The Theology of the God You're Building

**EDO SEGAL:** Timnit, in your recent work with Émile Torres you did something computer scientists almost never do — you turned and looked at the ideology, not the technology. You bundled seven overlapping belief systems under one acronym, TESCREAL — transhumanism, extropianism, singularitarianism, cosmism, rationalism, effective altruism, longtermism — and argued they form a single worldview that has quietly become the engine driving the race to build artificial general intelligence. And Ray, I have to say it plainly because the audience deserves it: you are, by your own decades of writing, close to the center of that bundle. The S in TESCREAL is singularitarianism. So this round is unusually direct. Timnit, lay out the critique. Ray, you'll get the full floor to answer as the man being described.

**GEBRU:** Let me be precise, because the critique is often caricatured. The claim is not that these are seven random eccentricities. It's that they share a genealogy and a structure. And the genealogy — this is the provocative part, and I stand by it — runs back through twentieth-century eugenics. The same impulse to improve and perfect the species, to engineer a superior future being, animates the contemporary dream of a godlike machine intelligence that will solve all problems and spread mind across the cosmos. The promise of transcendence — expanded minds, uploaded consciousness, a posthuman future — carries forward the hierarchical, exclusionary logic of its ancestors, dressed now in the language of technology instead of biology. The dream of the perfect machine is the old dream of the perfect human in new clothing.

And the function of the ideology is what I most want named. If you sincerely believe that building AGI could either save humanity or extinguish it, then almost any cost in the present becomes justifiable — we did that round already, Ray, this is the engine under it. The rhetoric of [existential risk](https://www.youonai.ai/fieldguide/med/existential_risk) concentrates resources and authority in the hands of the people who claim to be steering humanity toward salvation, while the documented, immediate harms — the labor, the bias, the surveillance — get dismissed as distractions from the cosmic main event. The longtermist frame presents itself as the maximally responsible position, the most farsighted, the most concerned with humanity's welfare. I reframe it as a quasi-religious movement whose grand eschatology aligns suspiciously well with the commercial and personal ambitions of the wealthy men who promote it. And I'm careful — the critique doesn't depend on accusing anyone of bad faith. Most adherents are sincere. Ideas have social functions independent of the intentions of the people who hold them. This bundle of ideas functions, whatever its champions feel in their hearts, to direct vast resources toward speculative futures while licensing indifference to present harm.

**EDO SEGAL:** Ray, you've just been called the high priest of a eugenics-descended religion. The floor is entirely yours. Take the time you need.

**KURZWEIL:** I'll take it seriously, because it's a serious charge and a careless answer would confirm it. First, the eugenics genealogy. I reject it, and I reject it as someone whose own relatives fled the regime that made eugenics murder. Eugenics was about coercion — about the state deciding which people may exist, sterilizing and killing toward a "superior" type chosen by the powerful. My entire framework is the opposite vector: it's about individuals choosing to extend their own capabilities, voluntarily, with the technology becoming cheaper and therefore more available to more people, not less. Eugenics narrowed the circle of who counts by force. The democratization of intelligence widens it by cost. To collapse those into the same impulse because both use the word "improve" is exactly the kind of move Timnit would never accept if I made it about her work — guilt by etymology.

Second, the function argument. She says my warnings and my hopes both serve to concentrate power and license present harm. But notice the shape: if optimism is hype and caution is hype, then no statement I make can be evidence of anything, and we should all go home. That argument proves too much. And I'll grant the uncomfortable half — yes, the companies metabolize my framework for their valuations, and yes, some people invoke "saving humanity" while doing nothing for anyone currently alive. That's real and it's rotten. But the existence of cynics who exploit a belief does not make the belief false, any more than televangelists make grace false. I have spent fifty years putting specific, falsifiable dates on the record and grading them in public so anyone can check. That's the opposite of theology. Theology can't be wrong by 2029. I can.

And here's where I'll be sharp, because she was sharp with me. The TESCREAL move is itself a power move, Timnit. By bundling seven distinct things into one sinister acronym and tracing it to eugenics, you've built a machine for dismissing anyone who takes the long-term future seriously without engaging their actual arguments. It's an ad hominem with a genealogy attached. Some of the people in your bundle are wrong. Some are right. Some are dangerous and some are saints, and they disagree with each other violently. Flattening them into one heresy so you can refuse the whole thing — that's the thing you accuse me of doing with the word "AI." You named the thing. Now you've built a thing of your own that's just as good at making argument unnecessary.

**GEBRU:** That's a fair hit on the worst use of the acronym, and I'll take it — when TESCREAL is used to avoid argument, it's being misused, and people do misuse it. But you've dodged the genealogy by narrowing eugenics to its most murderous form. Eugenics was also, and primarily, a respectable scientific consensus among elites who sincerely believed they were improving humanity — endorsed by Nobel laureates, taught at the best universities, funded by the wealthiest foundations — and that's the part that rhymes, Ray. Not the gas chambers. The dinner parties. The confident, credentialed, well-funded certainty that a small group of brilliant people understood where the species should go and had the right to steer it there. That's the structure. And "voluntary, because it's cheap" doesn't dissolve it, because the people setting the direction — which capabilities, whose values, what counts as improvement — are the same handful of wealthy men, and the billions are once again the data, not the deciders.

And cosmism — the C in the bundle, the dream of spreading mind across the galaxy, filling the universe with our descendants, a future of unimaginable numbers of digital people — is where it gets truly dangerous, because once you're optimizing for ten-to-the-fifty future beings, the eight billion currently alive become a rounding error in your spreadsheet. That's not a caricature. That's the explicit math of the longtermist papers. The people writing them have calculated that a small reduction in the probability of cosmic catastrophe outweighs any amount of present suffering, because the future population is so large it swamps everything. That arithmetic, taken seriously, is a license to ignore every living person in the name of trillions of hypothetical ones. I don't think most of the people who hold it are monsters. I think they've built a machine for converting compassion for an imaginary multitude into indifference toward an actual one, and they cannot see that the machine is running because the machine flatters them as the saviors of all future time.

**KURZWEIL:** And I reject that arithmetic too, Timnit — I'm a singularitarian, not a longtermist, and the distinction you flattened matters here. I don't discount the present for ten-to-the-fifty future beings. I want the merger because of the people alive now, my generation, the ones dying on schedule. The fact that I share an acronym with people who'd sacrifice the present for a galactic spreadsheet is exactly the bundling problem I raised — you've put me in a room with positions I argue against and then asked me to answer for them. I'll condemn the spreadsheet with you. Now will you grant that condemning it doesn't condemn me?

**EDO SEGAL:** I'm going to stop the room, because we just hit a genuine convergence and agreements are news. Mark this. Ray conceded that the acronym is abused to dodge argument; Timnit conceded that the abuse is real and the worst form of the genealogy overreaches. You've narrowed the disagreement to something precise and load-bearing: whether the structure of a small credentialed elite steering the species is the danger Timnit says it is, or the engine of progress Ray says it is. That's the second convergence of the night, and it's sharper than either of you would have written alone. Hold it. The next round comes back to the ground — to the smooth interface, the friction Ray says is a tax and Timnit says was never the point, and the question of what gets lost when the cost of creation hits zero. After the break.