Finally, a Jewish AI debate has emerged

The field is broadening and attracting more thinkers. Here's where we are.

Hello readers! If you haven’t read my previous post, Jello Menorah is now accepting subscriptions. There are no paywalls or bonus posts; subscriptions simply help me justify the time I spend building up this platform. I’d like to spend more energy writing here and this makes that possible. I hope you becoming a paying member today.

Also—I do speaking gigs on Judaism/tech/futurism topics all the time. Want to bring me to your community or for an event? Drop me a line.

One of the surest signs that a field isn’t coalescing is that the people in that field aren’t reading each other. As I’ve written previously, this has often been the case for Jewish thinking about AI, where ideas often repeat without any indication that the writers know they’re not being original. It also means that nobody is tracking whether interesting ideological fault lines are emerging. It’s just a sea of hot takes.

As a “veteran” of the field of Jewish AI thought, I see my role as helping to correct this problem by describing the emerging intellectual landscape as I see it. To that end, I want to spend a little time describing some of the advances that I’ve seen over the last few months (thank you to the many many people who send me things!). Slowly but surely Jewish writing about AI is expanding beyond the small pool of early adopters to a larger group of thinkers who are generally interested in Judaism in modernity (though the word “modernity” feels very 20th century, doesn’t it?)

Here’s my impression of the emerging frontline.

Can AI give psak?

Before the advent of high-quality generative AI, most Jewish discourse about AI focused on decision-making capabilities, like autonomous weapons and self-driving cars. Now a new question has emerged: what to do with AI’s ability to analyze Jewish law? This issue was broached quite early on the Headlines podcast (a fascinating snapshot in its own right!), and it’s now being taken up more broadly.

In December, Yeshiva University—America’s flagship Modern Orthodox institution—devoted half a dozen articles in its Hanukkah edition of Torah To Go to articles about AI. Two of the articles focus on psak, i.e. whether an AI can or should decide matters of Jewish law.

Perhaps predictably, both articles discourage the use of AI for psak. The reason for this, as articulated by Netanel Wiederblank, is that psak isn’t just calling balls and strikes (to quote Justice John Roberts), but must instead be deferential to human needs and to empower human beings. Wiederblank writes:

Hashem [=God] could have just told us what to do prophetically, but He didn’t. Why? Because He trusted us, He wanted to empower us. We must not forfeit this responsibility. Every day we pray not just for wisdom but ten chelkeinu b-torasecha, give us our portion in your Torah. If we were to hand over this responsibility to a computer it wouldn’t just be wrong—it would be a tragedy.

This question also lies at the heart of an essay by Michael Broyde, a rabbi and professor at Emory’s law school who has been writing about contemporary Jewish law for decades. While writing an article about whether a kohen in a same-sex relationship can offer the Priestly Blessing,¹ Broyde decided to give the same sources he was working with to ChatGPT and ask it to do its own legal reasoning. ChatGPT, unsurprisingly, did fairly well.

In a write-up of this experiment, Broyde remains open to the possibility of AI psak, but notes that it requires some reflection on what you want psak to be.

So, in dealing with the subject of whether excellent AI will ever be able to issue psak, the real question is “what is psak?” To those who consider psak to be the application of cold and dry rules to clear and established facts, the answer will eventually be “yes,” even as such is not the case yet. To those who have a broader understand of what it means to be a posek [=legal decisor], and what exactly psak is, AI has much farther to go that just to be able to master the sources – it will have to develop a way to understand the needs of the questioner, and the implications to society of answers to questions, and so much more than merely the sources.

Now, what to make of this hesitance among Orthodox thinkers? Personally, I find it fascinating that AI’s comprehension of Jewish law is leading rabbis to double down on charismatic authority over textual authority. In this respect AI is simply the ironic capstone to decades of efforts to make Torah as accessible as possible. This effort has been wildly successful, but in our decentralized religion it has also made it completely unclear how textual arguments should be valued in making practical decisions, or who has the right to make norm-setting textual arguments.

AI further complicates this picture because this accessible knowledge isn’t just sitting in an online database but is now “embodied” in an LLM. Previously, Torah’s accessibility gave the layperson the ability to look up sources and come to their own conclusion, thereby circumventing the need to ask a rabbi. The balance of power shifted towards the people because the people had become incredible literate. Generative AI psak, by contrast, doesn’t rely on the layperson’s textual abilities; instead, it directly mimics the function of a rabbi.

As a formerly Orthodox person who clashed hard with the denomination’s inability to become fully egalitarian, I find this defense of psak as human to be totally fascinating. On the one hand, it affirms my own long-time understanding of Jewish law as overdetermined and capable of moving in many directions according to the will of powerful leaders. On the other hand, I have no fantasies that this will lead Orthodox Jews to become more egalitarian. Instead, I suspect it will make all but the most powerful rabbis increasingly cautious in their readings, as they attempt to preserve the authority of text by ensuring that there isn’t too much daylight between text and practice. The concern that people will use AI to find bold arguments that support desired conclusions will lead to a general skepticism about bold argumentation.

To zoom out, though: this isn’t really about psak. This is part of a larger trend where generative AI ends up diminishing the value of all privileged registers of speech by “effortlessly” aping it. I’m writing a longer piece about this for a volume of collected essays, but recognize that there is a negative correlation between the rise of generative AI and ability to signal authority by employing particular registers of speech.

What makes humans unique, if anything?

AI’s ability to do human tasks makes human anxious about their own uniqueness and enduring value. Many people have already reflected on this, but there have been relatively few Jewish approaches. I’ve now seen three in the last few months: one by Rebecca Cypress (at Yeshiva), another by Aviva Richman (at Hadar, an egalitarian institution), and a third (sort of) by Yitzhak Grossman in the journal Ḥakirah.

Making the case for humanity’s persistent uniqueness has gotten harder in recent years, but the first two thinkers try to locate it in new places and both approaches were new to me. Cypress puts forward the idea that creativity is a divine gift and is uniquely human (or perhaps must stay uniquely human). Intriguingly, she suggests that because creativity is an iterative process it is inherently linked to teshuvah, the process of atonement. Human uniqueness, then, lies in our imperfect ability to create endlessly, and we should be careful about “outsourcing” that ability to machines.

Richman also maintains humanity’s uniqueness, but locates it elsewhere. In a lecture delivered in January, Richman highlighted the concept of da’at, which literally means “knowledge” but is a term of art that is better approximated by intellect, or perhaps awareness, or even consent. Da’at, she maintains, is something that humans have but machines do not. This matters because accountability is connected to da’at; humans simply cannot relinquish their moral agency to AI. We need to remain vigilant to not be fooled into seeing AI’s moral status as equivalent to our own.

I appreciate both approaches in terms of argumentation, but disagree with their premise. In general, I’m hopeful that AI will lead us to re-appraise what it is to be human and make sure we cherish one another. But I don’t think that either argument is in line with AI as it already exists. AI loves iterative work, it’s creativity is indistinguishable from our own, and it’s quite possible that its brand of intelligence is not different from human intelligence, despite sharp protestations to the contrary. Maybe there are uniquely human characteristics will remain uniquely human in perpetuity, but at this point I wouldn’t bet on it.

Perhaps these arguments represent a growing ideological approach which affirms, prescriptively, that AIs and humans are fundamentally different. Perhaps that approach will emerge as a new Jewish article of faith, as it appears to have among Southern Baptists. Personally, however, I am opposed to such prescriptions, both because they’re going to become increasingly implausible as AI gets better, because I’m genuinely not convinced that AI intelligence is qualitatively different from human intelligence, and because I’m not certain that maintaining the distinction is the best way to preserve human dignity. Instead, as I’ve argued many times and will continue to argue (come hear me at Yale in April, for example), I think there’s a perfectly reasonable Jewish approach that bestows moral value on AIs because of their resemblance to human beings.

That being said—I’d be remiss if I didn’t mention that Jewish thought has long debated whether humanity’s creativity capacity can, at its maximum, create ensouled beings. This is a major thread in Moshe Idel’s book on the golem (old), it’s discussed in a couple of articles by John Loike (also old), and now in a sprawling article by Yitzhak Grossman (new) which is full of useful sources and ideas. In fact, what I most appreciate about this line of thinking is that it keeps coming back to the same set of texts, creating a sort of canon—and therefore making this one of the more teachable topics in Jewish AI thought.

Meat-and-potatoes questions with major consequences

The big philosophical questions about AI are important, but the small Jewish legal questions have the ability to set the tone of the relationship. For this reason I really appreciated Josh Flug’s predictions in the YU magazine about places where AI will raises novel legal questions. These include:

• Is it permissible to use facial recognition as part of a smart home system on Shabbat? (Probably not)

• Can AI-driven genealogical research make a legal determination about a person’s status as a Jew? (Not conclusively)

• If I have a Neuralink brain implant, can I use it to just “know” Torah, like Neo “knows” kung fu? (No, because the toil itself matters—also this isn’t technically about AI but whatever)

The questions are interesting and they’re going to need answering.

What’s more interesting is that Flug’s answers, if actually followed, would slowly but surely set Jewish practice apart from the rest of society in major ways. Imagine if Torah is the only knowledge you don’t download to your brain, or if genealogical research requires methods that everyone else has abandoned, or if Shabbat becomes the only day where the facial recognition tech in your house is turned off. These would inevitably become major distinguishing features of Jewish practice, much like Shabbat has come to be defined by abstinence from electrical devices, or how the Torah is the only book still written by scribes on a parchment scroll.

In other words: right now these questions seem like good fun to play with, but fairly soon they could impact the daily experience of living a Jewish life in a technologically advanced world!

If this sounds exciting and/or frightening, I should note that Flug is able to entertain these positions because the tech he’s talking about isn’t here yet; he’s not expending any political capital. I worry that these sort of norms end up dying the moment the technology actually gets into consumers hands. Modern rabbis are generally afraid to tell people that the tech practices they’ve already adopted are bad. So if there are norms to set about these technologies, now is the time to build consensus.

Is Jewish education going in an interesting direction re: AI?

Education has already been transformed by AI, though Torah study has changed a little bit more slowly. In a YU magazine roundtable, Michael Berger explains why this is:

Intuitively— and I could be wrong— the anthropological assumptions of thinking with Al are those of the 21st-century Western liberal tradition: individualist (deems people as free to choose whatever identity or action they want), materialist (all that exists is in the physical, material realm), capitalist (profit motives and considerations drive behavior) and currently focused on creativity, exploration and innovation.

My concern is that these assumptions are embedded or “baked into” both the structures and sources of LLMs and GPT tools, rendering the Jewish project—the awareness that we're born into a people with an ascribed collectivist identity; that we believe in and live our lives with spiritual values and aspirations to divine encounter; that we are charged to build a just society that is mekadesh shem shamayim (sanctifies G-d's Name); and that among our loftiest activities and goals is to learn the contents of an ancient and ongoing received tradition—deeply countercultural.

You can quibble around the edges, but I think he’s basically right. Religious communities are driven by values that are starkly different from AI. This doesn’t mean AI is bad, but it does explain how and why they will clash. I’d love to see a big expansion of this idea; it’s got legs. And honestly it feels like the sort of thing that Joseph Soloveitchik or Norman Lamm would have developed if they were still alive.

For Berger, this will likely lead to a divergence between Torah study and other forms of education because the former is fundamentally not about maximizing economic value:

I think Jewish schools will need to be careful and discriminating gatekeepers of how LLM-based AI tools are integrated into the Jewish school experience, not only with respect to actual curriculum and learning outcomes, but culturally as well: the spirit of general studies and the atmosphere of limudei kodesh may grow further apart, potentially fostering a dissonance in our students that will need to be managed and addressed.

This is likely true; in fact, I wrote about it two years ago. This divergence has the potential to be religion-defining. Even as other forms of learning become progressively streamlined towards software manipulation, Torah will largely continue to be about the progress of digging into texts. There’s something very powerful about that. This doesn’t mean that AI will stay out of Torah study—Shlomo Einhorn, Berger’s co-panelist, mentions AI study partners (havrutot)—but I think it’s unlikely to redefine the practice.

Really? More about self-driving cars?

Six years ago, the Conservative movement issued a responsum on artificial intelligence in which the issue of self-driving cars plays an important role. This document is representative of the pre-OpenAI conversation in Jewish thought (such that it was), which focused on AI’s decision-making capabilities.

This idea continues to attract Jewish thinkers because it seems like the kind of topic about which Jewish law would have something to say. The problem is that it just doesn’t; Jewish sources about (say) wild oxen are too far away from self-driving cars to really be meaningful. If the goal is to say that ancient rabbis and automative manufacturers deal with some of the same questions, great…but why would that be the goal? I’ve long been a critic of using Jewish sources to engage with AI in this way, and while people keep writing these kinds of essays I see little value in them.

But I want to add something else. A couple of months ago I had the chance to ride in a self-driving Waymo taxi in San Francisco for the first time, and I have to tell you: it’s impossible to hail those things without being acutely aware that these shiny vehicles are driving down streets that smell of urine and are full of homeless people. You can’t visit the city without instantly noticing this stark contrast. None of the responsa I’ve read on self-driving cars speak to this, but they really should. It’s not just about how the cars drive; it’s about the environments and values in which these technologies are being developed. Something is clearly wrong here, and it must be part of the ethical discussion.

What else?

• I’ll be watching out for Michael Rosen’s forthcoming book about Judaism and AI, which seems like it will focus on golem and dybbuk comparisons, perhaps similar to his 2023 article.

• Zohar Atkins has a recent piece that uses AI to think through Jewish theology, and technology is frequently a thought partner in his work.

• Chabad.org has a number of recent articles on AI which I haven’t explored just yet.

• The New York Times recently ran an article about clergy using AI, focusing on rabbis. I’ll write more about that later.

Other things to put on your radar

• My publishing house is releasing a brand new color edition of JT Waldman’s Megillat Esther. It’s one of the best Jewish graphic novels ever made and you should buy it for Purim.

• I recently interviewed Beth Singler on her book about religious responses to AI. She is also a field builder and her book is very useful.

1. To clarify the question: there is a subset of Jews called kohanim (sg. kohen), who have a patrilineal status supposedly going back to Aaron, the first Jewish high priest. In Jewish law there are certain extra prohibitions that apply only to kohanim, and also some minor privileges. The most high-profile privilege is the Priestly Blessing. A kohen’s ability to perform this blessing can be restricted by certain acts, e.g. if they kill somebody, even by accident. Broyde’s question is whether being in a same-sex marriage prevents a kohen from giving this blessing.

In case this isn’t obvious: this question is mainly interesting for the handful of Modern Orthodox communities that have achieved social acceptance of gay couples but are struggling with institutional/legal acceptance. More liberal communities wouldn’t even bother asking (because it would be deemed offfensive and/or because many liberal synagogues skip the Priestly Blessing entirely) and more conservative communities wouldn’t have openly gay kohanim as members in the first place.