AI Isn't a Technology. It's a Belief System, and That Distinction Has Policy Consequences

There's a story policymakers tell themselves about artificial intelligence, and it goes something like this: China is winning because it doesn't care about privacy. Its government can hoover up data at scale, direct resources centrally, and race ahead while liberal democracies tie themselves in ethical knots. Whoever gets to superhuman intelligence first wins the century. Privacy, in this telling, is a luxury the West can't afford.

It's a tidy story. It's also probably built on a faulty premise.

What if the West's real vulnerability isn't its regard for privacy, but its uncritical embrace of the idea of AI itself? Consider what the term actually describes. "Nanotechnology" refers to an objective measure, scale at the level of atoms and molecules. "Artificial intelligence" refers to something far squishier: tasks that we, subjectively, decide count as intelligent. Fifteen years ago, the face-altering filters now common on Snapchat and Instagram were called image processing. Today the same underlying work gets marketed as AI. The shift in language did more work than any shift in the underlying science.

That matters more than it sounds. If "AI" is partly a marketing term, then it's also a philosophy, a particular lens for thinking about computation, and philosophies carry consequences. Here's an alternative lens: when a program sorts cats from dogs, don't describe it as a machine learning to see. Describe it as people contributing thousands of labeled examples, for the first time rigorously defining what visually separates a cat from a dog. Every situation framed as "AI doing something" has a second, equally valid framing: people doing something, through a tool. The AI framing tends to erase the humans. That erasure has a cost, and it isn't a small one.

The Economics of Erasure

Here's where the public health and labor angle comes in, because this isn't just an abstract philosophical quibble. Less than 10 percent of the US workforce works officially in the technology sector today, compared with 30 to 40 percent employed in the leading industrial sectors of the 1960s. Part of the explanation is structural: when people generate data, behavioral signals, or problem-solving input online, that labor isn't counted as work. It's treated as an unspoken barter for free services like search or social media. Meanwhile, investors chasing "AI is the future" narratives reward automation and overlook the less glamorous, more labor-intensive innovations that actually put people to work using networked tools.

Think of it like an iceberg. The visible tip is the slick algorithm, the chatbot, the recommendation engine. Below the waterline is an enormous, uncounted mass of human labor: people labeling images, moderating content, writing the text that trains language models, teaching the system what "good" output even looks like. China's real edge in AI isn't really surveillance. Recent reporting suggests it's a vast shadow workforce actively labeling data that feeds the algorithms, a workforce whose contributions mostly go unrecognized and undercompensated. That's not a formula unique to China. It's a pattern.

History offers a useful counterexample. Toyota's Kaizen production system, the heart of its manufacturing success through the 1970s and 1980s, worked because workers understood and improved the processes they were part of, and were recognized for doing so. Compare that to today's data-labeling workforce, kept invisible so the "pay no attention to the man behind the curtain" illusion of autonomous AI can hold. Bringing that labor into the light of acknowledged, paid work isn't a threat to privacy. It's arguably the only real alternative to a culture of ubiquitous, unpaid surveillance. Workers who get paid become fuller citizens: they spend money where they choose, gain leverage in the labor market, and historically, that's how working conditions improve in the first place.

There's a deeper point about capacity, too. Some estimates suggest a single human brain holds more raw computational capacity than every computer on Earth combined. With Moore's Law slowing and processor gains no longer doubling at the old pace, that gap isn't closing anytime soon. The idea that machines are poised to render human judgment obsolete doesn't hold up well against that math.

None of this is an argument against the underlying techniques, deep learning and convolutional networks are genuinely powerful tools. But they run on human-generated data. OpenAI's celebrated text-generation systems were trained on millions of pages written by people. Research in machine teaching shows that when the humans supplying training data are actively, thoughtfully engaged rather than anonymous and uncredited, the resulting systems train faster and cheaper. That kind of engagement only happens if contributors, not just the engineers at the top, are treated as essential and paid accordingly.

The usual rebuttal is that AI is starting to train itself, no humans required. But that's only true for a narrow slice of problems, ones that can be defined with mathematical precision rather than drawn from messy, statistical reality: chess, certain math proofs, game-playing systems. Even there, human-machine teams routinely outperform machines working alone. Self-training systems are the exception, not the template for how AI actually works in the world.

Why It Matters

Seen this way, "AI" looks less like a technology and more like a political ideology: the belief that a small technical elite can and should build systems that eventually operate independently of, and ultimately replace, human judgment altogether. That belief has close historical cousins in technocracy and central-planning socialism, both of which assumed a small expert class should supersede distributed human agency. It's worth sitting with the discomfort of that comparison, because it explains something odd: why a government built on centralized control would find this framing so congenial, and why Western institutions, amid their own crisis of public trust, have been so quick to adopt it too. The real policy question isn't how to out-surveil anyone. It's whether we're willing to pay, credit, and empower the people whose work actually makes these systems function.