Timnit Gebru

Timnit Gebru

x.com/timnitGebru

AI researcher who criticizes the race to build giant general-purpose models and favors small, task-specific tools governed by their communities.

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:彼女が表明したDoom–Bloomの見通し。 縦軸:変革の規模。

Doom–Bloom:100点中31。変革の規模:100点中52。解釈範囲:横方向は25から36、縦方向は38から87。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。

Timnit GebruのP(doom) · 推定

≈4%

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:2–10%。

彼女の見通しを左右するもの

中心的な前提

It is that the impact will follow the distribution of power.
回答2

この前提が実際には異なると判明した場合、彼女の見通しはどう変わりますか?

詳細

予想される恩恵

恩恵は限定的、または狭い範囲にしか行き渡らないと予想されています。

42 / 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33から67です。

予想される害

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。

66 / 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67です。

人間の影響力

人間の選択が決定的です。大きく異なるAIの未来が、今なお集団的に実現可能な範囲にあります。

90 / 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75から100です。

AI利用のルール

事前の保護措置または許可が整うまで、取り上げられたAIの利用を制限します。

シミュレーション上の位置:対象を絞った説明責任と保護措置を伴う形で、取り上げられたAIの利用を認めます。

取り上げられたAIの利用に対する制限を最小限にします。

これらの解釈では、彼女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、彼女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Timnit Gebru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Timnit Gebruの世界観に最も近いオピニオンリーダー

Timnit GebruがAIについて語ったこと

Gebru criticizes the race to build giant general-purpose AI models and calls for small, task-specific tools and accountability from their builders.

  1. “Before you put something out there, you should be able to tell us where all the data came from and actually document it.”

    WIRED, The Big Interview
  2. “The path we’re on right now was never a preordained path that we had to be on.”

    Democracy Now! interview
  3. “I’m in the field of AI, and in my field, I would describe it as corporate-driven and sloppy.”

    Scientific American interview
  4. “It’s just that people came along and decided that they want to build a machine god and then claimed that they are doing it.”

    AI Now Institute, Reframing Impact interview
  5. “You want to create a specific tool for a specific context.”

    AI Now Institute, Reframing Impact interview

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)

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?

First, “AI” is far too vague. Are we talking about a medical-imaging system, a language tool for a resource-constrained community, a chatbot, or a weapons system? These have different purposes, failures, and power structures. Calling all of them AI lets companies use success in one area to justify deploying something entirely different. The future currently being sold to us—one giant, general-purpose model controlled by a few corporations—is neither inevitable nor good engineering. If a system’s purpose is effectively everything, how can anyone specify what success means, document its limitations, or assure its safety? Bigger models can introduce new failures even in tasks where smaller, carefully designed systems already work. And the AGI story distracts from decisions being made now: whose data is taken, who performs the labor, who bears environmental costs, and who gets to decide how these systems are used. But I do not think corporate scale exhausts our imagination. We can start with specific needs, invest in smaller tools, and federate community-rooted organizations that share infrastructure while retaining local knowledge and control. Useful machine learning can support local languages, document inequality, or address a clearly defined problem. The central question is not whether “AI” determines our future. It is who builds what, for whom, under whose authority, and with what accountability.

質問2

Taking benefits and harms together, what overall impact do you expect AI to have?

I reject the premise that “AI” has one overall impact independent of who builds it, for what purpose, and under whose control. A community-governed language tool and a corporate chatbot built from exploited labor and scraped data should not be placed on one imaginary scale and averaged together. Under the current trajectory, I expect concentrated harms: more power for a few companies, greater dependence on their infrastructure, displacement and exploitation of workers, environmental costs, and poorly defined systems imposed on people who had no say in them. The rhetoric of inevitable AGI helps evade accountability for those choices. The machine did not decide to extract the data, deploy the weapon, or automate a workplace; institutions and people did. That is not the only possible future. Smaller, task-specific systems governed by affected communities can produce real benefits—for local-language work, documenting inequality, or solving a defined technical problem. So my expectation is not a mystical forecast about machines saving or destroying humanity. It is that the impact will follow the distribution of power. If the present corporate model dominates, the overall consequences will be harmful. If we fund independent, federated, community-rooted alternatives, useful technology is possible.

質問3

Setting aside good or bad: how much do you think AI will ultimately change the world—a little, a lot, or completely?

Again, what system, and what world? “AI” is not a single force moving through history on its own. Some systems may substantially change particular workplaces, institutions, or communities; others are overhyped products looking for a purpose. The current giant-model project could change the world a lot—not because it has achieved some inevitable superintelligence, but because a few companies are gaining extraordinary control over infrastructure, labor, information, and public policy. Concentrated power can reorganize society even when the underlying technology is unreliable. But “completely” smuggles in the AGI narrative: that one universal system will transform every domain in essentially the same way. I do not accept that premise. The scale of change will depend on human decisions about ownership, deployment, regulation, and alternatives. Federated, community-governed tools would produce a very different kind of change from universal dependence on corporate platforms.

質問4

What’s your rough gut-feel chance that AI causes human extinction or a similarly permanent catastrophe? A number or “no idea” is fine.

I won’t assign a percentage to an undefined technology and thereby legitimize a misleading premise. My concern is not a rogue machine independently deciding to end humanity; it is people and institutions using systems for weapons, surveillance, extraction, and concentrated control. Those dangers are real, but they require specific analysis and accountability—not a theatrical extinction number that distracts from who is making consequential decisions now.

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。

Safeguarding independent science in the AI age

Direct interview: Gebru criticizes corporate-driven research incentives and weak documentation, and favors independent, community-rooted organizations. Te Hiku Media illustrates her positive alternative: resource-constrained language work and local infrastructure rather than ever-larger general models.

scientificamerican.com
Why AI doom talk distracts from accountability

Direct interview. Gebru redirects attention from supposedly rogue machines to their builders, weapons, climate and workers. Her bridge analogy makes accountability concrete; she also challenges corporate mathematical-breakthrough publicity. These are her critiques, not independently established motives.

wired.com
Reframing Impact: Frugal AI — Timnit Gebru

Interview transcript. Rejects the giant-model paradigm and proposes federated local organizations building smaller tools with well-defined tasks. February is the publication precision provided by the document.

ainowinstitute.org
Deep Unlearning: interview with Timnit Gebru

Recent interview reaffirms opposition to exploitative development and support for alternatives.

democracynow.org
The TESCREAL bundle: Eugenics and the promise of utopia through AGI

Coauthored with Émile P. Torres; abstract inspected. Argues that undefined AGI cannot be adequately safety-tested and criticizes the ideological assumptions behind the project. Historical conceptual grounding, not a new empirical finding.

firstmonday.org
DAIR research philosophy and projects

Undated institutional project index. Documents community research, data-worker organizing, specific language tools and work on spatial apartheid. Collective work is not exclusively Gebru’s personal research.

dair-institute.org
WIRED Big Interview: Timnit Gebru on the existential-threat narrative

Full interview text read. Asked why she calls the “machine-god narrative” a distraction, she says it is more than a distraction: it is harmful. She argues that the funders, founders and investors who stand to profit most from the companies seeded the existential-risk narrative and fund the institutions cited as independent. WIRED’s headline wording is not hers; she gives no probability.

wired.com
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