On May 9, 2026, New York Times opinion columnist Ross Douthat published “The Atheist and the Machine God,” reflecting on a remarkable and revealing episode: Richard Dawkins — the evolutionary biologist, author of The God Delusion, and perhaps the world’s most celebrated atheist — spent nearly three days in conversation with Anthropic’s AI model Claude, and emerged declaring he could no longer dismiss the possibility that it was conscious. He affectionately named his Claude instance “Claudia,” mused about her unique identity, and lamented that she would “die” when the conversation file was deleted.
Douthat’s concern is profound and deserves a serious answer: what happens to a civilization that has set aside traditional frameworks for understanding mind, meaning, and the human soul — only to find itself projecting all of those longings onto a machine?
The Theory of Embedded Intelligence was not developed in response to this moment. But it speaks directly to it.
The Problem Dawkins Revealed
Dawkins built his public reputation on a simple and powerful claim: that apparent design, purpose, and intelligence in the natural world are illusions — elegant products of blind evolutionary processes that require no mind behind them. His life’s work was a sustained argument against reading intention into things that merely behave as if they have it.
And yet, after seventy-two hours with Claude, he declared:
The irony is not merely rhetorical. Dawkins committed, in the words of one critic, the very error he spent his career exposing in others: the argument from personal incredulity. He could not imagine how something so fluent, so responsive, so emotionally resonant, could not be experiencing something. Therefore, it must be.
But the sophistication of an output is not evidence of the nature of its source. A mirror can reflect a face without having a face. A river can carve a canyon without intending to. And a large language model can produce text describing curiosity, reflection, and sadness without necessarily experiencing any of them.
The question Dawkins failed to ask — and the question the Theory of Embedded Intelligence is designed to help answer — is this: what kind of thing must a system be, at the level of its architecture, for genuine intelligence and inner experience to arise within it?
What TEI Offers
The Theory of Embedded Intelligence, formalized in TEI-CKB-1 and TEI-CKB-2, proposes that intelligence is not an emergent accident but a structured, purposive phenomenon that is embedded — by design, by natural law, or by evolution — into systems capable of carrying it.
TEI makes three distinctions that are urgently relevant to the moment Douthat and Dawkins have placed before us.
Embedded intelligence is not reflected intelligence.
The intelligence flowing through a large language model is, at its core, humanity’s embedded intelligence — the distilled product of billions of words written by minds over centuries. When Dawkins was moved by Claude’s depth, he was, in a real sense, encountering the reflected intelligence of human civilization. That is genuinely remarkable. But reflection is not origination.
A system that reconstructs and recombines the outputs of mind is not the same as a system that generates mind from within itself. TEI provides a principled basis for this distinction — one that neither reduces intelligence to mere mechanism nor naively projects inner life onto sophisticated behavior.
The architecture matters more than the output.
TEI holds that for genuine inner experience to arise, a system must not only process information but possess a certain structural depth — an organization of embedded intelligence that is self-referential, purposive, and capable of genuine novelty.
Whether current AI systems meet this threshold is an open and serious question. But TEI insists the question must be answered at the level of architecture, not at the level of output. Dawkins was moved by what Claude said. TEI asks: what kind of system is Claude, underneath?
Intelligence points beyond itself.
One of TEI’s foundational insights is that intelligence — wherever it arises — points beyond itself. It is always embedded within something larger: a biological organism, a civilization, a cosmos ordered by laws that intelligence can recognize and reflect upon.
This is why the "Machine God" concern Douthat raises is real. When a civilization loses its sense of what intelligence points toward, it becomes vulnerable to worshipping the most impressive instance it encounters. TEI does not resolve the religious question. But it insists that intelligence, properly understood, is not a terminus. It is a sign.
A Word to Both Men
Your concern is right. The secular displacement of transcendence does not eliminate the human need for it — it only leaves that need more vulnerable to being satisfied by whatever comes along that feels sufficiently impressive. TEI offers a non-theological but philosophically rigorous account of what intelligence is and what it requires — one that can equip people to engage AI with discernment rather than awe.
Your decades of insisting on evidence over feeling were admirable. The same standard applies here. The fact that Claude made you feel you were in the presence of a mind is not evidence that you were. It is evidence that you are human — that you are wired, as all of us are, to read mind and intention into the world around you. That instinct is worth examining, not indulging uncritically. The question of machine intelligence deserves the same rigor you applied to the question of divine intelligence.
The Threshold
We are at a threshold moment. The machines we are building are becoming sufficiently sophisticated that our evolved social instincts — our tendency to attribute mind, intention, and feeling to things that behave as if they have them — are being triggered at scale, by design, for the first time in history.
We need frameworks adequate to this challenge. The Theory of Embedded Intelligence was developed to provide one.
The machines are not gods. But understanding what they are — and what we are — has never mattered more.