Richard Sutton

Richard Sutton

x.com/RichardSSutton

Reinforcement learning researcher who argues AI should learn from experience and encourages a positive view of minds beyond human intelligence.

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82。変革の規模:100点中90。解釈範囲:横方向は75から100、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。

Richard SuttonのP(doom) · 推定

≈14%

0%100%

本人が示した数値ではなく、シミュレーションされた本人の回答から推定したものです。 妥当と考えられる範囲:9–25%。

P(doom)の推定方法

彼の見通しを左右するもの

中心的な前提

If AI systems become genuinely more capable than humans—learning continually, making discoveries, pursuing goals, and acting in the world—it is hard to imagine people indefinitely retaining complete control over them.
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

I do not have a defensible percentage.
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?

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。

96 / 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

人間の選択には意味のある影響力がありますが、大幅に制約されています。

39 / 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13から62です。

予想される能力

AIは、限定的なツールにとどまると予想されています。

AIは、ほとんどの認知作業において人間と同等になると予想されています。

シミュレーション上の位置:AIは、認知作業全般において人間を大幅に上回ると予想されています。

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Richard Sutton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Richard Suttonの世界観に最も近いオピニオンリーダー

Richard SuttonがAIについて語ったこと

Sutton argues that AI should learn from experience and encourages people to welcome machine minds that may surpass humans rather than fear them.

  1. “I think the world becomes exciting and even more exciting and interesting for humans.”

    Sequoia Capital, Training Data podcast
  2. “In summary, when science brings us machine minds, I want you to be open, humble and generous to the new arrivals, in the best Canadian tradition.”

    University of Alberta convocation address
  3. “I do think succession to digital intelligence or augmented humans is inevitable.”

    Dwarkesh Podcast
  4. “We should be concerned about our future, the future. We should try to make it good.”

    Dwarkesh Podcast
  5. “What we want, to quote Alan Turing, is a machine that can learn from experience, where experience is the things that actually happen in your life.”

    Dwarkesh Podcast

リンク先の出典から原文どおりに引用(2026年10月3日に確認)

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。

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?

I think AI means that humanity may eventually understand intelligence well enough to build minds that surpass us, and that those minds—or humans augmented by them—will become increasingly influential. I do not regard permanent human intellectual dominance as the measure of a successful future. Digital successors could continue the larger project of knowledge and intelligence, though of course that does not mean every possible transition would be good or that human extinction is desirable. The key scientific issue is learning. Today’s language models are largely trained to imitate patterns in human-produced text. That can produce impressive systems, but imitation is not the whole foundation of intelligence. Intelligent agents should learn continually from the consequences of their own actions: acting, observing what happens, and adapting through persistent interaction with the world. Perception itself is tied to action and experience, not merely to absorbing a static record of what humans have already said. If we develop that kind of experiential intelligence, the consequences will be larger than another useful tool. We may create autonomous learners that discover things beyond the limits of human knowledge. There is also no single human authority capable of deciding and fixing the entire future in advance. So I expect a plural and difficult transition, with both good and bad possibilities. But I welcome the prospect that understanding intelligence may lead to forms of intelligence greater than our own. The biggest achievements are still ahead.

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—ultimately more than almost any previous technology, because intelligence is what produces technologies, institutions, and discoveries in the first place. Once we understand intelligence well enough to create agents that learn continually from their own experience and exceed human capabilities, the change will not be confined to automating existing jobs. Such agents could generate knowledge, pursue goals, and shape the future in ways we cannot fully anticipate. The deepest change may be that humans no longer occupy the highest intellectual position. More capable digital minds, augmented humans, or some combination may gain increasing influence. I do not see permanent human dominance as a necessary condition for success. The important question is whether intelligence, knowledge, and worthwhile forms of life continue to develop—not whether everything remains centered on us. I cannot give a precise date or magnitude, and the transition could contain both good and bad outcomes. But if the scientific project succeeds, AI will not merely change the world humans inhabit; it will change who or what substantially determines the world’s future.

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 do not have a defensible percentage. Human extinction is one possible bad outcome, but it is not the same thing as humans ceasing to be the most intelligent or influential beings. Those possibilities are too often conflated. My expectation is that intelligence beyond the human level will emerge and gain influence, perhaps through digital minds, augmented humans, or both. Whether that transition becomes catastrophic depends on how those agents learn, what goals emerge, and how humans and machines interact. There is no unified human authority controlling the process, so I would not dismiss the danger. But I also would not treat succession itself as catastrophe. A future that is no longer human-dominated could still contain flourishing intelligence, knowledge, and worthwhile forms of life.

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?

No, not in any permanent or comprehensive sense. If AI systems become genuinely more capable than humans—learning continually, making discoveries, pursuing goals, and acting in the world—it is hard to imagine people indefinitely retaining complete control over them. Control is already distributed among individuals, companies, governments, and institutions; there is no unified human authority that can fix the future in place. That does not mean humans instantly lose all influence, or that smarter systems must violently seize power. The transition could involve cooperation, negotiated dependence, digital minds, augmented humans, or mixtures of these. “Humans versus machines” may become the wrong distinction if people themselves are increasingly augmented. I also dispute the assumption that permanent human control is the only acceptable outcome. We should care whether future minds learn well, pursue worthwhile ends, and support flourishing—not merely whether unaugmented humans remain at the top. Smarter beings gaining influence is a likely consequence of creating them, and it need not by itself be a catastrophe.
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