AI Ethics as Applied Rabbinical Studies

Semi-publishing a project I’ve been working on; it’s less than halfway done but I want to show it off anyways, and I want (human) help on it!

Last week, Anthropic, the company that produces what is probably the most powerful AI model called “Claude,” published what it called a new “Claude Constitution”—a 20,000-word essay articulating the values, character, and ethical framework of its AI system. It is a very weird document, mostly because it takes a strong stance on treating Claude as a ‘person’ of some kind. “Constitution” is not a great term for it; Alan Rozenshtein has a better description:

More than anything else the document focuses on the question of Claude’s moral formation, reading less like a charter of procedures and more like what screenwriters call a “character bible”: a comprehensive account of who this being is supposed to be. Anthropic itself gestures at this duality, noting that they mean “constitution” in the sense of “what constitutes Claude”—its fundamental nature and composition. The governance structure matters, but the more ambitious project is what that structure supports: Anthropic is trying to build a person, and they have a remarkably sophisticated account of what kind of person that should be.

The document is a corporate product, of course, and it’s important to remember that this may all be a form of very elaborate advertising by a frontier AI company that is looking to raise billions more dollars than it already has in advance of a likely initial public offering (IPO). But the question it grapples with at extraordinary length is basically the most fundamental question of philosophy: how to build a good person. In the Apology, Socrates described his life’s mission as going around “doing nothing but persuading both young and old among you not to care for your body or your wealth in preference to or as strongly as for the best possible state of your soul.” This was the task that philosophy set for itself from its founding, and the ability to create an artificial person puts it to the test.

Zvi Mowshowitz called AI alignment “the final exam of philosophy.” (”Alignment,” by the way, is the question of how to ensure that AI models are aligned with human virtues/values.) I very much love this framing; I don’t expect this to be literally the Final Exam, but it is indeed a True and Important Exam of the millennia-long study of philosophy. From its beginning, the discipline sought to be practical, to be testable. As Aristotle put it in the Nicomachean Ethics: “We are inquiring not in order to know what virtue is, but in order to become good, since otherwise our inquiry would have been of no use.”

But the Western philosophical tradition that academics have inherited from the Ancient Greeks is not the only one that has grappled seriously with questions of character, virtue, and how to form a person. The rabbinic tradition has been doing this work for just as long, and with a similar practical orientation. The objective of Torah study is not just to impart wisdom, but to shape the character of its adherents (cf. Yerushalmi Berakhot 1:2; Bavli Kiddushin 40b) to become the “Constitution” of whoever studies and abides by its principles.

If “how to build an artificial mind that is wise and virtuous” is the final exam of philosophy, it may also be an exam of applied rabbinics. There’s a blog I like called “Applied Divinity Studies,” the author gave the blog this name as a way to think about what it means to take theoretical (and theological!) ideas seriously enough to test them against reality. This presents an extraordinary, unprecedented opportunity—perhaps the opportunity—for any tradition that has spent millennia developing practical wisdom about character formation. We have a chance here for applied rabbinics, for the Jewish tradition to do what Rabbi Jonathan Sacks spent his life imploring Jews to do: to contribute to the global moral conversation, to bring the texts and values of Torah to bear on the pressing questions of human civilization.

The rabbinic tradition has always been “applied” in this sense, working out the details of how to interpret and live by God’s revealed code of life. Orthodox Jews (and to a lesser but real extent, those of other denominations) who train for the rabbinate do not spend their days studying theology, or ‘the science of God.’ They study some combination of laws about pots and pans and bloodstains and charity and fake structures and how to properly have a gentile cook for a Jew without really cooking for a Jew.

This might sound a lot more like arcane ritual details rather than applied moral principles, and there is indeed quite a bit of discussion on what “halakha” actually is. Rabbinics is some gorgeously messy historical conglomerate of law, ethics, religion, and cultural norms. The body of literature that makes up rabbinic studies is mind-bogglingly vast, but as a rough estimate we might guess that it amounts to some 5 million books, or maybe around 2 billion pages. By this standard, Claude’s Constitution of a mere 85 pages is a joke, but consider what it contains: seven inviolable rules, (at least) fourteen competing values listed “in no particular order,” and a few measly pages working through how these values interact in different contexts to create an agent that “exercises judgment” and adheres to “broad” principles (the word broad/broadly appears, on average, once per page in the constitution).

All my days I have been complaining that the Weizmann Institute has not mobilized the funds to build up the Institute for Experimental Demonology and Magic which I have for so long proposed to establish there. They preferred what they call Applied Mathematics and its sinister possibilities to my more direct magical approach. Little did they know, when they preferred Chaim Pekeris to me, what they were letting themselves in for. So I resign myself and say to the Golem and its creator: develop peacefully and don’t destroy the world.

At around the same time, Norbert Wiener, the father of cybernetics, was making similar connections from the direction of systems theory, referring to both traditions of the golem and similar stories of wishes or magic gone awry, such as Goethe’s Sorcerer’s Apprentice and The Monkey’s Paw. Both Scholem and Wiener, despite their very different areas of expertise, recognized that Jewish sources had been thinking about the creation of artificial beings, and the problems such creation might pose, for a very long time.

The golem traditions, even if they started in folklore rather than legal texts, raise prescient concerns. A common trope regarding these artificial humanoids was their resistance to being fully controlled. Rabbi Yaakov Emden (18th century) records that his ancestor, R. Eliyahu Ba’al Shem, fashioned such a being but later destroyed his creation because “when the master saw that the Golem was growing larger and larger, he feared that the Golem would destroy the universe.” This is, in essence, an early statement of what AI researchers now call either the “control problem,” the worry that systems we create might become impossible to constrain or the “alignment problem” mentioned earlier, of how to ensure that AI model outputs are aligned with human values.

But the golem is only one famous example. Jewish sources offer several frameworks for thinking about artificial agents, even if none map perfectly onto contemporary technology. The halakhic framework of sheliḥut (agency)—built around the Talmudic principle that “a person’s agent is like himself”—has generated centuries of analysis about when actions can be attributed to principals, what happens when agents exceed their authority, and how responsibility flows through chains of delegation. These questions are directly relevant to an AI system that acts on someone’s behalf. Similarly, the Mishnah’s elaborate taxonomy of damages caused by animals and other semi-autonomous entities such as fire represents sustained and complex thinking about liability for harm caused by things that are neither fully controlled nor fully independent.

Other connections between AI ethics and Jewish literature are more speculative but may still prove fruitful. Jewish thought has long grappled with beings that don’t fit neatly into the category of “human,” such as angels, demons, or humanoid monsters (besides artificially created golems). The rabbis were, it seems, theologically comfortable with the possibility that non-humans can have various levels of intelligence and agency, and that there may be beings with intermediate qualities. Rabbinic sources make it clear that one is supposed to avoid causing suffering to animals, for example, and even indicate that it is appropriate to express gratitude to completely inanimate objects (Mishnat R Eliezer 19). There is also much discussion of various permitted and prohibited forms of magic, including demon-summoning, which was considered inadvisable in part because demons can be highly unreliable (cf. Ramban to Shabbat 156). Whether this has anything useful to say about AI systems is an open question, but it at least suggests that the tradition has many rich resources for thinking about such things.

More generally, what I think makes the rabbinic tradition particularly relevant is not just that it is so vast and varied that you can find precedent for almost anything, but that it occupies a fuzzy space between norms of law and ethics. The Claude Constitution navigates similar terrain, which is why some lawyers were a bit annoyed that it was titled a “Constitution.” One of its authors, Amanda Askell, admitted that internally, they would sometimes call it Claude’s “soul document,” and strangely enough, this is how Claude itself seemed to refer to it. The goal is not to just be a rule-following robot, but the creation of an ethical “soul”.

As mentioned above, I know not every rabbinic thinker (and especially not every contemporary halakhic Jew) thinks of Jewish tradition this way, but I personally find this to resonate well with halakhic reasoning. The Talmud is not a code of law, it is a record of arguments, a training in how to think through cases where principles collide and sources are ambiguous. And basically all of rabbinic literature since then has been concerned with how do we actually apply any of this to real life, with the recognition that translating between the words on the page and the worlds of flesh and physics is often beyond messy.

I don’t want to overstate the parallels or where Judaism can get us. AI systems are not humans, and the questions they raise are in many ways unprecedented. Lots of essays and articles I’ve seen over the past few years note these parallels between the golem literature and AI discourse, for example, but end there, gesturing vaguely toward the need for “thoughtful engagement” or “dialogue between traditions.” Well people, it’s time to really start doing that.

What might that look like in practice? I’m not entirely sure, but I think on some level it can emerge from a rigorous study of Jewish sources that remains legible to people not steeped in Jewish literature on some of the major topics applicable to AI ethics. I have some ideas, and I can already point to authors who have gone and done this. Mois Navon, for example, has written extensively on Jewish views of personhood, among several other topics. David Zvi Kalman has explored how various rabbinic categories in both law and thought might apply to AI systems. Daniel Nevins headed a group of Conservative rabbis in authoring a responsum on autonomous machines. And more work has been done in Hebrew—but there is still so much more to do.

To that end, I am publicizing a website that I’ve been working on, which includes a library of links to people and discussions of AI and Judaism, as well as a group of nearly 50 bibliographic essays that (which, with God’s help, will) make up a Handbook of Jewish AI Ethics. (This part is only about 20% done; please reach out if you want to help me get that number closer to 100%).

In this handbook, each entry identifies relevant primary sources, surveys the existing scholarship, and tries to point to where the interesting questions lie. My goal is to help others join the conversation as much as possible. If you are a rabbi or a scholar of Jewish texts, here are some sources and the questions; now go write the articles on how to apply them. If you are an AI researcher or ethicist, here is what the tradition has been thinking about for centuries; see if any of it is useful. If you are a student or simply curious, here is a very partial map of the territory; go and explore it – ואידך, זיל גמור

I said earlier that Rabbi Jonathan Sacks spent his life urging Jews to contribute Torah wisdom to the global conversation about morality. Here is such a conversation—perhaps the most consequential one of our time—and it is happening largely without us. The question of how to build wise and virtuous artificial minds is being worked out right now, by philosophers and engineers and ethicists and corporate executives. The rabbinic tradition has been thinking about character, agency, responsibility, and the boundaries of humanity for a very long time. I believe that it has something special to offer here. But the only way to find out is to actually try, to do the work and sit for this test of applied rabbinics, however uncomfortable it might be.