The Machine Knows You Better Than You Know Yourself. That's the Business Model.

It takes a particular kind of audacity to announce that you have built a system to understand human beings better than they understand themselves, to monetize that understanding by selling access to their future behavior, and then to insist—with apparent sincerity—that this represents progress toward human flourishing. This week, as every week, we witnessed that audacity. Google's latest AI model, Gemini 2.0, was unveiled with the now-familiar promise: it would anticipate your needs before you articulated them. Meta announced further refinements to its recommendation algorithms, which already determine what two billion people see, read, and believe. Apple, ever the aesthetic packager of surveillance, introduced new "intelligence" features that would, it promised, make your digital life seamless by making your choices for you.

The language is always the same: personalization, optimization, assistance. The reality is rather different. What these companies have built is not artificial intelligence in any meaningful sense. It is a global infrastructure for behavior modification at a scale that would have made B.F. Skinner weep with envy. The difference is that Skinner had the intellectual honesty to call his work what it was.

The Precedent We Have Forgotten

This is not, of course, without precedent. In 1957, Vance Packard published The Hidden Persuaders, an exposé of how advertising agencies were using psychological research to manipulate consumer behavior below the threshold of conscious awareness. The public was horrified. Congressional hearings were held. The industry promised reform. What followed was not reform but refinement. Today's persuasion architecture makes Madison Avenue's subliminal advertising look like a child's lemonade stand next to ExxonMobil.

The difference is scale, speed, and asymmetry. Packard's ad men worked with surveys and focus groups. Today's platforms work with billions of behavioral data points collected every second: what you click, how long you pause, what makes your pupils dilate (if you're using a device with eye-tracking), what time of day you're most susceptible to impulsive decisions. They don't guess at your psychology. They measure it, model it, and then they architect environments designed to exploit it.

What Prediction Really Means

The tech industry speaks of prediction as though it were meteorology—passive observation of natural phenomena. This is a category error bordering on fraud. Weather forecasting does not alter the weather. But when Facebook predicts you will click on a particular post, it doesn't wait to see if you do. It adjusts the environment—what appears in your feed, in what order, with what emotional valence—to make that prediction come true. The prediction and the outcome are not independent variables. They are parts of the same engineered system.

Shoshana Zuboff, professor emerita at Harvard Business School, calls this "surveillance capitalism." In her 2019 book The Age of Surveillance Capitalism, she documents how tech companies discovered that the real value wasn't in the data itself but in the behavioral predictions derived from it—predictions that could be sold to anyone willing to pay for guaranteed outcomes. Advertisers were the first customers. But the model works just as well for political campaigns, insurance companies, and authoritarian governments. The product being sold is not your attention. It is your future behavior, rendered predictable and controllable.

One is tempted to point out that this violates every principle of informed consent, autonomy, and human dignity that liberal democracies claim to uphold. One would be correct.

The Evidence Is Not Subtle

The tech companies respond to such findings with a familiar script: the algorithms merely reflect human nature; they give people what they want; any problems are bugs, not features, and are being addressed by the very engineers who designed the systems in the first place. This argument collapses under minimal scrutiny. If the algorithms merely reflected demand, they would not require constant adjustment to maintain engagement. The truth is that human attention is not naturally infinite, and human outrage is not naturally renewable. Both must be cultivated.

The Argument They Won't Make

There is, to be fair, a serious counterargument that Silicon Valley never makes, perhaps because making it would require admitting what the business model actually is. It goes like this: human beings are not particularly good at making decisions. We are subject to cognitive biases, emotional volatility, and weakness of will. Left to our own devices, we eat badly, exercise rarely, consume misinformation, and vote against our own interests. Perhaps algorithmic guidance—backed by more data and better models than any individual possesses—would lead to better outcomes. Perhaps autonomy is overrated.

This is the argument for what philosophers call "soft paternalism"—the nudge, the default option, the architecture of choice that makes good decisions easy and bad decisions hard. And there are, indeed, contexts where it works: organ donation registries with opt-out defaults, retirement savings plans with automatic enrollment. The key difference is transparency and intent. A public health campaign that makes healthy food more visible in a cafeteria is not the same as a social media platform that makes anger more visible in a feed because anger increases engagement and engagement increases revenue.

The latter is not paternalism. It is exploitation. And it is not designed to improve your life. It is designed to monetize it.

What History Suggests

When industries build their business models on the systematic manipulation of human behavior, liberal democracies have historically had two responses: regulate or nationalize. The food industry's use of addictive additives led to the Pure Food and Drug Act of 1906. The tobacco industry's suppression of cancer research led to the Surgeon General's warning in 1964 and decades of litigation. The financial industry's predatory lending led to Dodd-Frank in 2010. Each industry fought regulation with the same argument: innovation, consumer choice, economic growth. Each industry lost, eventually, because the harm became undeniable.

The tech industry is different in scale but not in kind. What makes it more dangerous is the asymmetry. When Philip Morris lied about nicotine, the victims knew they were smoking. When Facebook manipulates your feed, you don't know which of your thoughts are yours and which were curated for you. The manipulation is invisible, continuous, and—this is the crucial part—it gets better over time. Every interaction trains the model. Every click refines the prediction. You are not merely the product. You are also the unpaid laborer improving the system that controls you.

The Policy We Need

So what is to be done? The European Union's Digital Services Act, which took effect in 2024, is a start—it bans certain forms of manipulative design and requires transparency in algorithmic systems. But transparency is not enough when the systems are too complex for any individual to audit and too profitable for companies to abandon voluntarily. What is required is a fundamental reframing of what we allow machines to do to human minds.

First: a ban on behavioral futures markets. Companies should be prohibited from selling predictions about individual behavior to third parties. You can sell me a product. You cannot sell my future decisions to someone else without my knowledge. This is not radical. It is the application of existing principles about bodily autonomy and informed consent to the digital realm.

Second: algorithmic transparency with teeth. Platforms should be required to disclose not just that they use algorithms but how those algorithms rank, recommend, and suppress content. More importantly, users should have the right to a non-personalized feed—a chronological timeline, a random sample, anything but a feed optimized to maximize engagement at the cost of everything else. Let people choose their own cognitive environment.

Third: a fiduciary duty for platforms. Doctors have a duty to act in their patients' interests. Lawyers have a duty to act in their clients' interests. Platforms that shape billions of people's perceptions of reality should have a duty to act in users' interests, not advertisers'. This means no A/B testing human emotions without consent. No radicalization pipelines. No dark patterns designed to defeat rational choice.

The Question We Must Answer

The fundamental question is not whether artificial intelligence will transform society. It already has. The question is whether that transformation will preserve or destroy the possibility of human autonomy. Right now, the answer is being written in code by engineers in Menlo Park and Mountain View, optimized for quarterly earnings and user engagement metrics. The result is a world in which your thoughts are not entirely your own, your decisions are not entirely yours, and your future is not entirely open.

This is not a future we must accept. It is a present we can still change. But doing so requires acknowledging what we are up against: not artificial intelligence, but a business model that profits from making human beings more predictable and more manipulable. The machine doesn't know you better than you know yourself. It's just trying to make sure you never find out who you might have been.

One is tempted to ask: What would it profit a civilization to gain the whole world of frictionless convenience and lose its capacity for self-determination? We are about to find out. Unless, of course, we decide not to.