The second claim is doing the real political work here, while the first is a philosophically substantive claim that, in this context, is a red herring.

My friend’s argument that night was a clean example of the slide. A camera is just an object; objects can’t be good or bad, they can only be used for good or bad; therefore the main question about adopting Flock cameras in communities is whether people, e.g. our officials, use it responsibly for our benefit. That is, Flock cameras can’t embody any values at all (moral, political, etc.) therefore they are value neutral in themselves, taking on the valences of their beneficial or harmful applications. The metaphysical shrug between the first clause and the second does the work of the political conclusion at the end, and the handoff happens so smoothly that no one really notices it.

The claim that technology cannot embody values at all is a substantive one that, perhaps to many people, seems intuitive. After all, a rock or a hurricane don’t “value” anything, so how could they embody values? How could a camera sitting on a pole? But this “how” is a question framed as an objection and, insofar as it is a question is adequately addressed by an account. As an objection, however, it incorrectly assumes that the only way values can be exemplified is by holding those values, perhaps as beliefs. Nobody credible in these debates thinks that rocks believe anything, so what is actually going on here?

When folks are talking about values embodied in an object or a system, especially as it relates to technology, they’re talking about a reification or materialization of a system of relationships that have shaped and designed that thing to bring it about and situate it for a context. Apple’s values for futurism, simplicity, prestige, etc. are not just held in the heads of executives and marketers, they’re read directly from the iPhone’s patented, and ferociously litigated, beveled corners. Langdon Winner argues that the atomic bomb is an example of a technology that can only exist within an authoritarian social structure with a centralized chain of command, since a democratized command apparatus would undermine the operational security purported by such a device in the first place. In Winner’s analysis, for example, political values can be imparted to a technology by contingent design choices in its engineering that reinforce certain sociopolitical interests over others, especially if more equitable alternatives exist, or by being a category of technology that, by nature, cannot operate safely or effectively without a specific social structure (Winner 1980).

With this in mind, we can understand that one reason a rock does not embody values is because it was not designed. A hammer, however, might. A hammer is shaped the way it is because someone wanted to drive nails. The shape of the head, the heft, the grip… each encodes various assumptions: About what a hand is. What a nail is. What work even looks like. These are not explicit values, but they are implicit ones. The hammer assumes a world in which certain things need fastening, in which force applied in a particular direction is an appropriate response, and in which the person holding it has the physical capacity to swing it. Change the person’s body, change the material, change the social context, and the hammer becomes a different object with different implicit relationships to the world around it. ²

We understand this on a crude level when we talk about what a technology is for. When a technology is used contrary to its intended purpose, we say it is being “abused”, and this language reveals more than it seems. I’ll come back to that.

This is obvious with a hammer, and it’s as true about other, more sophisticated technology. Indeed it should be surprising to think that more social context and conditioning should somehow reduce the values implicit in a thing. Yet exactly this becomes controversial when we scale up to software, to infrastructure, to systems that process millions of data points and make decisions about people’s lives, even though the principle is the same. A system designed to track license plates and share that data across jurisdictions, to flag “suspicious” patterns, to integrate with law enforcement databases, is not a hammer, clearly. But, as with technology generally, it is a specific kind of social relation made material. For a Flock camera, it encodes assumptions about what constitutes suspicion, about who has the right to monitor whom, about the proper relationship between the state and the individual, about which populations are risks to be managed and which are citizens to be protected. These are not abstract considerations: they are built into the architecture of the system from start to finish.

To see how, consider what a Flock camera actually does, mechanically. It is not merely a camera. It is a network node with a particular set of defaults. It is always on. It photographs every license plate that passes, not just the ones connected to a known crime. It converts those plates into structured data, tags them with GPS coordinates and timestamps, and transmits them to a centralized database designed for sharing across jurisdictions: city police, county sheriffs, state patrol, and federal agencies. The platform maintains “hot lists” of flagged plates, and the criteria for flagging are opaque, subject to change, and often exempt from public records laws. The system does not ask whether surveillance is appropriate in a given moment; it assumes surveillance is the baseline, and the only question is whether a particular vehicle is on a list.

These are design choices. Someone decided that always-on was better than triggered-by-suspicion. Someone decided that sharing across agencies was a feature, not a bug. Someone decided that the everyday subject of the system is a potential suspect, and that innocence is something you prove by not being on a list. Those decisions are not neutral. They constitute a political theory: the state has a legitimate interest in tracking the movements of the innocent, privacy is a burden on security, and institutional friction between agencies is an obstacle to be overcome rather than a safeguard to be preserved. The camera does not merely record reality: it constructs a particular kind of reality, one in which mass monitoring is the water we swim in. This is true even if the Flock camera remains in its box – these political choices are writ into the being of the camera itself.

So, not only can technologies embody values generally, they often embody particular political values and incentives that are far from being determined merely by their post hoc applications; they enable and incentivize particular sociopolitical realities over others.

This does not mean that the values embodied in a technology determine how it will be used. A fascist can use Signal, a communication tool designed to resist state surveillance, for secure messaging. The values are in the design affordances, not the end-user’s intentions. But the fact that adversaries sometimes use tools designed against them does not make those tools neutral. It just means that its embodied values are not its destiny.

There is also an asymmetry here. Signal’s architecture resists centralized control. It was designed to make mass surveillance difficult, and it does this regardless of who uses it. A fascist using Signal gains secure communication; they do not gain the capacity to surveil others, because the tool’s affordances resist that repurposing. Flock cameras and Palantir systems are different. Their architecture is designed to slot into existing state institutions such as police departments, immigration agencies, and intelligence services that already possess the legal authority, budget, and organizational capacity to conduct surveillance. The technology does not create the machinery of control on its own: it amplifies and streamlines machinery that already exists. ³ exists at all, its lethal properties demand that it be controlled by a centralized, rigidly hierarchical chain of command… The internal social system of the bomb must be authoritarian; there is no other way." That is, the bomb is strongly and inherently compatible with specifically authoritarian social structures and incompatible with democratic control. Marcuse put it more bluntly in One-Dimensional Man (1964): “the traditional notion of the ’neutrality’ of technology can no longer be maintained.” Technological rationality, he wrote, has become political rationality. Technologies that reinforce existing concentrations of power are naturally stabilized by that sociopolitical context, whereas liberatory technologies are always opposed by such existing powers, practically by definition.]

The Evidence of History

All of this is abstract until you look at the history. And since we’re talking about cameras, let’s start with a history of modern cameras and related imaging systems and evaluate the plausibility of intrinsic technological neutrality as played out in the real world, with real people.

Kodak’s Shirley cards, the reference images used to calibrate color film in the 1950s, were designed for white skin. For decades, photographers trying to capture darker skin tones got muddy, indistinct results not because of some law of physics but because the chemistry and the calibration standards had been optimized for a particular range of melanin. The film could not “see” the shades and tones of Black skin properly. This was not a bug that emerged in application. It was a feature of the design, built into the emulsion chemistry and the testing process. While it’s almost certainly true that Kodak scientists weren’t sitting around explicitly trying to make a product that performed poorly on dark skin, it’s demonstrably the case that the ability to capture dark skin was deprioritized and the criterion for an acceptable product made no particular effort to ensure equitable performance on dark skin as for white skin. Kodak only changed its film when furniture and chocolate manufacturers made a sufficient stink that they could not accurately photograph their dark-colored products. The market for Black people as photographic subjects was not large enough to matter.

Early television cameras inherited this problem. The lighting standards, the gamma curves, the contrast ratios, all were optimized for the same reference point. Black faces disappeared into the background not because of how the technology was being used but because of how it was built. You could not “use it well” enough ⁴ to solve this problem without fundamentally rebuilding the technology.

Nikon’s anti-blink detection, a digital camera feature from 2009 designed to warn photographers when a subject had blinked, failed on Asian faces. The algorithm had been trained on a dataset that did not include enough variation in eye shape, and so consistently misidentified Asian subjects. Like Kodak, this was not misuse. It was use exactly as designed, on populations the designers had not bothered to include in their model of “people.” Modern facial recognition systems continue to have higher error rates on Black and Asian faces for similar structural reasons.

And lest we should think that this is a result of the complexity of the technology at hand, pulse oximeters, those little devices that clip onto your fingertip to measure blood oxygen using LEDs and photodetectors, overestimate oxygen saturation in darker skin. This was known in the medical literature for years before COVID-19, but the devices remained in widespread use in part because the FDA did not require racial equity testing. During the pandemic, Black patients had treatment delayed or were sent home with readings that said their oxygen was fine, when it was not. People died because a technology designed for white skin told doctors that Black patients were healthy.

These examples are not aberrations or unforeseeable exceptions: they are the systematic, predictable outputs of systems that reflect existing social inequalities, designed by teams that do not include the people those systems will be used against, deployed in contexts where the costs of error fall disproportionately on vulnerable and disempowered populations.

Exculpatory responses to historical examples like the above typically fall into a few categories: that the technology had unintended consequences, that “bad design” is to blame, not the technology itself, or that the technology was merely being misused or misapplied. Each of these aim to deflect any value assignment to the technology itself, to preserve the claim that technology really is innocently neutral despite appearances. Unfortunately, none of these responses are adequate on inspection. Let’s evaluate these together.

Deflections and Defenses: Unintended Consequences, Poor Design, and Abuse

Unintended Consequences

The gentlest version of this deflection is that the harm in a particular case was an unintended consequence. Kodak did not set out to make Black skin unphotographable. The Nikon engineers were not trying to insult Asian faces by failing to correctly detect them. The oximeter was not designed for the purpose of killing anyone. All true, and yet all beside the point. Intent is not the thing that determines whether a technology embodies values. A hammer’s designer need not consciously intend anything about disabled hands for the hammer to be built around a particular kind of body. What matters is what got prioritized, what got treated as the default case, and what got left out of the model of “people” the thing was built to serve. “Unintended” describes the designer’s state of mind. It says nothing about the object.

And there is a pattern to which consequences turn out to be “unintended.” They are reliably the ones that fall on populations the designers did not consult, did not include, and did not have to answer to. When the same category of person shows up on the losing end of decades of independent design decisions, across film chemistry and camera firmware and medical devices, “unintended” stops functioning as an exculpation and starts functioning as a description of whose interests were structurally invisible. A consequence that is unintended but perfectly predictable, and predictably borne by the same people every time, is not an accident. It is a value, expressed through neglect rather than malice. The technology still embodies it. The word “unintended” just tells you the designers were comfortable enough not to check.⁵

Poor Design

The next move concedes more ground: fine, the technology was bad, but that is a failure of design, not of the technology itself. The oximeter could have been calibrated for darker skin. The dataset could have included Asian eyes. Fix the design and the technology is redeemed. Its badness was contingent, a bug, not a property of the thing.

At first this sounds like an admission and concession, but it incoherently smuggles the neutrality thesis back in through the word “itself.” It imagines a pure Platonic oximeter floating above its actual instantiations, innocent of what every real oximeter did. But there is no technology “itself” apart from its designs. A technology is not an ideal form that gets imperfectly realized; it is the actual object, with actual calibration standards, actual training data, actual defaults, deployed in an actual world. To say “the design was bad but the technology is fine” is to point at an object that does not exist and call it the real one. The design is the technology. There is nothing else there.

And once you grant that, the defense collapses into the historicist’s own position. If the values live in the design, and the design is the technology, then the technology embodies values. Again, you have not rescued neutrality. You have just relocated the problem one word to the left and hoped no one would follow it. Worse, “poor design” implies the fix is a straightforward technical matter, a better dataset, a wider calibration range, when the reason those things were missing in the first place was a structure of incentives that made certain people not worth designing for. You can patch a given oximeter. You cannot patch the market logic that decided whose blood oxygen was worth measuring accurately, and that logic will produce the next biased device whether or not you fix this one.

Abuse

And, lastly, we have the defense that the technology was abused, or, at least that some form of “abuse” occurs that is exculpatory for the technology in question. What it means to abuse a technology can be broken into at least two apparent categories.

Design-contrary Abuse

We have technologies that are, knowingly or not, used against their (presumably benevolent or at least neutral) design. This covers the cases where, for example, a hammer is used to smash the head of one’s neighbor as opposed to striking a nail. Because the hammer was used contrary to its design, it was abused and should not be seen as an exemplar of any values in the hammer itself. Pointing to this as a defense has several problems.

First, the abuse defense concedes the point it is trying to refute. If a technology can be “abused,” then there must be some sense in which it can be “used well.” But “used well” and “abused” are normative categories. They presuppose that the technology has some proper function, some set of values it is supposed to serve. You cannot abuse a rock. You can only abuse a technology because technologies are designed for purposes, and purposes entail values. So the moment someone says “the technology was abused,” they have already admitted that technology embodies values.

Second, the abuse defense trivializes the objections people raise. When someone points out that facial recognition is less accurate on Black faces, that pulse oximeters give false readings on darker skin, that predictive policing algorithms send police back to the same over-policed neighborhoods, calling these problems “abuse” implies that the solution is simply to use the technology more carefully. It shifts the burden from the designers and the institutions that build, sell, and deploy these systems to the communities they harm. If only the police department had used it correctly. If only they had been the right kind of subjects. If only your body had matched the design assumptions.

Third, and most importantly, the abuse defense ignores the structural incentives built into the technology itself. Technologies do not emerge from nowhere. They are developed by organizations with particular interests, funded by capital with particular returns in mind, deployed by states with particular agendas. Flock Safety cameras are not just cameras. They are a business model built on data extraction. The company leases the hardware to police departments on subscription, a few thousand dollars per camera per year, because the real product is the data: license plate readings, vehicle patterns, associations between cars and locations, all fed into databases that can be queried, shared, sold, or used to train further systems. The incentive is not to protect communities. The incentive is to maximize the amount of data collected and the number of agencies with access to it.

This is the same move as “guns don’t kill people, people kill people.” The slogan is superficially true in a trivial sense: guns are inanimate objects, they do not have intentions. But the argument is deployed to block any discussion of whether the design of guns, the marketing of guns, the political economy of gun manufacture and distribution, the lobbying that prevents regulation, and the cultural narrative that equates firearm possession with freedom and masculinity, might have something to do with the rate at which people are shot. It reduces a structural problem to an individual moral failure and then declares the matter closed. The technology is innocent; only users are guilty. And since we cannot regulate users (freedom, you see) we must accept the carnage as the price of liberty.

The surveillance version works the same way. Flock cameras don’t surveil people, people surveil people. Palantir doesn’t deport immigrants, analysts deport immigrants. The system is innocent; only the operator is guilty. And since we cannot perfectly control every operator (bureaucracy, you see) we must accept the surveillance as the price of safety. The argument is not offered in good faith: it is offered to prevent the question from being asked.

This is why the “abuse” framing is so convenient for the companies that build these systems. It allows them to maintain the fiction that the technology is separable from the political economy that produced it. Accepting the attributions of history but devaluing them as an obstruction to technological progress is as obscurantist as it is self-serving.

Abusive Design

For completeness, we can consider the plausibility of technological neutrality in the most extreme case: technologies knowingly designed to cause harm. I presume readers here would not find it particularly contentious that technologies knowingly designed to cause harm are neither neutral nor being abused in their application. For those most curious, I’ve relegated such analysis to a footnote. ⁶ Suffice it to say that I take at least this case to be clear without argument for all but the most strident. For someone unfamiliar with Flock, Palantir, or other manufacturers of surveillance technologies, it may be unclear whether considerations of abusive design are relevant at all. However, we’ll discuss precisely what intentions these companies and their CEOs have for their surveillance technologies shortly.

The Ideology Behind the Infrastructure

This brings us to the people building these systems, because the values embodied in technology are not always implicit. Sometimes they are stated outright.

Alex Karp, the CEO of Palantir, has been remarkably candid about the political vision underlying his company’s work. In The Technological Republic, his 2025 book written with Nicholas Zamiska, and in his shareholder letters, Karp argues that Silicon Valley owes a “moral debt” to the American state and should rededicate itself to national purpose. He has described Palantir’s purpose in terms of “hard power” built on software, and in a 2025 earnings call, he stated that the company’s products exist to “scare our enemies and, on occasion, to kill them.” This is not the language of neutral tools. This is the language of someone who believes that technology should serve the projection of state power, and that the private sector owes its loyalty to the military and intelligence apparatus. The alternative, in his view, is nationalization or irrelevance.

Karp’s ideology is not hidden. He quotes St. Augustine to justify his service to the state. ⁷ He warns that AI companies who refuse defense work will be left behind. He presents Palantir as an “anti-surveillance company” while its platforms are used to conduct mass surveillance by ICE, the LAPD, and agencies across the world. The contradiction is only apparent if you accept his framing. If you understand that the company is against surveillance by competitors and in favor of surveillance by its clients, the contradiction resolves into a coherent market position.

Peter Thiel, the co-founder of Palantir and chairman of its board, has been even more explicit. In a 2009 essay for the Cato Institute, Thiel wrote: “I no longer believe that freedom and democracy are compatible.” He called “capitalist democracy” an oxymoron and argued that the extension of the franchise to women and welfare beneficiaries had made capitalism harder to sustain. This is not a neutral observer of technology. This is someone who believes that democratic governance is an obstacle to the kind of society he wants to build, a society in which technological elites are freed from the constraints of popular will.

Thiel has described surveillance AI as “communist totalitarian technology” while simultaneously building the infrastructure for it through Palantir and Anduril. The apparent contradiction is resolved if you understand that his objection is not to surveillance per se but to surveillance by the wrong people. Surveillance by a technocratic elite, wedded to state power and shielded from democratic accountability, is not totalitarian in his framework. It is necessary. It is the architecture of the world he is trying to build.

These are not fringe figures. Karp runs the company; Thiel chairs its board. I don’t claim a one-to-one map from either man’s beliefs to particular lines of code or to this-or-that bolt or spring, but their public statements jointly describe the political orientation of the enterprise, and the architecture tracks it. Palantir is a multi-billion-dollar company with contracts across the federal government and with police departments worldwide. Flock Safety, the company installing license plate readers in our neighborhoods, operates in the same ecosystem. Its cameras feed into databases that Palantir and other analytics companies can access. The infrastructure of surveillance is not a collection of independent tools. It is a network, and the people building it have been clear about what they want it to do.

Karp’s vision of “hard power” built on software is not a metaphor. Palantir’s platforms are architecturally designed for data fusion: breaking down the institutional barriers between police departments, immigration agencies, intelligence services, and military units. In a world with stronger democratic safeguards, those barriers are features. They create friction. They prevent the kind of total information awareness that makes authoritarian control efficient. Palantir’s technical achievement is to remove that friction, to make cross-agency surveillance seamless, and to turn the fragmented oversight of liberal democracy into a bug to be engineered around. The technology does not just reflect Karp’s ideology: it implements it.

Thiel’s belief that democracy and freedom are incompatible finds expression in his autonomous systems. Anduril builds drones and surveillance towers that reduce human decision-making in the loop. The political rationale is efficiency: machines are faster, cheaper, and less encumbered by rules of engagement or judicial oversight than human operators. But the technical choice to automate is itself a political choice. It is a choice to remove democratic accountability from the deployment of force, to make the use of surveillance and violence harder to see, harder to challenge, harder to subject to the slow, messy processes of popular will. The system does not just serve anti-democratic ends, it is designed to make democratic intervention difficult.

I am not claiming these systems work perfectly to achieve this vision or that it necessarily ever will. Thankfully they currently do not: Palantir is a buggy, expensive product like any enterprise software. Anduril’s drones crash. The point is not that the architecture flawlessly executes the designer’s vision. The point is that the vision is built into the design, and the design shapes what is possible. A system built to remove democratic friction will remove democratic friction even when it fails at everything else.

The Baby and the Bathwater

The argument that we should not “throw out the baby with the bathwater” when criticizing and opposing harmful technologies presumes that there is a baby, some essential core of the technology that is separable from its problematic applications. But this is precisely what is at issue. When the design of a system encodes surveillance logic, when its business model depends on data extraction, when its deployment patterns track racial and economic lines of inequality, there may not be a baby left once you drain the bathwater.

Take the Flock camera again. Drain out the cross-jurisdictional data sharing, the always-on collection of innocent plates, the opaque hot lists, and the data-extraction business model that leases the hardware on subscription because the readings are the real product. Take it all out. What is the neutral core that remains? A camera on a pole that photographs a passing car. Even if we were to grant neutrality for even that, which I don’t, nobody is selling that. There is no market for a license plate reader that does not read every plate, does not retain the data, does not share it, does not flag. The features you would have to remove to make it benign are the entire reason it exists. The product is the bathwater. Maybe someone could build a genuinely different tool: triggered-only, locally stored, unshared, auditable. That would be a different technology, with different values in its bones. It is not the one being bolted to poles in my neighborhood.

Don’t get me wrong. I’m not a technological determinist. I’m aware of the benefit various technologies bring to our lives. I also know technologies can be repurposed, resisted, regulated, and redesigned. But the fantasy is that we can simply step outside the history of a technology, evaluate it optimistically in some pure present moment, and decide to use it well. The belief that we can is itself a kind of naive ideology. It is the ideology of people who have likely never been the target of the systems they are evaluating, who have the luxury of treating history as an inconvenience rather than a structure of ongoing harm.

On Reductionism

A related charge is that this kind of historical analysis of surveillance technology is “reductionist”, that by refusing to evaluate this technology de novo, on its own terms, we are reducing it to nothing more than its past, stripping away its potential, treating it as fixed and determined rather than open and dynamic, perhaps despite my previous attestation.

This charge gets the direction of reduction exactly backwards, particularly in this case. It is not the historicist who reduces technology merely to its past. It is the ahistorical evaluator who reduces technology merely to its glassy-eyed potential, who treats the shiny promise of what the thing could do as if it were separable from the material reality of what it is and where it came from. The refusal to drop inconvenient historical context is not reductionist. It is the opposite. It is an insistence that the technology is more than its circuits, more than its marketing, more than the press release, more than whatever story gets told on a TED stage. It is an insistence that the technology is embedded in a history of power, profit, and harm, and that you cannot understand what it is without understanding what it has done.

Consider the pulse oximeter again. The ahistorical evaluator looks at it and sees a device that measures blood oxygen. Clean, useful, neutral. The historicist looks at the same device and sees something that measured some people’s blood oxygen accurately and gave others false readings that sent them home to die. Which of these is the reduction? The first description is not the richer one. It is the stunted one, stripped of the very facts that tell you what the device is and whom it serves. The historicist is not narrowing the object down to its worst moment; they are refusing to let the object be narrowed down to its promises.

A person who says “evaluate it on its own terms” is not asking for a fuller picture. They are asking for a narrower one. They are asking you to bracket the political economy, the design choices, the deployment patterns, the documented harms, and judge the thing as if it were an isolated artifact in a vacuum. This is not intellectual openness. It is a kind of willful blindness dressed up as neutrality. And it is no accident that this blindness consistently benefits the people who stand to profit from the technology being adopted.

To refuse this framing is not to deny that technologies change. It is to insist, though, that change does not happen in a vacuum: It happens through struggle, through resistance, through the very political contestation that the “evaluate it on its own terms” crowd wants to preempt. The technology is not reducible to its history, but it is certainly not reducible to its potential, either. The demand that we choose one reduction or the other is itself a deeply political move.

The Privilege of Ignoring History

This brings me to a recent discussion between Ezra Klein and Ta-Nehisi Coates on Klein’s podcast, prompted by Klein’s column praising Charlie Kirk as “practicing politics the right way” after Kirk was assassinated. Klein’s argument was one of grief and commonality: Kirk was murdered for participating in our politics, and something in that shared vulnerability should bond us across our divisions. He was showing up, engaging, doing the work of persuasion. Klein framed his admiration as strategic too: the left has lost badly, and we need to understand why Kirk was effective if we want to become competitive again. For Klein, Kirk’s appeal is a genuine mystery waiting to be exploited for a new liberal political strategy.

Coates was having none of it. He knew what Turning Point USA actually was, what Kirk had actually said about immigrants and trans people, what kind of movement he had built. Coates’s response was direct: “I believe hate is a powerful force. I believe it’s a powerful, powerful, powerful unifying force. And I think Charlie Kirk was a hate monger” (emphasis mine). He continued, “I’m all for unifying, I’m all for bridging gaps. But not at the expense of my neighbor’s humanity. I just can’t.” Coates correctly points out Klein’s naivety in underestimating hatred as a genuine political force, a mistake that uncovers a functionally callous detachment from history and from contemporary experiences of the very groups used by this power as its targets.

What struck me most was Coates’s historical framing and its effectiveness in cutting through Klein’s lionizing of Kirk’s politics. Coates wrote in Vanity Fair about the parallel to Reconstruction, how the explicit words of men who built an empire of slavery were ignored, how they were transfigured into gallant knights, how the country chose to bury the truth of the Confederacy in the name of some larger unity. There was a fatigue with Reconstruction, he noted, a desire to move on, to find common ground. And the cost of that common ground was paid by the people whose humanity got negotiated away in the bargain. ⁸

I think Coates is right, and I want to push the point further. What Klein sees as a contemporary mystery, as well as the fact that he sees Kirk’s success as a contemporary mystery, is directly a result of Klein’s ahistorical and anti-intellectual approach to the present. The bias in a pulse oximeter or a facial recognition system is not a matter of perspective. It is mathematically observable. The harm is objective. What is politically distributed is the incentive to look. The idea that one can step outside history, that one can evaluate a policy or a technology in a pure present moment untainted by what came before, is a form of epistemic privilege. It is available only to people for whom history has been a source of benefit rather than harm, for whom the structures of the past are foundations to build on rather than weights that drag them down.

For Klein, bringing history forward as Coates does is a kind of sandbagging. It’s a way of overburdening the present with the weight of a past that it cannot bear. He’s willing to admit all the terrible things about the past, perhaps suspiciously too much, in order to declare the present moment to be fundamentally different because, surely, now is not then. Klein’s position is an ironic admission that both sees the weight that history has and also an Oedipal wish oneself blind oneself to it, as if the same forces at work in the debates today were not the same forces at work then, as if these forces had not earned these weights for themselves.

This is the same move in the surveillance technology debate. In both cases, someone standing safely outside the blast radius of a history proposes to bracket that history, because attending to it interferes with a present project. Whether the project is national unity or technological adoption, the bracketing costs nothing for the person proposing it. The bill goes to someone else. “Let’s not throw out the baby with the bathwater.” “Let’s focus on what works now.” “Let’s not let the perfect be the enemy of the good.” These are all ways of saying: your historical grievances are getting in the way of my present convenience. They are ways of giving oneself permission to ignore a history that is not one’s own to ignore, because it gets in the way of what gets defined as progress.

Progress for whom?

The Cartesian ego, the disembodied mind that evaluates the world from nowhere, is not a universal human condition. It is a particular philosophical fantasy, and like most philosophical fantasies, it is more available to some people than others. Women have historically not been able to maintain the illusion of disembodied rationality in quite the same way, because their bodies have marked them, constrained them, reminded them of their corporeality in ways that men’s bodies have not. People of color, queer people, disabled people, the poor, anyone whose body is read as a problem by the dominant culture, knows that one cannot simply think one’s way out of the material conditions of one’s existence. ⁹

The fantasy of the neutral evaluator, the person who can stand outside history and judge technologies on their merits, is not just wrong. It is privileged, if not outright supremacist. It assumes that the perspective from nowhere, the absence of historical weight, is the same thing as objectivity and that the people who have benefited from the structures of the past are the ones best equipped to design the structures of the future.

Skepticism Is Not Sandbagging

I want to be clear about what I am not saying. I am not saying that all technology is bad. I am not saying that we should reject every innovation because it might have harmful applications. I am not saying that progress is impossible.

What I am saying is that skepticism based on history is not sandbagging. It is not a reflexive opposition to change. It is a necessary response to the fact that technologies are not neutral, that they encode the values and interests of the people who build them, that the harms they produce are often not accidental but structural, and that the people who benefit most from their deployment are the people least likely to experience those harms. History provides us with weight not to keep us from leaving the ground but to keep us stable in the sky.

When someone tells you that a surveillance technology is value neutral, they are asking you to ignore the history of surveillance, which is a history of control. When someone tells you that the question is only about proper use, they are asking you to trust the same institutions that have consistently used these tools against marginalized communities. When someone tells you not to throw out the baby with the bathwater, they are assuming that there is a baby, that the essential core of the technology is separable from the system of power that produced it.

Sometimes there is no baby. Sometimes the bathwater is the point.

I keep returning to the conversation about the Flock cameras. The person I was talking to is not a bad person. They were trying to be reasonable, to find a middle ground, to believe that our local government would use this technology responsibly. I understand the impulse. It is exhausting to maintain skepticism, to refuse the easy answers, to keep saying no when being sold the reasons for saying yes, or at least saying “maybe, with the right safeguards.”

But the right safeguards are structurally unlikely for a system designed to maximize surveillance. Oversight is hard to sustain when the company providing the technology has a financial incentive to expand its data collection. And “right use” is hard to even define when the technology itself encodes assumptions about which people count as suspicious, which populations pose a risk, and which communities deserve to be watched. The failure to recognize the risk that comes inherently from the values built into these systems leaves folks blind thinking that deployment is fine, we just need to shape the application of this neutral technology appropriately to contain and manage its downsides.

This is why I am skeptical of containment strategies. It is not that I believe regulation is impossible in principle, it’s that the specific architecture of surveillance technology makes meaningful regulation structurally unlikely. The data sharing agreements are opaque and often exempt from public records requests. The systems are sold for narrow purposes (e.g. finding stolen cars) but the data they collect is inevitably used for broader purposes because the data exists and the cost of querying it is zero. Once multiple jurisdictions adopt the same platform, opting out becomes politically difficult; it’s not just rejecting a vendor, it’s now cutting entire departments or agencies off from a regional data-sharing network that neighboring departments now depend on. The software is proprietary, which means civilian oversight bodies lack the technical capacity to audit what it actually does. And the vendor’s business model depends on expansion; every new city, every new camera, every new data-sharing agreement increases the value of the network. Regulation threatens that growth, so the vendor lobbies against it, often through front groups and police unions that present industry interests as public safety concerns. ¹⁰

Containment is not a strategy, it is a fantasy that allows us to feel moderate while the infrastructure expands around us.¹¹

This is not a technical problem to be worked on by shaping the regulatory landscape just so, to rearguard against foregone conclusions about progress and technologies designed to resist restraint. It is a political problem, and political problems are not solved by redesigning the surveillance architecture into a friendlier shape. Redesign fails for the same reason blocking a single camera fails: both accept the terms set by whatever keeps producing these devices, and the market simply routes around the patch and ships the next one.¹² Preventing one deployment matters only as a wedge, not as a finish line. What actually counts is going after the thing upstream of the devices, the incentives that make surveillance the default profitable answer in the first place. That is what I mean by organizing alternatives outside the system’s current logic, and by refusing the premises of the debate as it has been framed by the people who stand to profit from it.

The values embodied by technology are not abstract, they are a material reality, built into the infrastructure of our lives, shaping what is possible, deciding what is visible, determining who gets to be seen as fully human. The question is not whether technology can embody values - it does. As history shows, the real question is whose values it embodies, and whether we are willing to accept them.

Sources and Further Reading

On technology and values:

• Aristotle, De Anima
• Winner, Langdon. “Do Artifacts Have Politics?” Daedalus 109:1 (Winter 1980): 121-136.
• Marcuse, Herbert. One-Dimensional Man. Beacon Press, 1964.
• Schmitt, Carl. The Concept of the Political. The University of Chicago Press. 2007. Based on the 1932 edition of Der Begriff des Politischen published by Duncker & Humblot.

On the ideology behind the infrastructure:

• Karp, Alexander C., and Nicholas W. Zamiska. The Technological Republic: Hard Power, Soft Belief, and the Future of the West. Crown Currency, 2025.
• Karp, Alex. Letter to Shareholders. May 5, 2025.
• Karp, Alex. Palantir shareholder letters and public statements.
• Allyn, Bobby. “How Palantir, the secretive tech company, is rising in the Trump Era.” NPR. 3 May 2025.
• Palantir Technologies (PLTR) Q4 2024 Earnings Call Transcript. The Motley Fool Transcripts. 4 Feb 2025.
• PalantirTech Twitter Account. “The Technological Republic, in brief…”. 18 Apr 2026
• Thiel, Peter. “The Education of a Libertarian.” Cato Institute, 2009.
• Thiel, Peter. Various interviews and public statements on democracy, surveillance, and technology.

On racial bias in technology:

• Roth, Lorna. “Looking at Shirley, the Ultimate Norm: Colour Balance, Image Technologies, and Cognitive Equity.” Canadian Journal of Communication, 2009.
• Lewis, Sarah. “The Racial Bias Built Into Photography.” The New York Times, 25 Apr. 2019.
• “Are Face-Detection Cameras Racist?” Time Magazine, 2010.
• Sjoding, Michael W., et al. “Racial Bias in Pulse Oximetry Measurement.” New England Journal of Medicine, 2020.
• Fawzy, Ashraf, et al. “Racial and Ethnic Discrepancy in Pulse Oximetry and Delayed Identification of Treatment Eligibility Among Patients with COVID-19.” JAMA Internal Medicine, 2022.
• Buolamwini, Joy, and Timnit Gebru. “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.” Proceedings of Machine Learning Research, 2018.
• Angwin, Julia, et al. “Machine Bias.” ProPublica, 2016. (On COMPAS algorithm)

On history, progress, and whose progress:

• Klein, Ezra and Derek Thompson. “Abundance.” Simon & Schuster, 2025.
• Coates, Ta-Nehisi. “The Case for Reparations.” The Atlantic, 2014.
• Klein, Ezra, and Ta-Nehisi Coates. “Ta-Nehisi Coates on Bridging Gaps vs. Drawing Lines.” The Ezra Klein Show, New York Times, 28 Sept. 2025. Podcast.
• Newsom, Gavin and Charlie Kirk. “And, This is Charlie Kirk”. This is Gavin Newsom. Apple Podcasts. 06 Mar 2025.
• Newsom, Gavin and Steve Bannon. “And, This is Steve Bannon”. This is Gavin Newsom. Apple Podcasts. 10 Mar 2025.

On AI regulation and existential risk rhetoric:

• Future of Life Institute. “Pause Giant AI Experiments: An Open Letter.” March 22, 2023.
• Altman, Sam. Testimony before the Senate Judiciary Committee. May 16, 2023.
• Center for AI Safety. “Statement on AI Risk.” May 2023.
• Perrigo, Billy. “OpenAI Lobbied the E.U. to Water Down AI Regulation.” TIME, June 20, 2023.
• “OpenAI CEO’s Threat to Quit EU Draws Lawmaker Backlash.” Reuters, May 25, 2023.
• Hern, Alex. “What should the limits be?” The Guardian, June 7, 2023.
• Corporate Europe Observatory. “Big Tech Lobbying Is Derailing the AI Act.” November 2023.
• OpenSecrets. Federal Lobbying data for OpenAI, 2023-2024.

On Flock and surveillance infrastructure:

• Electronic Frontier Foundation. “Automated License Plate Readers (ALPRs).” Street-Level Surveillance project.
• 404 Media. Investigative reporting on Flock Safety data sharing and law enforcement use, 2024-2025.
• Various news reports on Flock Safety camera controversies in cities across the United States.