The revolution nobody advertised
The argument for artificial intelligence has settled into a familiar rhythm. First comes the promise: machines will automate drudgery, cure shortages, and free human beings for higher purposes. Then comes the warning: some jobs will disappear, many will change, and society must manage the transition. But this framework may be too narrow, and too generous. It imagines AI chiefly as an engine of production. In practice, one of its most immediate uses is control.
The deepest consequence of AI may not be that it replaces workers outright, though in many places it will. It may be that it allows employers, platforms, and governments to observe, classify, predict, and nudge human behavior at a scale previous technologies could only approximate. This is not merely automation. It is administration by machine. The more data AI consumes, the more it can infer: who is likely to quit, who is likely to steal, who is likely to organize, who is likely to buy, who is likely to dissent. In the language of the market, this is efficiency. In the language of politics, it is the architecture of compliance.
That is why the coming era of AI is best understood not as a duel between humans and machines, but as a consolidation of power. The winners will be those who own the models, the data, and the infrastructure that translates ordinary life into prediction. The losers may not just be workers whose tasks are automated. They may be citizens who discover that the spaces once thought private, casual, or forgotten have become legible to institutions that never used to see so much, so often, so cheaply.
From labor-saving tool to behavioral machine
Every major technology arrives with a morally flattering story about itself. The assembly line was supposed to make goods cheaper. The spreadsheet was supposed to make firms more efficient. The internet was supposed to democratize information. In each case, the promised benefits were real, but the distribution of power was never neutral. AI fits that pattern, only more sharply. It does not merely speed up decisions; it improves the machinery of judgment itself.
That matters because organizations do not use technology only to do more. They use it to know more. A warehouse does not deploy computer vision only to move boxes faster. It deploys it to track workers' pauses, posture, and pace. A retailer does not use predictive analytics merely to stock shelves. It uses it to infer which employees are underperforming before a manager has even noticed. A delivery platform does not just match riders to orders. It learns who will accept an assignment, who will cave to pressure, and how much it can squeeze before labor turns expensive. The point is not simply output. It is leverage.
This is why the prospect of AI-driven job displacement should not be treated as a discrete labor-market event, as if factories were shutting down one by one and workers could be neatly retrained. The more important shift may be subtler. AI enables what might be called partial automation of authority. Managers do not have to understand a system to rule through it. They only have to trust the dashboard. Once that happens, workers face not a boss but a scoring apparatus: attendance ratings, productivity rankings, risk scores, customer sentiment models, fraud flags. The human supervisor becomes a ceremonial figure in the middle of an increasingly automated hierarchy.
For many firms, this is irresistible. AI does not just cut costs. It lowers the price of suspicion. A company can monitor more behavior more continuously, and then call the result “optimization.” The same logic already governs the gig economy, warehouse logistics, and digital advertising. AI extends it into ordinary employment. The future of work may not be one in which everyone is replaced by software. It may be one in which everyone is kept, but watched more closely.
The age of surveillance capitalism has matured
The phrase “surveillance capitalism” once sounded theoretical, even overstated. It now sounds like plain description. Firms collect data from phones, cameras, apps, browsers, cars, doorbells, and connected devices. Much of this collection is unrelated to the service being sold. The data is aggregated, traded, enriched, and fed into models that can make unusually intimate predictions: what a person wants, fears, hides, or is about to do. What began as targeted advertising has evolved into a broader commercial logic of behavioral extraction.
AI makes that logic more powerful. If data was the raw material of surveillance capitalism, machine learning is its refinery. It can find patterns in scale where human analysts would see noise. It can fuse location histories, purchase records, device signals, and social graphs into a single probabilistic portrait. The result is not merely personalization. It is preemption. The system no longer waits for you to act. It tries to know in advance.
That changes the political economy of privacy. Privacy used to mean control over information. Now it increasingly means control over inference. Even when a person opts out of a service, changes settings, or declines permissions, the extraction often continues through adjacent data brokers, third-party trackers, or institutional partnerships. The individual is no longer a customer so much as an environment in which the system operates. The more AI improves, the more that environment becomes searchable.
This is where the job-displacement story and the surveillance story converge. The same tools that help firms automate labor also let them automate suspicion. Employers do not need perfect knowledge of their workers. They only need enough predictive power to discipline them. If an algorithm can identify who may quit, it can target retention. If it can identify who may unionize, it can escalate management. If it can infer who is stressed, isolated, or financially fragile, it can tailor pressure with unnerving precision. AI does not abolish human power. It makes it more granular.
“The miracle of AI is not that it sees everything,” one privacy scholar told me recently. “It is that institutions can now act as if they do.”
Why governments are tempted
The private sector did not invent this logic alone. Governments have long admired any technology that lowers the cost of monitoring populations. The difference today is that AI makes surveillance scalable enough to be bureaucratically seductive. Instead of hiring more investigators, states can buy more data. Instead of building a case from scratch, they can purchase behavioral traces collected elsewhere. Instead of waiting for a warrant, they can often locate commercially available information that would have been hard to obtain directly.
This raises a constitutional and democratic problem that is more serious than the usual privacy debate suggests. A government that can cheaply purchase location histories, browse histories, facial imagery, or network maps is not the same as a government that must ask courts and citizens before intruding. The practical result is a bypass of older legal safeguards. What the state may not always collect cleanly, it may sometimes acquire by proxy.
AI deepens the temptation because it converts abundance into plausibility. Raw data can be messy, but models make it actionable. A mass of signals becomes a list of suspects. A set of correlations becomes a risk assessment. A location trail becomes a story. And once an institution has a story, it tends to treat it as truth. That is how digital authoritarianism can spread without a single dramatic rupture. It arrives as a service improvement, a fraud-prevention tool, a public-safety partnership, a better way to allocate scarce attention.
There is an especially modern cruelty to this arrangement. Earlier forms of surveillance demanded expensive human labor and generated visible institutions. Today's systems are often ambient. Doorbell cameras, license-plate readers, workplace analytics, ad-tech identifiers, and social platforms all produce fragments of a shared record. People participate because the devices are useful, the convenience is real, and opting out is difficult or socially costly. The result is a crowdsourced observability regime. Everyone helps watch everyone else.
The labor market will be disciplined before it is transformed
Economists often ask whether AI will destroy jobs faster than it creates them. That is a necessary question, but not the most revealing one. A more instructive question is: what happens to bargaining power before employment levels fully adjust?
History suggests that labor markets do not wait politely for a technological transition to finish. They absorb shocks asymmetrically. In the first phase, management gets the new tools, while workers get uncertainty. Even where AI fails to replace jobs outright, it can weaken workers' ability to negotiate pay, schedules, and dignity. Employers can benchmark performance more aggressively. They can outsource routine judgment to models. They can fragment tasks so that each person becomes easier to substitute. The result is not always mass unemployment. Sometimes it is mass precariousness.
This matters because workers are not only producers; they are citizens. A labor market that is hyper-monitored trains people to accept asymmetric scrutiny as normal. It habituates them to machine assessment and constant availability. It makes surveillance feel like professionalism. Once that norm spreads, it can seep outward from the warehouse or call center into schools, hospitals, finance, logistics, and public services. AI does not need to eliminate employment to reshape social life. It only needs to make obedience feel efficient.
The irony is that the public debate often treats AI as if its main social consequence were technical unemployment. But the first-order effect may be political. A society in which workers are ranked, tracked, and algorithmically managed becomes one in which power itself is more opaque. When no one can explain a score but everyone must obey it, the language of accountability begins to rot.
The great bargain is getting worse
For two decades, the implicit bargain of the digital economy was this: users would surrender data in exchange for free services. That bargain was always lopsided, but at least it was legible. AI has made it more dangerous by expanding what can be extracted from what is surrendered. Data is no longer merely a record of past behavior. It is a predictor of future conduct, a proxy for psychology, and a weapon in markets and institutions.
Once data becomes predictive enough, the incentive to collect it becomes almost limitless. Why ask a question directly when a model can infer the answer? Why rely on consent when ambient tracking can do the work? Why preserve privacy when opacity is so profitable? These are not rhetorical questions for the industry. They are business strategy.
And yet the public debate remains oddly sentimental about choice. People are told to read privacy policies, tweak settings, and be careful online, as though individual discipline could offset a system designed to make ordinary life continuously measurable. This is like telling commuters to avoid traffic by choosing a better personality. Structural problems require structural answers. If AI intensifies surveillance capitalism, then the remedy cannot be limited to better disclosure. It must include stronger limits on collection, retention, sharing, and inference.
That is a hard political sell because the benefits of surveillance are diffuse and immediate to the powerful. The harms are delayed, distributed, and often invisible until they are not. A consumer rarely sees the model that decides which ad to show. A worker rarely sees the model that decides their schedule. A citizen rarely sees the model that flags them for extra scrutiny. Secrecy is not a bug in the system. It is part of how the system works.
What a less dystopian future would require
If this sounds alarmist, consider the alternative: not a dramatic cyber-totalitarian future, but a normal one in which people are simply known too well by institutions with too many incentives. That future is more plausible, and in some ways more frightening. Digital authoritarianism does not need jackboots if it has dashboards. It does not need overt censorship if it can shape what people see, buy, fear, and do. It does not need to ban dissent if it can make dissent costly.
A less dystopian future would require more than AI safety as conventionally discussed. Model alignment matters. So do misinformation, deepfakes, and autonomous weapons. But if the central political economy of AI remains extraction and surveillance, then technical guardrails will not be enough. Democracies would need to treat data rights as civil rights, and workplace monitoring as a labor issue, and commercial data brokerage as an accountability problem, not a nuisance.
They would also need to rethink the corporate myth that every improvement in prediction is a public good. In many contexts, better prediction is simply better power. The challenge is not to ban intelligence from machines. It is to stop letting machines deepen the asymmetry between those who observe and those who are observed.
“The question is no longer whether AI can think,” a policy expert said. “It is whether institutions will use AI to think for us, decide for us, and eventually rule us.”
The industrial age gave society factories. The internet gave it platforms. AI is giving it a layer of inference beneath both: a machine-readable society in which behavior is continuously translated into risk, value, and control. That may not be the future anyone wanted. It may still be the one we are building.
