Three camps in the AI harm debate

A summary of Deep Questions with Cal Newport, “An Insider’s Guide to AI Doom” — published 1 October 2026, 28 m 39 s. No guest: a solo explainer, drawn on a tablet as he talks. Audio: podcasts.apple.com; video, same title, at youtu.be/McsQ7hXzd3M. Timestamps link into the video at that moment.

Structure, from the show’s own chapters: intro (0:00), why the debate is so heated (1:01), the doomers (2:01), the doomer civil war (9:12), the accelerationists (14:03), the centrists (17:51), the full landscape and the fight over regulation (24:00).

This is the framework episode that precedes the Torres interview summarised in A stark warning from a former AI doomer, published a week later. Where that episode is one camp’s critique argued at length by a defector, this one is Cal’s own map of all three camps — and it places him on it. See Method and caveats for what was checked and what was not.

The gist in one paragraph

Cal’s claim is that the AI-harms argument is unreadable unless you know which of three camps a given speaker belongs to, because two of them are asking for regulation and mean entirely different things by it. The doomers descend from 1990s Bay-Area transhumanism via rationalism and effective altruism; they hold that superintelligent AI is inevitable and will be either a good god or a bad one, and Anthropic and OpenAI sit inside this camp rather than opposite it — which is why the loudest attacks on those two labs come from their own intellectual family. The accelerationists — VCs, the non-doomer AI companies, the White House — think the doomers are weird and the race worth winning. The centrists want AI regulated as a normal technology, share the accelerationists’ rejection of doomer metaphysics and the doomers’ appetite for intervention, and agree with neither about why. Cal’s payoff: Congress is currently being briefed mostly by the doomers, and a doomer-shaped regulation is the wrong instrument for the harms a centrist is worried about.

Two caveats he opens with (0:50): he is “basically just winging this episode,” the boundaries are “a little bit porous,” and he is trying to keep his own opinion out of it — an attempt he abandons explicitly in the third camp, which is his.

The three camps at a glance

Camp Who Core belief Motive Cal ascribes
1 Doomers rationalists, AI-focused EAs, plus Anthropic and OpenAI superintelligent AI is inevitable; it will be a good god or a bad god an eschatological futurist utopianism dating to the 1980s–90s
2 Accelerationists VCs, the other AI firms, the White House the doomers are weird; go fast, the benefits are real and China is racing money and power
3 Centrists AI ethicists, “normal technology” computer scientists, cybersecurity experts AI is powerful but ordinary; regulate it like any other unsafe technology — (Cal’s own camp)

On the asymmetry between those motives (17:04): the accelerationist story is the familiar one — “companies want to make more money and they don’t want stuff to stop them” — and so it reads as legible. “The doomer story is a little bit more complicated and I think it’s sort of thrown a lot of people off … way stranger in the American experience than just being driven by I want to get a return on investment.”

Camp 1: the doomers

The lineage

Back to the Extropians, the futurist and “very utopian” Silicon Valley movement of the 1980s–90s (2:33). They wove two technologies together into one transhumanist utopia: nanotechnology, via Eric Drexler — nanomachines that repair bodies, reverse aging and disease — and AI, in the form of a superintelligence acting as “essentially a benevolent digital deity” (3:26). The Extropians were, on Cal’s telling, “pretty positive.”

Eliezer Yudkowsky is the hinge (3:26): dropped out after the 8th grade, read science fiction, found the Extropian mailing list, wrote heavily for it, got funding for his own techno-utopian institute — and then concluded that superintelligent AI “could also just as easily kill us all.” Inevitable, but not inevitably good: “more eschatological than the Extropians realized” (4:16). That strand became the rationalists.

Effective altruism arrives from a different direction (5:07): William MacAskill and Toby Ord at Oxford, influenced by Peter Singer, applying logic and rationality to altruism — rating charities by good done per dollar, and a pledge campaign telling high earners to keep the lucrative job and give a large share away. Nick Bostrom is the join (5:57): an Oxford philosopher who was also on the Extropian list, convinced by Yudkowsky-style arguments, and one of several voices that moved EA from effective giving toward longtermism — “the most altruistic thing you could do is prevent the extinction of humanity” (6:49). He founded an existential-risk institute at Oxford where, over time, AI killing everyone became the dominant risk.

The arithmetic that did the work, from a MacAskill talk Cal places at Google (7:41): why weight hundreds of millions of present lives over the tens of trillions of future lives that extinction forecloses? “So, they pulled on this logical thread long enough, and a lot of effective altruists said, ‘Preventing superintelligent AI from killing us all, that is what effective altruism should focus.’”

Cal is careful that EA is the broader movement — “incredibly well-funded mainly by tech centimillionaires and tech billionaires,” with AI only part of the portfolio; Dustin Moskovitz is cited as giving heavily to foreign aid after USAID was gutted. But in the Bay Area the two labels have fused, which is why Cade Metz of the New York Times writes “EA/rationalist circles” as one phrase (8:32).

The creed, as Cal summarises it: superintelligent AI is inevitable; they believed this long before LLMs; carelessly built it kills us all; built right it delivers a transhuman utopia that is “the ultimate good” — and therefore deserves to be the main object of altruism.

The labs are inside the camp

The structurally important move (9:25). Many companies have hyper-scaled LLMs; Anthropic and OpenAI are distinguished by coming out of the rationalist/EA tradition and being founded to execute its programme. OpenAI exists because DeepMind — also from these circles — was bought by Google, and Musk funded an alternative on the view that “we can’t trust a big company to shepherd superintelligent AI the right way.” Anthropic is breakaway OpenAI employees, leaving two years after OpenAI’s for-profit subsidiary was declared, on the grounds that OpenAI was “not sufficiently pursuing the rationalist effective altruist goals.”

Anthropic in particular Cal calls “completely incestuously connected” with these movements (10:15): Dario Amodei lived in a group house with major effective altruists; his sister, an Anthropic co-founder, married one of them; early funding came from EA individuals including Sam Bankman-Fried and from EA investment funds.

Rather than walk the connections, he points at sources (11:05): his own New York Times op-ed from a few weeks earlier, Cade Metz on The Daily, a Michelle Goldberg column, a Wall Street Journal column from the previous week, Adam Becker’s book, and his own appearance alongside Becker on Ed Zitron’s podcast.

The civil war

What makes the picture confusing (11:55): the labs and the rest of the camp are at each other’s throats while agreeing on the premises.

“They all believe the same general things. Superintelligent AI is inevitable. It will either be a good god or a bad god. They differ, however, in whether they believe we can get the good god right now or whether we’re not ready and we need to wait.” The people at the labs think they can; the people outside think not yet.

Hence the misreading Cal wants to correct (13:39): when Nate Soares and EA figures appear around Bernie Sanders’ anti-superintelligence legislation, it looks like an outside force constraining the frontier labs. “But they all come from the same doomer mentality. They just differ on where we are on this timeline that they both agree about.”

Camp 2: the accelerationists

Defined by two beliefs (13:39): the doomers are weird, and we should go as fast as possible — the benefits are large, and there is an economic race with China to win.

And the discount to apply: “You shouldn’t fully trust the accelerationist. They have an obvious motive here, which is money.” A VC does not want impediments to returns; Huang does not want chip export bans; the White House has “a lot of connections to … the right-aligned corners of Silicon Valley.”

Camp 3: the centrists

Named only for their position on his diagram, between the other two (17:55). They want controls or safeguards, so they reject “let her rip”; they have no interest in doomer philosophy, which they find “anti-humanist and quite frankly kind of strange.” The positive programme: “regulate AI like a normal technology … like we would regulate any other technology that we’re worried about being unsafe. And if we think companies are acting weirdly, then we need to investigate and find out why.”

So they overlap both neighbours and agree with neither (18:47): the doomers’ appetite for regulation without its style or motivation; the accelerationists’ rejection of doomer ideology without their let-it-grow conclusion.

Three overlapping constituencies:

AI ethics (19:38) — a field predating LLMs, focused on proximate harms: bias in training data propagating into automated decisions; AI-driven surveillance, “which is happening um in China right now … to hunt down and persecute … minority groups” (20:29). Cal gives them priority on the ideological critique too: “they were the first to really point out early on the unsavory aspects of the futurist ideologies that drives the doomer” (21:21) — naming TESCREAL and Timnit Gebru, and characterising the argument as these philosophies have deep anti-humanist aspects, almost eugenicist: it doesn’t matter how many people die along the way if idealised transhumanist individuals survive as the apotheosis of the race.

“Normal technology” computer scientists (22:11) — his own box, with Arvind Narayanan and Gary Marcus in it, wary of hype and looking at the actual technology: “We’re less worried about superintelligence than we are about good old-fashioned straight-up harms that it could or couldn’t cause right now.” His own work: the long-horizon LLM-agent experiments OpenAI and Anthropic run and other AI companies do not — run, he believes, as part of the race to superintelligence first. Hence his call to investigate and possibly regulate — “not because I’m trying to stop the bad god from coming, but because I think these are bad actors. That their motivations are leading to dangerous behavior” (23:03). He also claims a foot in AI ethics via Georgetown’s Center for Digital Ethics.

Cybersecurity (19:38, 23:03) — not his expertise, and he reports their position as two flat statements: the superintelligence material “is a fairy tale,” and the hacking capabilities now shipping are “very dangerous for very realistic pragmatic reasons” — advanced “just so you can do well on some benchmarks or … reach some … vision you have about where you should be on the route to superintelligence.”

The payoff: two camps asking for regulation

The part Cal calls “the most confusing thing at the moment” (24:45): the centrists and the doomers are both calling for regulation, with “very different visions and motivations for what that means.”

His reading of where Congress currently sits, attributed to congressional staffers he says he has heard from directly:

The centrist objection is not that intervention is wrong but that doomer-shaped intervention misses. Normal-technology intervention asks “What are you doing? Why are you doing it? What’s your motivation?” — and, he argues, would bite the two doomer labs harder than anyone else, precisely because their conduct is the anomaly. Legislating against “superintelligence” or “recursive self-improvement” — “an idea that comes out of the futurist circles not computer science” — means missing “actual meaningful probably workable style interventions,” which should instead target specific technologies and techniques with specific safeguards (26:27). “We’re roughly aligned with the doomers who are saying we need to intervene, but we’re really not aligned on why we need to intervene and what that should look like.”

How to use the map

The closing instruction (27:19): whatever you are reading — article, tweet, Bluesky thread, podcast, “a buddy of yours at a dinner party” — “you got to first categorize that person somewhere in here to understand what they believe, why, and what their agenda is.”

And a disclaimer of the authority the map might otherwise claim: “I’m not going to tell you right now which of these you should agree with. I obviously I … I’m a centrist, but that doesn’t mean I’m right. Doesn’t mean I’m right. I don’t even really want to be a part of this debate, to be honest. I find it very stressful. But I do think clarity really matters.”

Method and caveats