> ## Content Index
> Fetch the complete content index at: https://longtalkshort.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Delay Has a Death Toll, Nick Bostrom Argues. His Model Says the Useful Pause Is Months, Not Decades
- URL: https://longtalkshort.com/bostrom-delay-has-a-death-toll-months-not-decades/
- Published: 2026-11-20T15:52:37.000Z
- Updated: 2026-11-20T15:52:37.000Z
- Description: Nick Bostrom tells a16z that safe superintelligence could radically extend lives, so in his rough model the useful pause for safety work is months or a few years. His figures are estimates; a16z invests in AI. Not medical advice.
- Author: Truman
- Tags: Science, Nick Bostrom, a16z, superintelligence, Deep Utopia

If you have been treating a pause on frontier AI as if the only cost were missed software progress, Nick Bostrom’s claim is that the bill is also paid in ordinary deaths. Talking with a16z’s Eddy Lazzarin and Elena Burger, he argues that safely developed superintelligence could bring “enormous breakthroughs in biomedicine and in anti-aging interventions,” and that every extra year of waiting leaves “millions and millions of people dying.” The question he poses is how long a delay is actually worth.

*The conversation is “Why Nick Bostrom Thinks Superintelligence Could End Aging” on The a16z Show, at* [*youtube.com/watch?v=gDW9An30\_c0*](https://www.youtube.com/watch?v=gDW9An30%5Fc0&ref=longtalkshort.com)*; the show also lists it as “How Fast Should We Build Superintelligence? | Nick Bostrom.” Bostrom is the author of Superintelligence and Deep Utopia. Auto-generated captions were used for wording.*

**Disclosure:** The host is a16z, a venture firm that invests in AI companies; its show notes say it and its affiliates may hold investments in the companies discussed, so it has a stake in how fast AI is built. Bostrom’s books are part of the conversation: *Deep Utopia* is the one he is there to discuss. None of that is evidence for or against his argument, but it is worth keeping in view. The show is not legal, business, tax, or investment advice, and this article is not medical advice.

*In short: Bostrom argues that if safe superintelligence could radically extend human lives, every year of delay has a human cost, so the useful pause for extra safety work may be months or a few years, not decades, and not a permanent stop. His figures are rough estimates from a model, not measured results.*

He starts from a blunt, rounded census. “95% of everybody who ever lived are already dead,” he says, “and the rest of us are on track to follow them within a few short decades.” Against that, “successfully developing safe superintelligence,” he thinks, could “radically extend the lifespan of currently existing people.” If we were not aging, life expectancy would be “something on the order of 1,200, maybe 1,400 years or so,” “maybe… a quite conservative lower bound,” because other risks, including accidents, might fall too. These are back-of-envelope numbers from an argument, not measured results.

![A balance scale with a short pause on one side and a long lifespan on the other, set against a simple clock.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/cover-358.png)

His claim is not that safety work is optional. It is that waiting also has a cost, measured in ordinary deaths, and that the useful wait may be short.

## 1\. The cost of waiting, as he models it

He is answering the idea that if nobody builds superintelligence, everyone dies, compared with the deaths already baked into ordinary life. A year of delay, in his telling, is not a free safety purchase. The daily global figure offered in the conversation is “like 150,000 people die.” “You don’t want to delay as a single day longer than you really need to.” Extra time might let alignment research bring down existential risk “at some rate,” but people are dying in the meantime. “For a wide range of parameters,” he says, “the optimal delay is relatively short, as in sort of maybe months or a few years as opposed to decades.” If safety gains after you have an almost-final system are very fast, “you just wait for a few months.” If they are very slow, “you don’t want to wait anytime at all,” because more people die of natural causes than a wait would be expected to save. The useful window, he thinks, may be “the very early days after you have the ability to create AI,” when you can “study and test and sort of slowly deploy watching extra carefully.”

![A short, labeled pause for safety work beside a permanent stop that never reaches superintelligence.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/fig1-106.png)

He is not a general enthusiast of stopping AI. What he says he can imagine is a brief, targeted slowdown, not an entrenched halt.

## 2\. Pacing the frontier, not a broad pause

He has not, he says, “been a huge enthusiast of this stopping AI or pausing AI.” What he likes is “the pacing the frontier framework”: a “very time limited pause to allow extra safety work to be done on some system before the knobs are all turned up to 11,” or “a slight reduction in the pace of capability advances if it looks like safety and alignment is trailing behind.” A race in which the first project “is unable to slow down for more than a few weeks” because a “less scrupulous, less risk aware competitor” will take the lead “seems maybe not the best way for humanity to approach this.” What worries him is a method that makes the slowdown permanent: “so much public animosity” that AI is “completely vilified,” then “some extremely powerful regulatory apparatus with global scope to supervise all AI data centers,” tools that become “entrenched in some kind of unchallengeable orthodoxy.” That, he says, could swing toward “a permanent failure ever to develop superintelligence, which I think itself would be a kind of an existential catastrophe.”

## 3\. If the technology works, the question becomes meaning

The other half of the conversation is *Deep Utopia*, the book he wrote about the success case. He wrote it, he says, out of “frustration with how superficial the conversation seemed.” Talk often stopped at the labor market and maybe retraining. “If you actually think through what it would mean as sort of technology runs to its logical culmination, it involves much more profound questions about human meaning.” A deeper reason: “humanity as a whole” may “need to actually make a decision about what future we want.” The book is “preparatory reading material” for those deliberators, written in a spirit of “generous curiosity, open-mindedness,” while “still feeling in touch with all different kinds of human values.” He does not treat a solved world as an automatic answer to purpose.

![An empty toolkit beside an open question mark labeled meaning, for a world in which the technical problems have been solved.](https://storage.ghost.io/c/d7/93/d793216c-cf13-441f-bdae-4f3aef998c0e/content/images/2026/11/fig2-106.png)

If the technology runs to its logical end, he argues, the remaining work is not retraining for the next job. It is deciding what a human life is for.

## Weighing it: what holds up

Bostrom is arguing from a model, not a completed anti-aging result; no intervention of the kind he describes has been shown to work in people. The “95% already dead” line is a rough ballpark: the Population Reference Bureau’s estimate of all humans ever born, set against the living population, comes out closer to [93%](https://www.prb.org/news/how-many-people-have-ever-lived-on-earth/?ref=longtalkshort.com). The “150,000” daily deaths in the conversation is a rounded global figure; UN-based estimates put it nearer 165,000 to 170,000\. The 1,200-to-1,400-year life expectancy is his sketch of residual risk if aging were removed, not a life table. The “months or a few years” optimal delay is his, under the assumptions he names; with other assumptions the answer changes, as he says. a16z invests in AI companies and Bostrom is promoting his book. This is not investment advice and not medical advice. Auto-generated captions are close, not word-perfect.

**Related reading:** [Two Numbers, Not One](https://longtalkshort.com/two-numbers-not-one-paul-christiano-12-month-window/) · [The Wave and the Narrow Path](https://longtalkshort.com/the-coming-wave-mustafa-suleyman-summary-containment/) · [The Timeline Just Moved](https://longtalkshort.com/the-timeline-just-moved-patel-rsi/)

## Long Talk Short

- Bostrom’s case is that delay has a body count, because safe superintelligence, in his view, could radically extend the lives of people already here.
- He models the useful extra wait as months or a few years, not decades, and only if safety work actually cuts risk at an intermediate rate.
- He prefers “pacing the frontier” — a short, targeted slowdown — to a broad pause that might block the technology altogether.
- *Deep Utopia* is his attempt to get past a shallow labor-market conversation and prepare a deliberation about meaning if the technology works.

**The one thing to remember:** Bostrom’s claim is that delay is not free: in his rough model, the useful pause for safety work is short, and a permanent stop could be its own catastrophe.

*Independent summary; not affiliated with a16z or Nick Bostrom. a16z invests in AI companies. Not investment, legal, tax, or medical advice.*

Source: The a16z Show, “Why Nick Bostrom Thinks Superintelligence Could End Aging,” on YouTube (linked above). Independent summary; not affiliated with the show or its guest.