Jaron Lanier: The AI Danger Is the Ideology, Not

Jaron Lanier has a gift for making you feel like you've been thinking about something slightly wrong this whole time.

In a recent conversation with host Matt Kawecki on This Is The World, the VR pioneer and longtime Silicon Valley dissenter made an argument that sounds provocative on first pass but gets harder to dismiss the longer you sit with it: the danger of AI isn't the software. It's the story we've decided to tell about the software.

"AI in itself doesn't exist because it's an ideology," Lanier told Kawecki. "The ideology is the danger, not the software."

This isn't technophobia dressed up in intellectual clothing. Lanier spent decades building the foundational technology for virtual reality. He knows how software works. His argument is more precise than a simple warning about machines—and more interesting.

What's in a name

When you call something "artificial intelligence," you're not just describing a category of software. You're making a philosophical claim: that we've created something brain-adjacent, something that thinks, possibly something that deserves its own moral consideration. Lanier's point is that this claim is bundled invisibly into the terminology, and most people never notice they've accepted it.

He offers three ways to frame the same software: as a new creature (what "AI" implies), as a machine (like a tractor), or as a new form of collaboration between people. He advocates for the third frame, comparing AI models to Wikipedia or open-source software—outputs of collective human contribution rather than emergent alien intelligence.

This isn't just philosophical hair-splitting. The frame you choose has real consequences. If you believe you're dealing with a new kind of creature, human responsibility naturally diffuses. The system did it. The algorithm decided. The AI hallucinated. Each of those phrasings quietly moves accountability away from the people who built, trained, and deployed the software. Lanier's collaboration frame forces the question back where he thinks it belongs: on the humans whose data, decisions, and labor produced the output in the first place.

Whether you find this reframe convincing or not, the underlying problem he's identifying—that our vocabulary shapes our accountability—is worth sitting with.

Silicon Valley's new religion

Lanier goes further, and this is where some of his AI researcher colleagues get uncomfortable: he calls the culture around AI a religion.

Not as an insult. As a structural observation.

"For many people, particularly some of the younger AI people in the West Coast and in the Bay Area in particular, it definitely has all the qualities of religion," he said. "They feel it might give them immortality or it might create an apocalypse and something better might come after the apocalypse."

The tell, for Lanier, isn't the belief in extraordinary outcomes—it's the way those beliefs function. When the promise of superintelligence becomes a reason to deprioritize feeding hungry people today, or when disagreeing with the prevailing AI narrative feels like committing a sin, you've left the domain of engineering and entered something else. The future-oriented certainty becomes a reason to discount the present. That's not a technical posture. That's eschatology.

It's worth noting that many AI researchers would push back hard here. The effective altruism-adjacent community that Lanier is probably gesturing at represents a fraction of the people building AI systems. Most researchers are focused on narrow, concrete problems. Demis Hassabis, whom Kawecki mentions in the conversation, won a Nobel Prize for protein-folding work—not for prophesying machine godhood. There's a real risk that Lanier's religion critique, aimed precisely, hits a target much larger than it should.

Still, the phenomenon he's describing exists. Whether it's representative of "AI culture" broadly is the question he leaves more open than he probably intends.

From selling to controlling

The piece of Lanier's argument that I find most technically grounded—and most unsettling—is his distinction between capitalism as we've known it and what the major tech platforms have actually become.

The standard critique goes: surveillance capitalism, a term coined by Shoshana Zuboff, describes how companies harvest behavioral data to sell ads. Lanier respects Zuboff and her work, but he thinks the "capitalism" framing is already out of date. He points to something in OpenAI's investor prospectus that stopped me cold: it reportedly tells investors to expect to lose their money, because the company will probably make money itself irrelevant.

Elon Musk has said something similar. So have others in the industry. The endpoint of their vision isn't a more profitable capitalism—it's a post-monetary system where direct behavioral control replaces the market as the mechanism of power.

"Behavior modification no longer needs money as an intermediary," Lanier said. "That's really the most important thing."

This is an important distinction. If he's right, then the debate about whether tech companies are monopolies, or whether we should tax them more, or break them up, is a debate about the old game while the new game is already being played. Power through algorithmic feedback loops doesn't need to sell you anything. It just needs to move you.

Lanier draws a direct line from this to the global rise of personality-cult politics. He's careful not to call it a left-right shift—he describes it as something stranger, a form of governance-by-manipulation that algorithmic systems are structurally suited to amplify. The mechanism: human brains are wired to respond deeply to certain social and emotional signals, and a network of devices tracking behavioral feedback can tune itself to activate exactly those signals. The result isn't persuasion in any traditional sense. It's something closer to direct neural access.

You can agree or disagree with the political reading. The underlying mechanism is harder to argue with.

Experiments, not answers

When Kawecki asks whether Big Tech should be nationalized, Lanier sidesteps the question in an interesting way. He doesn't want nationalization. He doesn't want a single regulatory answer. He wants variety.

His concern is epistemological: we currently have one global experiment running—the American platform model—with no control group. We can't know if things could be different because we've never let them be different. He points approvingly at Australia's moves to restrict social media for minors, not because he's certain it's right, but because it generates data. He half-endorses EU tech regulation not because he agrees with every provision, but because friction creates comparison.

This is an unusual position for someone with strong views. Lanier is clearly not neutral on these questions—he thinks the current system is doing real damage. But his prescription is essentially: let a hundred models bloom and see what survives.

His data dignity proposal fits this framework. The idea—that people should be compensated for the data they contribute to AI training—has attracted criticism from privacy advocates who argue that economic participation in a data economy is incompatible with genuine privacy. Lanier accepts the tension. "There's no perfection in this life," he said. "The problem is there's no way out of it." He's not selling utopia. He's selling the least bad version of an ancient dilemma: the trade-off between self-determination and dependence on power.

That honesty about limits is, genuinely, not what you usually get from someone pitching a solution.

What's actually at stake

The most useful thing Lanier offers isn't a policy prescription or a rebranding exercise for machine learning. It's a prompt to examine what we've already agreed to without noticing.

We accepted "artificial intelligence" as a neutral descriptive term. It isn't. We accepted that the people building these systems are primarily in the business of selling products. They may not be—or may not be for much longer. We accepted that the risks of AI are mostly about what the software might do autonomously. Lanier thinks the risks are mostly about what the software licenses us to stop thinking about.

His mentor, Marvin Minsky—who did more than perhaps anyone to build the mythology of artificial intelligence—apparently loved arguing with him about this. Loved it. There's something worth holding onto in that image: the father of AI ideology and his most persistent critic, finding the disagreement itself to be the productive thing.

The question Lanier keeps circling without quite landing on is whether that kind of productive disagreement is still possible when one side has accumulated more centralized money and behavioral influence than any institution in human history.

By Rachel "Rach" Kovacs, Cybersecurity & Privacy Correspondent, Buzzrag