The seductive lie of frictionless intelligence
Artificial intelligence is usually introduced to the public as a productivity story: faster drafting, cheaper coding, better diagnoses, smoother logistics. That framing is true, but incomplete in the way that propaganda is incomplete. It captures the visible benefit and hides the structural consequence. AI is not merely a tool that helps humans work more efficiently. It is also a tool that helps institutions watch, sort, predict and discipline humans with unprecedented ease.
That is why the most consequential question about AI is not whether it will replace this or that profession, or whether it will produce a chatbot that sounds more charming than its competitors. The deeper issue is political. AI intensifies the age-old ambition of power to make society more legible, more measurable and more controllable. In the private sector, that ambition takes the form of surveillance capitalism: the extraction of behavioral data to infer, nudge and monetize human action. In the public sector, the same machinery can become a scaffold for digital authoritarianism, where states fuse algorithmic analysis with administrative power to monitor dissent, manage populations and narrow the space of private life.
The most provocative thing about AI is therefore not its intelligence. It is its administrative usefulness. AI does not need consciousness to reshape society. It only needs to be good at classification.
From prediction to power
Every industrial era promises liberation and delivers a new hierarchy. Steam reorganized labor. Electricity reorganized time. Software reorganized markets. AI is reorganizing prediction. It can estimate what a customer will buy, what a worker might miss, what a borrower may default on, what a driver is likely to do, and what an employee might be feeling. Once prediction becomes cheap, the temptation is irresistible: if a system can predict behavior, it can begin to shape it.
That logic is central to the critique of surveillance capitalism. Scholars and policymakers have warned that once large-scale data collection begins, it rarely stops, because the data become too valuable to relinquish and too useful to ignore. At a Harvard panel on surveillance capitalism’s threat to democracy, Suresh Venkatasubramanian argued that AI is a tool for predicting human behavior and that the only reliable response is to treat it as a dangerous technology requiring safety-minded regulation. The same discussion emphasized stronger privacy protections and public oversight of high-risk AI systems. The point is not abstract. It is that data accumulation creates institutional momentum, and institutional momentum has a politics.
Corporations do not collect data only to know customers better. They collect it to gain leverage. The more precisely a platform understands a person’s habits, anxieties and reflexes, the more effectively it can sell, nudge or exclude. What once looked like convenience—recommendation engines, location services, frictionless personalization—becomes a machine for behavioral influence. The user is not simply the consumer of the product. The user is the product being refined.
The workplace becomes a laboratory
Nowhere is this more visible than in the modern workplace. Employers once judged labor through direct observation, periodic reviews and crude metrics of output. AI changes the scale and texture of supervision. It can analyze keystrokes, voice tone, response times, route choices, facial expressions, productivity patterns and calendar activity. In warehouses, call centers and delivery networks, automated management can reduce workers to streams of quantifiable behavior. In offices, the same logic enters through software that measures attention, drafts performance scores and flags deviations from expected rhythms.
This is not simply automation in the old sense, where machines replace repetitive physical tasks. It is automation of judgment. A manager need not understand a worker to supervise one. An algorithm can infer “efficiency,” “engagement” or “risk” from data traces that workers cannot see and often cannot contest. That asymmetry matters because power becomes less accountable when it becomes statistical.
There is a seductive managerial argument here: if surveillance improves efficiency, and efficiency boosts competitiveness, then the technology is justified. But this is the moral logic of the spreadsheet, not of democratic life. A system that treats every person as an optimization problem will produce measurable gains while eroding the dignity that makes work tolerable. AI does not merely automate labor; it can automate suspicion.
Displacement is only the first-order effect
Public debate still gravitates toward job loss, as though the central drama were a contest between machines and workers. That frame is too narrow. Some jobs will disappear, others will be reshaped, and many will be degraded rather than eliminated. The more consequential transformation may be the redistribution of bargaining power. AI allows firms to centralize decision-making at the top while fragmenting human labor at the bottom. A smaller number of highly paid engineers, managers and owners may control a larger universe of monitored, modularized and interchangeable workers.
This is why the familiar reassurance—that technology creates new jobs eventually—is only partly relevant. New jobs do emerge. But they do not necessarily restore autonomy, income stability or social status. A society can replace a durable middle-class occupation with a precarious one and still claim it has “created jobs.” It has, but it has also altered the terms of citizenship.
That alteration is made worse when AI systems are deployed not just to replace labor but to classify people before they ever enter the labor market. Recruitment algorithms, credit scoring models, background-screening tools and automated credential filters can harden inequality under a veneer of objectivity. If the model says someone is a poor fit, who can prove otherwise? When the machine is wrong, the burden of appeal falls on the person who least understands the system.
Displacement, then, is not only economic. It is epistemic. AI changes who gets to define merit, risk and worth.
Surveillance capitalism’s quiet triumph
The greatest success of surveillance capitalism is not that it persuades people to click on ads. It is that it normalizes the extraction of human experience as a business model. The basic bargain of the digital age has become grotesquely lopsided: free services in exchange for exhaustive behavioral capture. The result is a market in prediction, where platforms profit from knowing not only what people have done, but what they may do next.
Harvard researchers and critics of the model argue that the underlying logic should not merely be regulated at the margins but challenged at its core. In one Harvard discussion, a speaker went further, calling for the abolition of the fundamental mechanisms of surveillance capitalism, starting with the secret, large-scale extraction of human data and its conversion into corporate asset. That language is intentionally severe because the system it describes is severe. It has normalized a world in which opacity is profitable and privacy is treated as a luxury good.
AI supercharges this regime by making inferences from data that are more invasive than the data themselves. A purchase record is ordinary. A prediction of emotional state, political susceptibility or health vulnerability is much more intimate. The leap from data collection to behavior manipulation is the defining move of the current era. People may consent to being tracked by clicking an unreadable terms-of-service box. They do not meaningfully consent to being shaped by models trained on every digital trace they leave behind.
“The line between observing people and governing them is becoming dangerously thin.”
Digital authoritarianism at scale
The authoritarian potential of AI is easy to see in the most obvious cases: facial recognition at protests, predictive policing, automated censorship, social scoring, mass data fusion across ministries and security agencies. But the larger danger is subtler. AI lowers the cost of coercion. A state no longer needs to terrorize everyone all the time when it can monitor everyone cheaply and selectively punish the few who matter.
This is the great advantage of digital authoritarianism: it can be less theatrical than older forms of repression. The citizen may not notice the pressure until the application is denied, the travel record is flagged, the post is throttled, the interview never arrives or the knock on the door comes after a pattern of online dissent. In such systems, control works best when it is ambient. People self-censor not because they are ordered to, but because the algorithm makes risk feel personal and inescapable.
Democracies are not immune. The difference is one of degree, not kind. A government that can assemble data from public records, private platforms, telecom networks, workplace sensors and consumer apps possesses an intimidating capacity for social mapping. Even if the state does not abuse that capacity today, it may be inherited by one that does tomorrow. The infrastructure of surveillance is rarely designed for the political climate in which it is eventually used.
This is why debates about AI governance cannot be reduced to model accuracy, benchmark scores or technical alignment. The issue is institutional design. Who owns the data? Who can access the model? Who audits the output? Who is harmed when the prediction becomes a decision? Technical excellence without democratic controls is merely efficient unfreedom.
The myth of neutral automation
AI is frequently described as neutral because it is mathematical. That is a category error. Math can be neutral; deployment is not. A model trained on past decisions inherits the assumptions embedded in those decisions. If a system is optimized for profit, it will favor profit. If it is optimized for state security, it will favor state security. If it is optimized for engagement, it will favor whatever keeps the user scrolling, regardless of civic cost.
This is why calls for “responsible AI” sometimes sound too timid. Responsibility without power is theater. The underlying incentives must change. If firms are rewarded for harvesting behavioral surplus, they will harvest it. If states are rewarded for administrative control, they will deploy it. Ethical guidelines can reduce some harms, but they cannot by themselves reverse an economic order built on extraction.
The harder truth is that convenience has become the moral camouflage of surveillance. Every frictionless feature—autofill, predictive text, face unlock, personalized feeds, smart assistants—feels benign in isolation. Yet each one extends the perimeter of data capture. The bargain is cumulative. People surrender fragments of privacy until the fragments amount to a system.
What would resistance look like?
The answer cannot be nostalgia. A pre-digital society is not available, and in many respects it would be less free, not more. The real task is to build institutions that can absorb the benefits of AI without permitting it to become a universal instrument of monitoring. That means restricting data collection rather than merely policing misuse. It means limiting automated management in the workplace. It means banning or sharply constraining high-risk applications such as mass facial recognition, opaque scoring systems and untargeted behavioral profiling. It means creating public-sector capacity to audit, contest and, when necessary, prohibit systems that are too dangerous to normalize.
It also means changing the economic model. So long as digital platforms are rewarded for capturing attention and behavior, they will continue to turn human experience into inventory. Privacy cannot survive as an individual preference when the surrounding market structure is built to defeat it. Regulation must therefore target incentives, not just disclosures. Transparency matters, but it is not enough if the underlying business model remains predatory.
The most important political question of AI is whether society will treat it as an inevitable wave or a governed technology. The first view is convenient for incumbents. It suggests that the only rational response is adaptation: reskill, relocate, optimize, comply. The second view is more demanding. It insists that some uses of AI are socially illegitimate even when they are profitable, and that democratic societies have the right to say so.
That may sound radical. It is actually the minimum condition of self-government. A society that cannot decide where machines may watch, where they may decide and where they must not enter is not technologically advanced. It is merely well-instrumented. The future of AI will not be defined by whether machines become human-like. It will be defined by whether humans remain able to keep machines from becoming a permanent form of rule.
