Artificial intelligence is routinely described as the most important productivity technology since electrification. That may be true and still miss the point. The more consequential effect of AI may be something darker and more durable: a shift from economies that sell goods and services to systems that extract, predict and shape human behavior. In that world, the workplace becomes a laboratory, the city becomes a sensor grid and the citizen becomes a data source.
The old anxiety about automation was simple enough. Machines would do the work; people would either move up the value chain or lose out. That story was always too neat, but it at least assumed that technology’s chief effect was on labor. AI complicates the picture. It does displace jobs, and likely many more will be quietly reorganized than dramatically eliminated. Yet its deepest impact may be on power: who gets to observe, classify and steer human action at scale. Once that possibility exists, the temptation to use it spreads from firms to states, from advertising to management, from convenience to coercion.
Shoshana Zuboff, who popularized the phrase “surveillance capitalism,” argues that the point is not merely to observe behavior but to transform it into a new kind of asset—behavioral data that can be sold, predicted and used to influence future conduct. Her warning has become more relevant, not less, as AI systems have made extraction more efficient and inference more powerful. In her account, the problem is not technology itself but the business model that rewards monitoring everything, everywhere, all the time.
That logic is now no longer confined to social media feeds. It sits inside workplace software that tracks keystrokes, retail systems that measure foot traffic, cameras that read license plates, health platforms that infer stress and productivity, and consumer devices that silently harvest data from the most intimate corners of life. AI does not create the appetite for surveillance. It industrializes it.
The end of consent
For years, the moral alibi of digital capitalism was consent. Users clicked “I agree,” regulators looked away and firms treated the resulting permission as a blank cheque. But consent in the digital economy was always less a contract than a fiction. The terms were unreadable, the alternatives weak and the data flows opaque. AI makes that fiction even flimsier, because the value of data increasingly lies not in what people knowingly reveal but in what systems can infer from patterns they never intended to disclose.
That matters because prediction is not neutral. A system that can infer pregnancy, political leaning, sexual orientation, depression, union sentiment or job-hunting intent has already crossed from observation into power. It can target, nudge, exclude or pre-empt. The user may have opted into a map, a fitness app or a workplace dashboard; what they have really entered is a generalized environment of continuous legibility.
Once that environment exists, the line between commercial surveillance and state surveillance blurs. Data brokers and private platforms accumulate records that governments can buy, subpoena or access through partnerships. The result is a legal and political shortcut around the protections that once constrained direct state collection. The state does not always need to spy in the traditional sense if markets will do the collecting for it.
That is why the strongest critique of AI is not that it is smart. It is that it is scalable. A manager can supervise ten people directly, but software can supervise ten thousand. A police officer can follow one suspect, but an algorithm can process citywide camera feeds, scrape social media and correlate movements, contacts and purchases. AI lowers the cost of watching, and when the cost of watching falls, watching becomes normal.
The workplace as sensor field
Job displacement remains the most visible fear surrounding AI, and for good reason. The technology is already beginning to compress layers of white-collar labor that once looked secure: drafting, summarizing, coding, scheduling, customer support, preliminary legal review, basic design, recruiting, compliance. But the more revealing trend is not pure replacement. It is managerial augmentation.
AI allows firms to break work into smaller measurable units, compare workers against machine-generated benchmarks and intervene in real time. The office becomes more like a warehouse. Performance is not judged by broad professional discretion but by fine-grained telemetry: response time, output volume, apparent sentiment, predictive risk. That may increase efficiency. It also changes the bargain between employer and employee.
In theory, technology can free workers from drudgery. In practice, it often frees management from trust. The software does not merely ask whether a task was completed; it asks whether the worker could have completed it faster, cheaper or with less autonomy. The worker is no longer a professional but a variable to be optimized.
This is where the rhetoric of augmentation becomes misleading. Augmentation suggests partnership. What many firms really want is asymmetry. AI gives the boss a richer picture of labor while giving labor little comparable visibility into the institution that governs it. The result is a one-way mirror: employees are made more transparent, while decision-making becomes more opaque.
That imbalance will not stay inside private companies. Public-sector employers are adopting similar tools under the banner of efficiency, and gig platforms have already normalized algorithmic management. The deeper precedent is political: once citizens get used to being scored, sorted and nudged at work, the same logic becomes easier to extend elsewhere.
From surveillance capitalism to digital authoritarianism
The phrase “digital authoritarianism” is often used to describe overtly repressive regimes, but the more unsettling possibility is that the infrastructure of control can be built in ordinary democracies without any dramatic rupture. It arrives piecemeal: a city contracts with a facial-recognition vendor, a school system deploys monitoring software, a welfare agency uses risk scoring, a police department buys predictive analytics, a border authority assembles data from multiple commercial sources.
Each step can be defended in isolation. Who objects to safer streets, faster service or fraud reduction? But the aggregate effect is a lattice of observation that is hard to reverse once installed. Surveillance infrastructures are sticky. They are expensive to build, convenient to use and politically difficult to dismantle once agencies have grown dependent on them.
There is a reason the language of safety is so powerful in these debates. Safety is the most persuasive bridge between private profit and public coercion. Platforms insist they are moderating content to protect users; employers say they are monitoring productivity to protect customers; governments say they are collecting data to protect citizens. In each case, the claimed purpose is restraint. In each case, the capability expands.
The danger is not just abuse by bad actors. It is mission creep by institutions that persuade themselves they are acting responsibly. A database assembled for fraud detection becomes useful for policing. A workplace productivity tool becomes a tool for discipline. A consumer profile becomes a political profile. AI does not have to be malicious to be dangerous. It only has to be useful.
That usefulness is what makes surveillance capitalism so resilient. It is not merely parasitic on democracy; it thrives on its habits. Democratic societies generate speech, mobility, commerce and dissent, all of which produce data. The more active a society is, the richer the behavioral exhaust. The very freedoms that liberal societies prize can be repurposed as raw material for prediction and control.
AI does not have to be malicious to be dangerous. It only has to be useful.
The productivity mirage
Defenders of aggressive AI deployment often retreat to a practical argument: whatever its risks, the technology raises productivity, and higher productivity is the foundation of prosperity. This is true in the abstract and incomplete in reality. Productivity gains do not automatically translate into broad-based welfare. They can just as easily become shareholder value, executive leverage and labor substitution.
History offers a useful caution. Industrialization eventually raised living standards, but only after prolonged struggle over wages, safety and political rights. The question was never whether machines would improve output. It was who would capture the gains and under what conditions. AI is already repeating that conflict in compressed form.
But even the productivity case may be overstated. A great deal of current AI investment is aimed not at generating new value but at reducing uncertainty, standardizing decisions and extracting more from existing relationships. That is a different economic model. It is less about invention than surveillance, less about making things than knowing things about people so that firms and states can intervene earlier and more precisely.
Seen this way, the real macroeconomic prize is not merely labor savings. It is control over attention, demand and conduct. If that sounds abstract, it becomes concrete the moment a bank denies credit, an insurer prices risk, an employer flags dissent or a political campaign micro-targets persuasion. AI is transforming markets into instruments of behavioral governance.
What resistance would require
The obvious response is regulation, and regulation is necessary. But it is not sufficient. Rules about transparency, consent and fairness matter, yet they often arrive after systems have already spread. A model that has become infrastructural is hard to govern through after-the-fact disclosure alone. If the underlying business model rewards extraction, compliance often becomes a paperwork exercise.
Zuboff has argued that the solution must be upstream: not just more oversight, but a rejection of the logic that treats human experience as a quarry. That is a more radical proposal than it first appears. It means refusing the idea that every signal emitted by a person is fair game for private capture. It means distinguishing between services that people use and systems that use people.
Some policymakers have begun to move in that direction, especially in Europe, where the Digital Services Act, the Digital Markets Act and the AI Act reflect an effort to constrain the most abusive forms of platform power. Yet the harder challenge is institutional. Democracies will need stronger public expertise, independent auditing, tighter limits on data brokerage and greater protection for workers subject to automated management. They will also need to treat certain uses of AI not as mere products but as governance systems.
There is a deeper cultural task as well: to stop treating omniscience as innovation. The most seductive promise of AI is that it can reveal what is hidden. But in human affairs, hiddenness is often not a bug. It is what allows intimacy, dissent, experimentation and autonomy to exist. A society in which every action can be recorded and every preference predicted may be efficient. It is also thinner, flatter and easier to control.
The broad debate about AI often splits into optimists and pessimists. The optimists see productivity, medicine, discovery and convenience. The pessimists see unemployment, bias, fraud and error. Both miss the largest issue. AI is not only a technology of doing; it is a technology of knowing. And when knowledge is centralized in the hands of institutions that profit from asymmetry, the result is not merely disruption. It is a new architecture of power.
The most provocative claim about AI, then, is not that it will think like us. It is that it will teach governments and corporations to organize society less like a public sphere and more like an intelligence operation. That possibility is not inevitable. But it is already profitable, and that makes it far more dangerous than the fantasies of sentient machines that dominate the public conversation.
The future of AI will not be decided by whether a chatbot writes a competent essay or a robot folds laundry. It will be decided by whether democratic societies allow prediction to become permission, and surveillance to become governance.
