An essay
The Noun Trick
How the Seven “Settled” Debates in the Jewish AI Discourse Hide Their Own Questions
Every one of the seven “already said well” debates is a trick performed in grammar. The titles look like questions. They are not questions. They are smuggled nouns dressed up as questions, and the work that would make them into actual questions — defining each term — has not been done.
“Will AI replace the rabbi?” sounds like a question because it contains a verb. But every noun in it — “AI,” “the rabbi,” and the implicit function under “replace” — is a portmanteau of practices, roles, technologies, and relationships, none of which has been pinned down. The headline asks nothing, because its words do not refer to anything stable enough to be asked about.
This is not a quibble. It is the reason the discourse feels exhausted while the underlying questions remain untouched. The first-order arguments are not exhausted. They have not been made. What has been exhausted is the recycling of headlines whose nouns have not been opened.
Below is what each of the seven headlines hides.
1. “Will AI replace the rabbi / posek?”
The headline has four contested nouns and one contested verb, and pretends each is settled.
“Rabbi” is not a role. It is a cluster of incompatible roles — posek, pastor, teacher, communal manager, marriage officiant, biographical witness, scholar — that almost no rabbi in history has performed evenly. To ask whether AI replaces “the rabbi,” you must first pick which of these roles you mean. The discourse never picks.
“Posek” is not an act. It is a chain: research, ruling, accountability, relational reading of the questioner, willingness to be wrong in public. Each link can be performed by different agents. The headline treats psak as if it were a single output.
“AI” is not a thing. It is a continually shifting set of systems with different capacities, training data, alignment regimes, and failure modes. The AI of 2023 and the AI of 2026 are not the same noun, and the AI of 2028 will not be either.
“Replace” is not a single operation. It can mean substitute, augment, render obsolete, supplement, or reorient. The headline collapses all five into a binary.
Until each of these is unpacked, the question is not a question. It is a vibe wearing a question mark.
2. “Beit Midrash + Shabbat as Structural Safeguards”
Koppel’s thesis names two practices as if they were architecture. They are not architecture. They are practices, and practices vary across communities and across time.
“Beit Midrash” is not an institution. It is a wide spectrum — Brisker yeshivot, Hasidic shtieblach, women’s batei midrash, Mechon Hadar, Conservative seminaries, online chavruta networks. Each operates on a different ontology of learning. To say “the Beit Midrash” safeguards Judaism is to pick one and pretend it is all of them.
“Shabbat” is not a uniform practice. The Shabbat of a Modern Orthodox family in Teaneck, a Haredi family in Bnei Brak, a Conservative family in Boston, and a Renewal community in Berkeley are different phenomenological objects. Whether any of them safeguards against AI depends entirely on which Shabbat you mean.
“Structural safeguard” presupposes a stable threat model and a stable defense. Technologies do not threaten practices from the outside; they reshape them, and are reshaped by them. Asking whether Shabbat safeguards against AI is like asking whether the printing press safeguards against literacy. The question takes for granted that the practice and the technology are separable, when in fact each is always remaking the other.
The thesis is elegant precisely because it skips this work.
3. “Ameilut ba-Torah in the Age of Instant Answers”
“Ameilut” is not one virtue. Brisker chiseling, Hasidic devotional immersion, Mussar interior labor, yeshivish iyyun, baal habayis shiur attendance — each tradition means something different and incommensurable by ameilut. To ask what AI does to ameilut is unanswerable until you say whose ameilut. The disagreement about what toil in Torah even is has organized Jewish learning for generations; the headline erases it in a single word.
“Instant answers” treats Torah study as information retrieval. The very schools that put ameilut at their center never treated learning as retrieval. The phrase imports a model of study (Q&A) into traditions that explicitly do not operate that way.
“The age of” presumes an epochal break. Most ameilut traditions developed under earlier technological transformations — the codex, print, the responsa literature, daf yomi, Bar Ilan, ArtScroll — and absorbed each one. The framing pretends AI is the first.
4. “AI Antisemitism in LLMs / Algorithmic Bias Against Jews”
“Antisemitism” is not one phenomenon. Content-level slurs, structural exclusion, anti-Zionism that crosses into Jew-hatred, classical religious anti-Judaism inherited from Christian and Muslim training corpora, and statistical underrepresentation are all different mechanisms with different causes and different remedies. Treating them as one thing is the move that makes the headline possible.
“Bias” implies a neutral baseline. Whose Jews are the baseline? Israeli Jews? American Reform Jews? Haredi Jews? Mizrahi Jews? Each looks like bias from the perspective of the others.
“LLMs” flattens vastly different systems — closed and open weight, English-trained and multilingual, RLHF’d in different jurisdictions — into a single noun.
“Algorithmic” hides the human labelers, the contracted RLHF workers, the corpus curators, and the policy teams. The bias is human all the way down. Calling it algorithmic relocates moral responsibility into a noun where no one lives.
5. “The Golem as AI Precedent”
“The golem” does not exist. There are golems — the Sefer Yetzirah golem, the Maharal golem, the Hasidic golem, the Yiddish theatrical golem, Singer’s golem, Wegner’s golem, Cynthia Ozick’s golem. Each is the product of its specific historical and theological moment, and each was built (literarily) to do different work in its tradition. Lumping them into “the golem” deletes the actual content of the tradition.
“Precedent” is a legal-historical category. Literary motifs are not precedents. To call the golem a precedent for AI is a category error that imports a juridical structure where there is only metaphor.
The headline collapses centuries of distinct golem traditions into one figure and then asks whether that single figure predicts AI. The answer can only be yes or no — both wrong, because the question is malformed.
6. “AI Translation Tools for Torah and Jewish Scholarship”
“Torah” is not one corpus. Tanakh, Mishnah, Talmud Bavli, Talmud Yerushalmi, Midrash, Hasidut, Halakha, Aggadah, Kabbalah — each has its own translation history, vocabulary, and conventions. Each pulls AI in a different direction.
“Jewish scholarship” names a vast and internally disagreeing set of disciplines. Treating it as a unified object is what the headline needs you to do for the question to even seem coherent.
The headline asks whether AI translates well. It cannot answer this without first asking what well-translated means in each tradition, and that question is ancient.
7. “AI Sermons / The Debate Over AI-Written Divrei Torah”
“Written by AI” is a spectrum, not a binary. Outline by AI, first draft by AI, edited by AI, polished by AI, prompted by AI, brainstormed with AI — all sit under the same headline and all do different things to the speech act. The headline pretends authorship is on/off.
“Divrei Torah” is a category whose authenticity-conditions are themselves contested. Some traditions hold that any insight that opens the text counts. Others require lineage, citation, or chazara from a teacher. Without saying which, the debate has no object.
What This Means
The reason these seven debates feel settled is that the headlines have done the work of pretending the nouns are settled. Once the nouns are opened, none of the seven debates is over. None of them has been started.
The move for any new entry is not to find a clever angle on a settled question. It is to refuse the noun trick. Pick one of the smuggled words. Open it. Show what it actually contains. The question that emerges from that work will be a real question — and the answer will be the kind of thing the reader cannot un-see.
· 7 min read