Economics blogger who expects AI to transform science, argues people can keep valuable jobs and worries about AI-enabled bioterrorism.

AIは世界をどのように変えるでしょうか?

文明規模の変化漸進的な変化DoomBloom
シミュレーション上の位置解釈範囲

横軸:彼が表明したDoom–Bloomの見通し。 縦軸:変革の規模。

Doom–Bloom:100点中47。変革の規模:100点中87。解釈範囲:横方向は42から75、縦方向は75から100。これらは解釈上の座標であり、事象の確率ではありません。

Noah Smithが示したP(doom)

≈10%

0%100%
“about a 10% chance of bringing down civilization”

Civilization collapse from AI-enabled bioterrorism; not human extinction

Here’s how we’re all going to die · 2026年8月

彼の見通しを左右するもの

中心的な前提

If we give an AI tools, credentials, memory, and permission to execute a long project, then a malicious human can use it to do enormous damage even if the system remains perfectly obedient.
回答4

この前提が実際には異なると判明した場合、彼の見通しはどう変わりますか?

未解決の問い

The key uncertainty is whether the relevant biological capabilities are actually feasible.
回答3

ここで考えられる結果を彼が見分けるうえで、何が役立ちますか?

考えを変え得るもの

If experts demonstrated that AI agents still cannot materially help design or execute a catastrophic biological attack—or that systems reliably refuse dangerous requests even under determined attempts—I would sharply reduce my catastrophe estimate.
回答5

どのような証拠なら十分で、それによって彼の見解はどちらの方向に変わりますか?

詳細

予想される恩恵

変革をもたらし、広く価値のある恩恵が予想されています。

97 / 100

影響が小さい変革をもたらす影響

質的尺度での解釈範囲は100から100です。

予想される害

複数の解釈が依然として妥当です:深刻または広範な害が、予想される将来の実質的な一部となっています。 / 破局的または不可逆的な喪失が、予想される将来の中心となっています。

81 / 100

影響が小さい変革をもたらす影響

質的尺度での解釈範囲は67から100です。

人間の影響力

回答に基づく暫定的な推定です。より広い範囲は、ほかにあり得る解釈を示しています。

53 / 100

影響力が小さい影響力が大きい

質的尺度での解釈範囲は23から100です。

これらの解釈では、彼が示した条件が維持されています。恩恵と害は、どちらも大きくなり得ます。この範囲は、統計的な信頼区間ではなく、彼のシミュレーションされた回答をどのように読み取ったかを示すものです。

あなたはNoah Smithと比べてどの位置でしょうか?
約3分で自分のAIに対する世界観をマッピングして、比較できます

似ている世界観

シミュレーションされた世界観がNoah Smithの世界観に最も近いオピニオンリーダー

出典

このシミュレーション対象者の根拠として使用された記事、インタビュー、著作です。

My thoughts on AI safety

The accessible introduction calls the consequences of godlike AI deeply unknowable. The remainder is paywalled; later public essays and interviews supply the substantive current position.

noahpinion.blog
Updated thoughts on AI risk

Publicly revises his earlier chatbot-focused dismissal after observing agentic coding. Autonomous tool use and economic pressure to remove humans from loops make catastrophic misuse more plausible. Remains comparatively skeptical of a near-term robot takeover.

noahpinion.blog
Superintelligence is already here, today

Argues that combining human-like reasoning with computers’ exceptional recall and calculation already creates superhuman capabilities. Expects major scientific transformation; this is his broad definition of superintelligence, not a claim that every system is an autonomous omnipotent agent.

noahpinion.blog
Noah Smith returns to explain his P(doom) update — Doom Debates

Speaker-labeled transcript: Noah at 07:21 endorses roughly 10% for catastrophic biological misuse. At 08:55 he expects survivors and is unsure about rebuilding. At 09:25 he declines to endorse the interviewer’s proposed 5% permanent-doom number. At 12:26–14:59 he identifies biological feasibility and resistance to jailbreaks as updates. At 17:49–18:18 he explains why malicious human use dominates his concern. Do not import Liron’s forecasts into Noah’s answers.

lironshapira.substack.com
Plentiful, high-paying jobs in the age of AI

Clarifies that plentiful employment under extremely capable AI is possible, not guaranteed. Relative costs and constraints on AI supply can preserve human comparative advantage even when machines outperform people at every task.

noahpinion.blog
Your future job will be to keep AI on task

Emphasizes humans’ comparative advantage in knowing what they want, with future work organized around directing AI. The accessible introduction supports this thesis; avoid inventing detailed empirical findings from inaccessible portions.

noahpinion.blog
Here’s how we’re all going to die

Assigns an informal 10% chance to AI-enabled bioterror bringing down civilization and 30% to world-changing destruction, without a precise horizon. Calls this much more serious than other AI harms while retaining enthusiasm for AI. These are not extinction odds; a separate annual-risk example is hypothetical, not his forecast.

noahpinion.blog
AI keeps stubbornly refusing to take our jobs

Distinguishes replacing tasks from eliminating occupations. Interprets contemporary employment evidence as inconsistent with an imminent job apocalypse; expects humans to continue directing AI, without claiming that no occupation will ever disappear.

noahpinion.blog
Two missing pieces in the AI safety discussion

Argues that safety advocates need concrete threats that persuade skeptics and must address Chinese leaders’ incentives for coordinated pacing. Revises his earlier idea that merely staying ahead would produce cooperation: domestic threats from China’s own models could instead change its incentives.

noahpinion.blog

シミュレーション評価

質問1

AIは私たちの未来にとって何を意味すると思いますか?また、なぜそう思いますか?

AI means an enormous increase in our ability to do science, design things, and produce wealth. Combining human-like reasoning with computers’ memory, speed, and calculation is already a kind of superhuman capability. And the standard objection—“look, the chatbot said something stupid”—is increasingly irrelevant once agents can use tools and carry out long, multi-step projects. I also reject the lazy jump from “AI can perform this task” to “humans will have no jobs.” Jobs are bundles of tasks, and humans still know what they want. Directing agents, choosing goals, checking results, and adapting output to real human needs can itself be valuable work. Relative cost matters too: even if machines outperform us technically, limited or expensive machine intelligence can leave humans with comparative advantage. That makes continued employment plausible, not guaranteed. But the biggest issue is catastrophic misuse. I’m less worried about an alien machine spontaneously becoming a machine god than about a malicious person ordering an obedient, highly capable agent to complete a dangerous project. Biological misuse is the nightmare because agentic systems could potentially help with an end-to-end effort, not merely answer isolated questions. I’m not a virologist, so the feasibility is a crucial uncertainty. Still, I put roughly a 10% chance on AI-enabled bioterror bringing down civilization and about 30% on world-changing destruction. Those are rough judgments, not extinction probabilities. So my view is simultaneously enthusiastic and alarmed: AI could deliver a scientific and economic transformation while preserving valuable human work, and it could also give terrible people unprecedented leverage. Anyone treating only one side as real is missing the future.

質問2

AIは最終的に世界をどの程度変えると思いますか?

Enormously—probably on the scale of the biggest general-purpose technologies, because it expands the supply of reasoning itself. Systems that can search, calculate, code, design, and run long projects could accelerate science and production across much of the economy. That does not mean benchmark gains instantly become factories, medicines, or functioning institutions; physical deployment and social adaptation remain bottlenecks. Nor does enormous change imply that humans become economically useless. Replacing tasks is not the same as eliminating occupations. People still choose goals, direct systems, judge whether results fit human needs, and operate where machine intelligence is costly or constrained. Some jobs will disappear and many will be reorganized, but “AI gets better, therefore nobody works” skips the actual economics. The darker possibility is equally world-changing. Capable agents could let malicious humans execute dangerous projects at a scale they could not manage alone, especially in biology. That could mean not just disruption but civilizational catastrophe. So I expect AI to transform science, production, and work—and I think there is a frightening chance it transforms the world through mass destruction instead.

質問3

AIが人類絶滅、または同様に永続的な破局を引き起こす可能性は、直感的にどのくらいだと思いますか?

I don’t have a useful number for human extinction or permanent collapse. My rough estimate is about a 10% chance that AI-enabled bioterror brings down civilization, but that is not a 10% extinction estimate. I expect there would probably be survivors; whether they rebuild is a separate question I cannot quantify responsibly. I’d also put roughly 30% on world-changing destruction from biological misuse. Again, that covers outcomes far short of extinction. The mechanism I worry about is not a machine god deciding to erase humanity. It is a malicious human using capable, obedient agents to carry out an end-to-end biological project. The key uncertainty is whether the relevant biological capabilities are actually feasible. So I’m very alarmed, but I’m not going to convert that alarm into a fake precise permanent-doom number.

質問4

人間より賢いAIシステムを、人々が引き続き制御できると思いますか?また、なぜそう思いますか?

“Smarter than humans” is too vague to answer with a clean yes or no. A system can be superhuman at coding, scientific reasoning, or planning while still functioning as a tool that follows human goals. Intelligence does not automatically generate an independent desire to seize power, and it certainly does not erase physical bottlenecks, institutional constraints, or human control over infrastructure. But agentic systems make control substantially harder. If we give an AI tools, credentials, memory, and permission to execute a long project, then a malicious human can use it to do enormous damage even if the system remains perfectly obedient. That is the control problem I find most concrete: not necessarily losing control *to* the AI, but losing control of what humans can accomplish *through* AI. So yes, I expect people can retain meaningful control over many systems smarter than themselves. I do not expect that control to guarantee safety. A compliant superhuman agent in the hands of a terrorist could be more dangerous than a mildly unruly one. The important distinction is between controlling the system’s behavior and controlling the dangerous capabilities it gives to people.

質問5

どのような発見や出来事があれば、AIが将来もたらす影響についてのあなたの見解が最も大きく変わりますか?

The biggest update would come from biology. If experts demonstrated that AI agents still cannot materially help design or execute a catastrophic biological attack—or that systems reliably refuse dangerous requests even under determined attempts—I would sharply reduce my catastrophe estimate. Conversely, a credible demonstration of an agent autonomously completing most of that end-to-end pipeline would raise it dramatically. That matters far more than another chatbot saying something embarrassing. On the economic side, I would update if widespread deployment actually produced sustained mass unemployment rather than merely changing tasks and occupations. I’d especially want evidence that humans cannot retain useful roles directing, checking, or complementing agents even when machine intelligence remains costly or constrained. Until then, “the model beat a human on a benchmark, therefore humans are economically obsolete” is not an argument; it is a missing argument. A truly autonomous system developing persistent goals, acquiring resources, and evading human control would also force me to take the machine-takeover scenario much more seriously. But right now the clearest path to catastrophe is simpler: a dangerous human using an obedient agent.
あなたはどの位置でしょうか?
いくつかの簡単な質問に答えて、自分のAIに対する世界観を探ってみましょう。
自分の世界観をマッピングする

あなたはどの位置でしょうか?

自分の世界観をマッピングする