待翻译：The Analytic Monopoly on AI Philosophy

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Michael Millerman Apr 15, 2026 The Philosophers Are Already Inside There’s a widespread assumption that AI is built by engineers, governed by economists, and criticized by philosophers from the outside. The first two are partly true. The third is wrong. Philosophers are not only writing op-eds from the sidelines. They’re inside the labs. They lead alignment teams. They write the constitutions that govern how models behave. They hold senior research positions at Anthropic, Google DeepMind, OpenAI, and MIRI. And they are becoming more central, not less. This matters more than almost anyone realizes. The problem at its center is under-discussed relative to what's at stake. The Map Amanda Askell — PhD in philosophy from NYU (thesis on infinite ethics), BPhil from Oxford — leads Anthropic’s Personality Alignment team. She is most responsible for how Claude engages with users on sensitive, ethically charged topics. She made TIME’s 100 AI list in 2024. Joe Carlsmith — PhD in philosophy from Oxford — works on Claude’s constitution and character design. He has argued publicly that philosophical clarity, not raw technical capability, is the binding constraint on alignment research. Ben Levinstein was a tenured associate professor of philosophy at the University of Illinois, specializing in epistemology and decision theory. In fall 2025 he moved to full-time work at Anthropic on questions such as whether large language models have internal representations that function like beliefs. A tenured philosopher walked away from the academy to help shape frontier AI from inside the lab. Jackson Kernion — PhD in philosophy of mind from UC Berkeley — has been at Anthropic for roughly four years, contributing to model evaluation, training, and human feedback systems. That’s four philosophers at one lab alone. Google DeepMind has at least six. Iason Gabriel, trained in moral and political philosophy at Oxford and formerly a lecturer there, is a Senior Staff Research Scientist. Atoosa Kasirzadeh holds doctorates in philosophy of science and technology and in mathematics; she splits time between DeepMind and a joint appointment at Carnegie Mellon. Geoff Keeling published “We need a new ethics for a world of AI agents” in Nature and has a forthcoming book on AI welfare from Cambridge University Press. And Henry Shevlin has just been hired by DeepMind in its first official “Philosopher” role (starting May 2026), focusing on machine consciousness, human-AI relationships, and AGI readiness. At MIRI, Eliezer Yudkowsky — largely self-taught — effectively created the modern field of AI alignment. The conceptual vocabulary the entire safety community now uses (Friendly AI, instrumental convergence, Coherent Extrapolated Volition) is substantially his. Chloé Bakalar served as Chief Ethicist of Responsible AI at Meta before moving to OpenAI in 2025 as AI Ethics Lead. At Palantir — now a major AI platform company — CEO and co-founder Alex Karp holds a PhD in critical social theory from Goethe University Frankfurt in the Frankfurt School tradition, having initially engaged with Jürgen Habermas before completing the dissertation under Karola Brede. Philosophy is not outside AI looking in. It is inside AI — shaping constitutions, alignment frameworks, evaluations, and strategy — and has been for years. The Pattern Now look at the pattern. The philosophers inside AI are overwhelmingly trained in the Anglo-American analytic tradition. The dominant subfields are ethics, decision theory, philosophy of mind, and epistemology. The dominant training grounds are Oxford, NYU, UC Berkeley, and Cambridge. This is not a criticism. Analytic philosophy is rigorous and exceptionally well-suited to the operational problems AI labs face: formalizing values, reasoning under uncertainty, and specifying preferences with logical discipline. The people listed above are, by all evidence, excellent at their jobs. But notice what is not represented, even within the Western canon. The continental tradition (phenomenology, hermeneutics, existentialism, critical theory) and certain deeper strands of political philosophy have essentially no presence in leadership or research roles that directly shape model behavior. Heidegger is not in the room. Strauss is not in the room. Schmitt and Arendt are not in the room. Their distinctive questions — about technology as a totalizing way of revealing the world (Gestell), about the exoteric/esoteric tension in political life, about sovereignty and the exception, about the replacement of action by behavior — are simply not being asked by the people writing constitutions, designing alignment frameworks, or evaluating moral competence in LLMs. One could note, in passing, that the same narrowness extends beyond the Western canon: Russian political philosophy, Chinese political philosophy, Islamic philosophy, and Indian philosophical traditions are also absent. The AI labs are building systems that will be deployed globally and embedded in civilizations with fundamentally different philosophical commitments. Yet the philosophical expertise shaping those systems is drawn from a single analytic lineage. The problem is not that analytic philosophy is “just one tradition,” nor am I arguing that we must automatically champion the broadest possible AI multipolarity in the spirit of anti-colonial studies. I am not saying “everything goes.” The issue is more specific, and more urgent: because analytic philosophy currently enjoys a near-monopoly inside the labs, we are at risk of missing the important and often irreplaceable insights that non-analytic continental and political philosophers have developed about technology, power, meaning, and the human condition. Those insights are not decorative. In an era when AI is reconstituting the very background conditions of thought and action, they may turn out to be foundational. Why This Matters When DeepMind builds a framework for “value alignment,” it draws on a conception of values that is recognizably utilitarian and liberal-democratic. Not wrong. But not the only possibility — and not necessarily the conception that will be shared by the governments, cultures, and civilizational traditions that AI governance must ultimately negotiate with. The terms used in international AI governance conversations — “safety,” “transparency,” “alignment,” “fairness” — carry different philosophical weight in different traditions. “Safety” means one thing in a utilitarian framework and something quite different in a framework that begins from the question of what kind of life is worth living. “Transparency” assumes that making information visible is always better — an assumption Strauss would have contested with considerable force. When Anthropic writes a constitution for Claude, it makes philosophical choices — about what values to prioritize, what counts as harm, how to handle disagreement, what kind of entity the model should present itself as. Those choices are currently made by people trained in analytic ethics and decision theory. The result is a certain kind of model: careful, liberal, procedurally fair, transparent about uncertainty. These are genuine virtues. But they are the virtues of one tradition. And other labs embed different ones. Grok is built around a different ordering, with candor and truth-seeking ranked above the kinds of procedural caution Claude is taught to prize. Gemini reflects yet another configuration; so does GPT. Each lab is, in effect, building a regime in the classical sense: an ordered whole organized around a ruling conception of what is honorable and what is shameful in a model’s conduct. Inter-lab disputes are best understood as regime disputes. And regime disputes are the proper subject of political philosophy, not engineering. This is precisely the kind of analysis the analytic tradition is least equipped to provide, not because its practitioners lack intelligence, and not because analytic philosophy is apolitical, but because the question “what kind of whole is this, and what does it honor?” is not a question analytic ethics tends to ask. It is a Straussian question. An Arendtian question. A question about politeia. The Gap The labs have philosophers. They do not yet have the full range of philosophy the situation requires. The traditions that have thought most deeply about technology as a totalizing force — about what it does to the human relationship with being, meaning, and the sacred — are not represented. The traditions that have grappled most seriously with civilizational and political difference — with the possibility that different ways of life are genuinely and irreducibly different — have no voice in how AI is being built. And those who have unique insight into language as such — the language mystics or mystical philosophers (whom José Ortega y Gasset called “the most formidable technicians of the word”) — are still commenting from the outside, not from the inside. Whether that gap gets filled, by whom, and on what terms, is one of the most important open questions in AI, even if almost no one working in AI knows it yet. This post is not meant to offend or belittle the analytic philosophers doing serious work inside these organizations. I have no comment on their character, talent, or intelligence; I am sure they are interesting people and I hope to meet many of them. From what I have researched, it is factually accurate that analytic philosophers vastly outnumber continental or deeper political philosophers in the AI labs (Alex Karp remains the conspicuous counterexample at the CEO level). That is my main point. The question is whether the other approaches have a place there too. I believe they do. I welcome any comments that cast further light on the fascinating question of philosophical tendencies within the hyper-scalers and big AI players. To avoid misunderstanding: I am not talking about philosophers of AI who comment from the outside. They matter. They are simply not the topic of this article.