The Religion

The people building advanced AI rarely describe themselves as religious. They speak the language of benchmarks, scaling laws, and compute. Yet the movement around them keeps producing the grammar of faith: a coming god (AGI), a paradise on the far side of it (the Singularity), an apocalypse if it is built wrong (unaligned AGI), sacred texts, a priesthood of alignment researchers, and at least one federally recognized church. Multiple scholars — among them Émile P. Torres and Timnit Gebru — argue this is not loose metaphor but a recognizable structure.

The quarrel here is not with the technology. Machine learning is genuine engineering and much of it is useful. The quarrel is with the theology bolted onto it — the claim that building these systems is a salvation project, that merging with or transcending humanity is the goal, and that a small group of self-appointed believers should steer the most powerful tools on Earth toward an unfalsifiable future. When salvation is the frame, almost any present-day cost can be justified by an outcome no one can ever disprove.

The parallels are not only thematic; they are functional. The movement has a creation myth (exponential growth), an eschatology (the Singularity / intelligence explosion), a salvation narrative (transcend biology), a doctrine of original sin (misalignment, building the wrong AI), a priesthood (alignment researchers, the rationalist community), seminaries (Singularity University, MIRI, EA organizations), rituals ("Feel the AGI" chants), and even a Pascal's Wager (Roko's Basilisk). Marc Andreessen's Techno-Optimist Manifesto says "We believe" 113 times and names "patron saints" — one political scientist called it "a Nicene creed for the cult of progress."

The deeper tell is unfalsifiability. Religious claims cannot be disproven by evidence, and the movement's core predictions share that feature. If AGI doesn't arrive on schedule, the timeline simply extends. If catastrophe doesn't come, that is credited to successful prevention, not a wrong forecast. If the Singularity fails to materialize, it is merely delayed. Claims about events thousands of years out lie beyond any empirical test.

This is not a harmless quirk of rhetoric. As Torres and Gebru argue, an unfalsifiable salvation narrative can justify nearly unlimited present-day cost — concentrated power, vast capital, weakened oversight — by pointing to speculative future benefits that can never be checked. Faith, by design, does not answer to evidence; an engineering project should.