The machine age has returned, but this time it watches
Every great technological era tells a flattering story about itself. Steam promised liberation from muscle. Electricity promised light, leisure, and the end of industrial drudgery. The internet sold itself as a decentralized commons that would widen access to information and power. Artificial intelligence is being marketed in the same register: as a force that will make us faster, smarter, and freer.
That story is not false. AI will automate tasks, compress costs, and remove some of the most tedious work humans have ever done. But the more important story is less cheerful and more politically revealing. AI is not merely a tool for doing old things more efficiently. It is becoming an infrastructure for deciding who is watched, who is replaced, who is nudged, who is denied, and who is deemed economically unnecessary. In that sense, the central AI revolution is not cognitive. It is administrative.
The provocative case for saying this plainly is that the technology’s most immediate uses are not about creativity or cure, but about classification and control. AI does not just write, summarize, and generate. It predicts, ranks, flags, scores, and recommends. That makes it the perfect instrument for a system that wants to extract maximum value from human behavior while minimizing the cost of human discretion. It is the engine room of surveillance capitalism, and, in less benign hands, a ready-made architecture for digital authoritarianism.
Automation is not new; invisibility is
Every generation of workers hears that machines will handle the repetitive tasks and leave the meaningful work to people. Yet the industrial history of automation shows a more sobering pattern. Machines do not merely eliminate labor; they reorganize it. They take apart occupations, reduce skilled judgment to standardized procedures, and shift bargaining power toward owners and managers. AI extends that logic into domains that once seemed protected by judgment: hiring, scheduling, customer service, underwriting, content moderation, logistics, even supervision itself.
The distinctive feature of this wave is not that it replaces all workers, but that it disaggregates work into fragments that can be automated selectively. A lawyer may still review contracts, but AI can draft the first pass. A teacher may still evaluate students, but software can monitor attention and homework completion. A warehouse worker may still move packages, but an algorithm decides where, when, and how fast. The worker remains present, but the decisive intelligence has migrated upward into systems that are opaque, centralized, and difficult to contest.
That matters because displacement is not only about unemployment. It is also about the erosion of agency. A human being can negotiate with a boss, appeal to a supervisor, or improvise around a bad policy. A machine-sorted workplace is less legible and less negotiable. When hiring filters, productivity dashboards, and automated discipline systems become the invisible governors of labor, the result is not just efficiency. It is a new kind of managerial absolutism.
One of the most telling facts about AI deployment is how often it is introduced under the language of assistance and ends up functioning as surveillance. Productivity software becomes keystroke tracking. Customer support analytics become performance scoring. Route optimization becomes real-time behavioral monitoring. The system is sold as help, but its actual value proposition is that it knows more about the worker than the worker knows about the system.
Surveillance capitalism is the business model behind the miracle
The deepest misunderstanding about AI is that it exists in a vacuum of technical progress. It does not. It is being built inside a political economy that has already learned how to monetize attention, behavior, and prediction. Scholarship and policy debates on surveillance capitalism describe a system in which raw digital traces are transformed into commercial assets, often by collecting more data than is strictly needed for the service being provided. Research on the subject identifies behavioral surplus as the underlying logic: the excess data that can be mined, modeled, and sold. [4][5]
That logic is especially powerful when paired with AI, because machine learning thrives on volume, texture, and constant feedback. The more data the system ingests, the better it becomes at pattern recognition; the better it becomes at pattern recognition, the more valuable it is to advertisers, employers, insurers, platforms, and states. This produces a self-reinforcing loop. AI rewards data hoarding, and data hoarding rewards AI. Once that loop takes hold, restraint becomes economically irrational unless it is imposed from outside.
Harvard’s discussion of surveillance capitalism and democracy makes the political consequence explicit: once institutions start collecting data, they are rarely inclined to stop, and the resulting machinery can be used to shape behavior and undermine democratic life. [1][2] The point is not that every algorithm is sinister. It is that the incentives of the system tend to favor opacity, retention, and expansion. If prediction yields profit, then the temptation is always to predict more; if prediction yields power, then the temptation is always to control more.
That helps explain why the most ambitious AI systems are being integrated into platforms that already know where users go, what they click, whom they call, what they buy, how long they linger, and how they respond to fear, novelty, and social pressure. AI is not creating surveillance capitalism from scratch. It is supercharging it.
“The only way to limit these negative outcomes is to treat AI like other dangerous technologies: by creating well-designed regulation to emphasize safety and social well-being.”
That warning, drawn from a discussion of AI’s role in surveillance capitalism, captures the scale of the problem: the issue is not one bad product or one rogue company, but the industrialization of observation itself. [1]
The workplace is becoming the first laboratory of digital authoritarianism
If digital authoritarianism sounds like a foreign-policy concept, it should not. The workplace is its most intimate testing ground. Employers have long monitored workers, but AI turns old oversight into continuous governance. Cameras detect posture. Software measures “engagement.” Scheduling systems infer reliability. Voice analysis claims to estimate sentiment. Hiring tools sort candidates before a human ever sees them.
This is not merely an efficiency story. It is an asymmetry story. The person being measured rarely knows the criteria, the thresholds, or the trade-offs embedded in the model. And unlike a human manager, the system does not tire, forget, or empathize. It can score thousands of workers at once, incorporate data from many sources, and update its judgments continuously. The result is a form of rule by machine that looks neutral precisely because it is procedural.
That neutrality is deceptive. AI systems inherit the priorities of the organizations that deploy them. If a company wants to reduce labor costs, the model will optimize toward that end. If a state wants to identify dissent, the model will learn which networks, phrases, and associations deserve attention. If a platform wants to maximize engagement, the model will feed users what keeps them present, even when that corrodes public discourse. The issue is not that algorithms possess political opinions. It is that they operationalize political goals without admitting they are political at all.
Maria Ressa, the Nobel laureate and journalist, has repeatedly warned that the same digital systems that connect people can also be used to manipulate them, heightening the surveillance ecosystem that shapes modern life. [3] Her point is less about any single platform than about the total environment: when human attention is continuously captured, categorized, and optimized, civic life becomes easier to steer and harder to trust.
The myth of productivity hides a transfer of power
AI evangelists often present automation as an argument about output: more goods, more services, lower costs, fewer bottlenecks. But productivity is a distributional question before it is a technical one. Who captures the gains? Who bears the risk? Who gets a say in deployment? Those questions are now being answered in a predictable direction. The owners of models, compute infrastructure, and data pipelines gain leverage; workers absorb the uncertainty; consumers receive convenience wrapped in dependency.
The rhetoric of augmentation is especially effective because it sounds humane. It promises that people will be freed from boring work and elevated to higher-value tasks. In some cases that will be true. But augmentation can also be a transitional stage in substitution. A system first becomes a copilot, then a benchmark, then the standard, then the replacement. Once managers see that output can be maintained with fewer humans, the political incentive to preserve human jobs weakens quickly.
What makes this moment different from earlier automation cycles is the speed with which AI can spread across sectors that once depended on tacit judgment. That raises the possibility of a labor market in which entry-level jobs, apprenticeship pathways, and routine professional tasks are hollowed out faster than new roles are created. A society can survive technological displacement. It struggles more when it destroys the ladder by which people were supposed to climb.
And because AI systems are often embedded in subscription services and cloud platforms, the cost of adoption can be low and the consequences diffuse. A firm does not need a grand strategy to automate; it only needs to follow the market logic of everyone else. That is how structural change happens now: not through dramatic decrees, but through thousands of uncoordinated decisions that add up to a reordered economy.
Why the state is tempted to love the same tools
Private firms are not the only actors drawn to AI’s predictive power. Governments are discovering its attractions as well. The state has always wanted legibility: census data, tax records, border controls, welfare databases, police files. AI intensifies that appetite by promising to find risk in the noise. It can triage welfare claims, identify suspected fraud, prioritize enforcement, and infer patterns across huge datasets. In theory, that can improve public administration. In practice, it can also turn suspicion into infrastructure.
Once a government learns that AI can locate “anomalies” among citizens, the temptation is to expand the search. The model may begin with fraud detection and end with social sorting. The same system that flags unusual transactions can also identify protest networks, monitor speech, or map relationships across digital spaces. When the line between public and private data becomes porous, democratic oversight becomes harder because the machinery of observation is distributed across contractors, platforms, agencies, and vendors.
That is why the long-term risk is not simply mass unemployment. It is institutional reflex. Bureaucracies love tools that promise precision. Security agencies love tools that promise preemption. Employers love tools that promise compliance. Markets love tools that promise prediction. The same technological stack serves all four instincts at once.
The political question is not whether AI is powerful, but who gets to refuse it
The most misleading debate about AI asks whether the technology is good or bad. That is the wrong level of analysis. AI is neither a moral agent nor a neutral force. It is a set of capabilities whose effects depend on ownership, law, labor power, and institutional design. The decisive question is whether societies can impose limits before the incentives to extract, automate, and monitor become too entrenched to reverse.
That means a serious response cannot stop at ethics statements or voluntary safety pledges. It has to include privacy rules that actually constrain collection, labor protections that prevent algorithmic management from becoming arbitrary rule, transparency requirements that make automated decisions contestable, and competition policy that reduces the concentration of data and compute in a few firms. It also requires a more radical idea that has been stated bluntly in policy circles: some parts of surveillance capitalism may need to be abolished, not merely moderated. [2]
That is an uncomfortable position because it challenges a widely shared fantasy: that the digital economy can remain simultaneously extractive and liberal, efficient and humane, predictive and private. It cannot. A system built to maximize behavioral prediction will always be tempted to turn public life into a dataset. A system built to maximize automated management will always be tempted to turn workers into variables. A system built to maximize engagement will always be tempted to turn democracy into distraction.
There is still time to choose otherwise, but only if the public stops treating AI as an oracle and starts treating it as an instrument of power. The new machine age will not be defined by whether machines can think. It will be defined by whether institutions use them to know too much, decide too much, and trust people too little.
That is the real provocation. Artificial intelligence is often described as a future technology. In practice, it is already reorganizing the present around a very old ambition: to make human beings more legible, more governable, and more profitable than they have any right to be.
