The Promise That Conceals the Power Grab
Artificial intelligence is being marketed as the most useful invention since electricity, a general-purpose tool that will write emails, diagnose disease, manage factories, and rescue overburdened workers from drudgery. That story is not false, exactly. But it is incomplete in a way that matters politically. The real significance of AI may not be that it does old tasks faster. It may be that it changes who gets to see, decide, and control what happens next.
The public debate still tends to split AI into familiar categories: automation and unemployment on one side, creativity and convenience on the other. That framing is too narrow. AI is also becoming the nervous system of a new kind of economic and political order, one built on ubiquitous data collection, behavioral prediction, and continuous assessment. In that order, the machine does not merely replace labor. It reorganizes trust, authority, and autonomy.
That is why the most consequential question about AI is not whether a chatbot can pass a test or a robot can pick up a box. It is whether a society that delegates ever more judgment to algorithmic systems can still defend privacy, due process, and democratic self-rule. The danger is not science fiction. It is administrative and banal: software that decides who gets hired, which neighborhood gets policed, which worker is flagged as underperforming, which citizen is treated as suspicious, and which population is nudged rather than persuaded.
Automation Has Always Been About Power, Not Just Efficiency
Every technological wave produces the same reassurance: some jobs will vanish, but better ones will appear. Sometimes that is true. The industrial revolution did, in the long run, raise living standards and create new forms of work. But it also generated violent labor displacement, urban misery, and a century of struggle over who captured the gains. AI belongs in that history. What is different now is not the existence of disruption. It is the speed of adaptation on the part of employers and the scale of surveillance built into the tools themselves.
Traditional automation substitutes machines for muscle or repetitive routine. AI goes further because it can be aimed at cognition itself: drafting, coding, screening, scheduling, evaluating, forecasting. That makes it useful not only for eliminating tasks, but for decomposing jobs into measurable fragments and then subjecting each fragment to algorithmic control. The result is a workplace in which fewer people may be needed, but those who remain are watched more intensely.
That distinction matters. A factory robot can displace a worker. An AI management system can do something more insidious: turn the worker into a stream of metrics. It can score tone of voice, typing speed, ticket resolution time, facial expression, route efficiency, and even the pauses between keystrokes. It can present itself as neutral while embedding managerial power in a black box no worker can inspect. And because the system is framed as data-driven, challenging it becomes harder. Questioning the machine sounds like questioning reality.
This is why the current AI boom should not be understood as a purely technological transition. It is an organizational one. Firms are not adopting AI only to automate labor. They are adopting it to centralize judgment, reduce discretion, and lower the cost of supervision. A human supervisor can be argued with. A dashboard cannot. A manager can be distracted, sympathetic, or inconsistent. An algorithm can be relentless.
Surveillance Capitalism Has Found Its Perfect Instrument
If automation is one half of the story, surveillance capitalism is the other. The modern digital economy has long depended on the extraction of behavioral data: clicks, dwell time, location traces, purchase histories, social networks, biometric signals. AI supercharges this model because it makes the data more useful. It turns raw observation into prediction, prediction into manipulation, and manipulation into revenue.
The logic is straightforward. Platforms collect data because data can be monetized. AI increases the value of that data by improving targeting, recommendation, personalization, and ad delivery. It also broadens the range of what can be inferred. The system no longer needs to know only what a user clicked. It can estimate what the user will believe, buy, fear, or do next. That is not just advertising. It is behavioral steering.
Once a system can predict behavior, the incentive is to shape behavior more aggressively. That is the trap. The more powerful the models become, the more tempting it is to build environments that keep people legible and manipulable. The feed becomes a laboratory. The app becomes a sensor. The smartphone becomes a wearable tracking device carried willingly in a pocket. From the perspective of the platform, the ideal user is not a sovereign individual but a statistically exploitable one.
Maria Ressa has argued that this business model heightens the surveillance ecosystem rather than merely reflecting it, and that is precisely the point. AI does not arrive in a vacuum. It plugs into an existing machinery of extraction and intensifies it. The old internet monetized attention. The AI internet monetizes inference. That is a more dangerous bargain because it reaches beyond what people reveal and into what they can be induced to reveal, infer, or become.
In that sense, surveillance capitalism is not a side effect of AI. It is the operating system. Once that system is in place, companies and governments alike discover that more data is always better, that privacy is always a friction cost, and that the public has little visibility into how the whole machine works.
Workers Are Becoming the Product and the Audience
The labor market will absorb some of AI’s shocks through productivity gains, new occupations, and changing demand. But it would be naïve to assume that the transition will be smooth or broadly shared. Historically, productivity booms do not automatically translate into wage growth. They translate into bargaining over who gets the gains. In a weak labor market, the answer is usually: the owners do.
AI intensifies this imbalance because it gives employers a credible threat to replace, downsize, or deskill. For white-collar workers, the threat is not only job loss but job hollowing. A paralegal may become an editor of machine-generated drafts. A journalist may become a verifier of machine-produced copy. A customer-service worker may become a supervisor of automated responses. The work remains, but the human contribution is redefined as a thin layer of oversight over machine output.
That shift has a psychological effect as well as an economic one. When a worker is told that a model can do 80 percent of the job, the worker’s bargaining position changes even if the model remains unreliable. Employers need not fully automate to discipline labor. They only need to make replacement plausible. AI becomes a lever in wage negotiations, staffing decisions, and performance reviews.
Meanwhile, the very systems used to manage workers often generate the data that justifies further automation. This is a self-fulfilling loop. The more a company measures output, the more it believes output can be fully measured. The more it optimizes for quantifiable targets, the more it ignores unquantifiable qualities such as judgment, care, institutional memory, and trust. In the short term, that looks efficient. In the long term, it erodes the human capacities organizations most depend on when things go wrong.
There is also a class dimension. AI’s benefits tend to accrue to firms and professionals who already have leverage: corporations that own data, industries that can scale digital tools, workers with skills that complement automation rather than compete with it. The costs, by contrast, are often borne by lower-wage employees, precarious contractors, and communities already exposed to churn. The result is not just displacement. It is stratification.
The Authoritarian Temptation Is Built In
AI’s political danger is not limited to democratic societies. It is especially attractive to regimes that want to suppress dissent while maintaining the appearance of modernity and efficiency. A government with enough data and enough models can monitor populations at industrial scale. It can scan messages, map relationships, identify gatherings, flag unusual movement, and rank citizens by risk. What once required a large secret-police apparatus can increasingly be done through software.
That is why AI is so compatible with digital authoritarianism. It lowers the marginal cost of watching people and raises the cost of disappearing into the crowd. It does not merely tell rulers what citizens have done. It can suggest what they might do next. Predictive systems are particularly dangerous when the state treats prediction as guilt. At that point, the rule of law begins to yield to probabilistic suspicion.
But even democracies are vulnerable to this drift. The machinery of public safety, immigration control, welfare administration, and intelligence gathering increasingly relies on automated scoring and risk assessment. Supporters argue that these tools are necessary to manage complexity. Sometimes they may be. But once governments have the tools, the temptation to expand their use is immense. As one observer of AI and surveillance has noted, once data collection begins, institutions almost never stop; they simply find new excuses to continue.
This is the quiet seduction of AI governance: the technology arrives wrapped in promises of efficiency, fairness, and objectivity, while the real outcome is a larger administrative state with less transparency. Models can be audited in principle, but in practice many systems are too opaque, proprietary, or dynamic for meaningful scrutiny. When decisions are automated at scale, accountability diffuses. No one made the final call, or everyone did, which amounts to the same thing.
Why Regulation Keeps Missing the Point
Public policy has so far treated AI mostly as a problem of safety, competition, or copyright. Those are real issues, but they are not the whole problem. A narrow focus on model misuse or existential risk can obscure the ordinary abuses that are already happening. The most damaging consequences of AI may not come from a rogue superintelligence. They will come from normal institutions using powerful tools to extract more value, discipline more workers, and observe more citizens.
That means the regulatory agenda must be broader than testing models for bias or mandating disclosure labels. It should restrict unnecessary data collection, limit behavioral profiling, require meaningful explanation of automated decisions, and create enforceable rights for people affected by algorithmic systems. Just as important, it should constrain the economic model that rewards surveillance in the first place. If firms can profit indefinitely from invisible extraction, they will.
There is a deeper issue here: a society cannot defend privacy while celebrating convenience as the supreme value. Nor can it defend labor rights while accepting that every human activity should be measurable, rankable, and optimizable. The logic of AI pushes toward total legibility. Democratic life requires the opposite in many domains: ambiguity, discretion, unobservability, and the ability to make mistakes without being permanently scored.
That does not mean rejecting AI wholesale. It means refusing the ideology that says every useful tool must also become a universal instrument of extraction. The claim that “innovation” demands omnipresent data is not a law of nature. It is a bargaining position.
The Real Battle Is Over Human Agency
The most persuasive defense of AI is that it will augment human capability rather than replace it. That can be true in the best cases. Doctors may diagnose better with machine assistance. Engineers may design faster. Scientists may search more efficiently. But the optimistic version of AI depends on a boundary that is being eroded in real time: the boundary between assistance and domination.
When systems predict, rank, and nudge us, they do not simply help us choose. They help determine which options we see, which information reaches us, and which choices are treated as legitimate. In the workplace, that means less autonomy. In the marketplace, less privacy. In politics, less deliberation. Across society, more dependence on systems that few people understand and fewer can contest.
That is why AI deserves to be treated not as a neutral productivity tool but as a contested political infrastructure. The central question is not whether machines can think. It is who gets to think with them, who gets watched by them, and who gets to say no.
The danger of AI is not that it will become human. It is that we will become easier to manage.
