Can AI ever replace a posek?

We are honoured this week to be hosting a fascinating piece by R. Gil Student (adapted from his recent book, Articles of Faith: Traditional Jewish Belief in the Internet Era)

Dicta. org.il recently released an AI “rabbi” that answers halachic questions in the style of a classical responsum. The AI is trained on rabbinic literature, including primary texts, commentaries, codes and responsa. It sees how poskim (halachic decisors) reason, interpret texts, compare cases, consider the unique circumstances of the question, and arrive at answers. While to my mind it does not yet perform at the level of a competent rabbi, it offers us a realistic idea of the possibility of a coherent technology that can answer halachic questions independently. Once it is perfected, may any individual Jew looking to fulfill God’s word ask an AI a halachic question and follow its answer? And once AI is available to do this, would we be obligated to ask an AI rather than a human, with all his frailties?

I. AI and Divine Assistance

One approach to take is to argue that a posek, a halachic authority, requires divine assistance, siyata di-shmaya. God will help a rabbi reach the appropriate conclusion but will not assist an AI. Therefore, an AI lacks the necessary siyata di-shmaya that a rabbi has. But is this true? If the siyata di-shmaya is for the rabbi’s sake, then an AI has no merit to deserve it. But if the siyata di-shmaya is for the questioner, who wants to fulfill the divine will, why would God not help the questioner via the AI? If the rabbi is a divine vessel, why can’t the AI be a divine vessel also?

Another approach is to compare AI to a gentile. Of course, a gentile is different. He is a child of God, created in the divine image. However, the comparison still has value as a kal va-chomer, an a fortiori logical argument. If a gentile, who is conscious and able to think and reason, is unable to issue a halachic ruling, then surely an AI is unable to do so. Yet, where do we see that a gentile may not serve as a posek? I am not aware of any text or precedent that says so, although I welcome any suggested texts. Absent a text, we need a strong explanation of if and why a learned gentile, who for whatever reason and in whichever way became a master of halachic texts, may not serve as a posek.

A few years ago, when people were debating whether women can be rabbis, one line of argument was that there is no longer any halachic requirements for a rabbi. As long as a woman knows halachah, she can serve as a rabbi. As a response to that argument, I asked whether a gentile who studies Shulchan Aruch can also serve as a rabbi. If he knows the material and can teach it to people, why can’t a gentile serve as a rabbi? Of course, a Jewish woman is part of the Jewish people and the Sinai covenant. She observes the commandments and the Jewish holidays, and is part of the age-long Jewish story. But does any of that matter in terms of issuing a halachic ruling?

II. Two Types of Pesak?

In order to begin answering these questions, we first need to distinguish between two types of paskening, issuing halachic rulings. The Gemara (Eruvin 62b) says that a student may not issue a halachic ruling in the presence of his mentor to the point that it is a capital offense (ibid., 63a). Rav Meir (Maharam) of Rothenburg (13th cen., Germany) is quoted as distinguishing between an original ruling and an existing ruling found in books (Hagahos Maimoniyos, Hilchos Talmud Torah 5:3). A student may offer in his teacher’s presence a halachic ruling that he found in a published book because he is not really ruling. He is just serving as a technical reference, offering the knowledge that he has acquired through his study. However, a new halachic ruling requires evaluation and consideration. A halachic reference is the regurgitation of information. A halachic ruling is the creation of new information. Significantly, this distinction is recorded in Shulchan Aruch (Yoreh De’ah 242:9) and explained by Shach (ad loc., 16). While many dispute Shach’s explanation, Rav Shmuel Wosner (21st cen., Israel) argues that this idea applies to post-Talmudic codes and responsa that are intended for practical teachings (Shevet Ha-Levi 2:113).

With this distinction between a halachic reference and a halachic ruling, we can understand another law. The Gemara (Avodah Zarah 7a) says that when one rabbi rules on a specific case strictly, another rabbi is not allowed to permit it. One approach to this seemingly technical rule is that pesak, halachic rulings, do not just teach the law. A rabbi’s ruling creates a religious status in the object under question. Is this chicken kosher or non-kosher? When you ask your rabbi, he defines the chicken’s status for you. Even if another rabbi disagrees and says that the chicken is kosher, it doesn’t matter. If he can show that the rabbi was mistaken on an explicit law (to’eh bi-dvar mishnah) then he nullifies the original ruling. Absent that, the first rabbi has changed the chicken’s religious status for you. He has not just taught you the law but determined what the law is for you. A pesak is a religious act, not just a technical reference.

III. Pesak as a Mitzvah

The Torah says regarding the priests: “that you are to teach the Children of Israel all the statutes” (Lev. 10:11). Rav Yitzchak of Corbeil (13th cen., France) counts this as a mitzvah on anyone capable of issuing a halachic ruling (if no one else more capable is available; Semak, no. 101). Issuing a pesak is a mitzvah, a religious obligation. While others include this as part of the mitzvah to study and teach Torah, Rav Yitzchak sees it as an independent mitzvah. Either way, issuing a halachic ruling is an act of religious devotion. It is a fulfillment of a divine command that allows the respondent to enter the religious life of the questioner and create for him a new halachic reality.

Is it any wonder, then, that a gentile may not serve as a posek, a halachic decisor? It is one thing to serve as a halachic reference, to direct people to existing rulings. Anyone who has studied halachah can tell people what, for example, Mishnah Berurah says about a specific case. A computer can do this, as well. A person is better than a regular computer at understanding the nuances of a question and offering the appropriate reference source. An AI might be even better at this than a person, with perfect recall of a massive library. However, this is just about referencing past rulings.

Issuing a new halachic ruling is not just about providing a reference — it is a religious activity. I suggest that only those within the religious community, only those who are part of the covenant and fulfill commandments, can create a halachic reality by issuing a ruling. Jewish men, women, and children fulfill commandments and take part in the halachic process as consumers. Therefore, if qualified and in the right circumstances, they may also take part in the other end of the halachic process as producers. A Jewish man, woman or child can serve as a posek. Someone outside the halachic covenant cannot create a halachic reality and therefore cannot serve as a posek. This would prevent someone from asking a gentile and, kal va-chomer, an AI for halachic rulings.

There is still room for AI in the halachic process. An AI can serve the broader public as a halachic reference for questions that have already been answered, directing people to the relevant texts and even providing them source sheets. An AI can also serve as a rabbi’s assistant, gathering relevant sources and proposing different approaches to answering a question. Ultimately, the posek will evaluate all the relevant texts and precedents, think through all the mitigating circumstances and personal considerations, weigh the halachic and meta-halachic issues involved, and deliver a halachic ruling.

For comments and discussion of this post on Facebook, click here.

Rambam's Messiah or a dystopian nightmare?

Yesterday morning I sat reading a book review in The Economist of Deep Utopia – a novel by Nick Bostrom which contemplates a future era in which all human work and activity is rendered redundant by progress in AI and robotics. The reviewer was deeply troubled by the nightmarish scenario and implications of the “terrible boredom” which would ensue once humans no longer have to dedicate the majority of their hours to earning a living and performing necessary chores. 

What struck me is the stark contrast of this reviewer’s outlook to that of Rambam in the closing statements of his Mishneh Torah:

“The Sages and the prophets did not yearn for the Messianic era in order to have dominion over the entire world, to rule over the gentiles, to be exalted by the nations, or to eat, drink, and celebrate. Rather, they desired to be free to involve themselves in Torah and wisdom without any pressures or disturbances, so that they would merit the world to come, as explained in Hilchot Teshuvah.

In that era, there will be neither famine or war, envy or competition for good will flow in abundance and all the delights will be freely available as dust. The occupation of the entire world will be solely to know God.”

It got me wondering how enthusiastic talk of “wanting to bring Mashiach now” should be translated practically. While a lot of focus is understandably placed on maximizing mitzvah performance and debating how and when a Mikdash might be built, how much of our nation would regard the Messianic era of Rambam and our sages as a utopian blessing rather than dystopian dullness?

One separate matter that this book review might resolve is how to reconcile and interpret some other difficult sources regarding the Messianic era. In a famous passage (Berachot 35b, see further discussion in Judaism Reclaimed), Rabbi Yishmael interprets the blessing contained in the Shema of “And you shall gather in your grain” to be envisioning an ideal future of great abundance. Rabbi Shimon objects that this would hardly be a blessing: if the super-abundance means that we are required to spend all day gathering in grain “what will become of Torah study?!”. Rather, he contends, the true future blessing can be found in Yeshaya’s messianic vision in which we will be able to sit and study while “strangers” will happily and voluntarily till our land.

Who might such strangers be? If it refers to non-Jews, how is this then to be reconciled with a further messianic prophecy of Yeshaya which describes how the nations will rush to learn the word of God from the Jews?

Well, if Nick Bostrom’s vision comes to pass, the “zarim” (strangers) who are tasked with harvesting the abundance and feeding humanity will be mechanized robotic ones, programmed with just the right amount of intelligence for the job. (I’ve recently wondered, during my weekly farm-volunteering in Southern Israel, if Rabbi Shimon would have been placated had he seen people in the modern era picking fruit while listening to hours of Shiurim and Torah podcasts!).

Finally, there are those who raise the question as to whether AI developments could even influence or render redundant the Maimonidean messianic vision of sages dedicating endless hours to Torah study and absorbing divine wisdom. With expected advances of technology, they ask, would the sages not just be able to plug a computer chip into their brains and thus instantly possess all the requisite Torah knowledge?

To understand the fallacy of such an assumption, I believe, one first has to be aware of the nature and function of Torah study (a matter explored at length in a chapter of Talmud Reclaimed). Torah knowledge is not simply a set of facts, laws and statistics which can be memorized or uploaded from a computer drive. Rather, as sources as diverse as Rambam, the Ba’al HaTanya and Rav Soloveitchik show, it is a relationship with God and perspective on the world which arises from immersing oneself in sophisticated Talmudic analysis. From absorbing such wisdom and training one’s mind sufficiently, a person can come to look at the entire world differently – from a more divine perspective.

In fact for Rambam, such a process is critical as a prior stage before one enters the esoteric areas of the Torah known as “Pardes” (Hilchot Yesodei HaTorah 4:13).

It is hard to imagine that delegating one’s Torah study to a robot can even begin to achieve these benefits. There is a derogatory concept in rabbinic literature of “Chamor Nosei Seforim” (donkey loaded with books) which refers to a putative Torah scholar who has merely memorized numerous sources without having properly understood or internalized their meaning. What we can hope for is that the advent of AI makes information so ubiquitous that all Torah rote-learning becomes unnecessary. If this occurs, it could allow for a shift of focus in the Messianic era to analysis of more profound principles and a utopian era in which we are free to develop our minds to that our “sons and daughters will prophesy” – a vision which represents the pinnacle of Rambam’s Mishneh Torah and vision of Judaism.

First posted on Facebook 2 June 2024.

AI and the future of robo-rabbinics

I must admit that I’ve never been a great one for sci-fi and the anxious talk of civilisation-ending supercomputers overtaking and enslaving humanity. Aside from my pocket chess computer defeating me with irritating regularity, the computers in my life have generally maintained an unthreatening distance and respected my personal space. 

If my Facebook feed is to be believed, however, my idyllic ignorance is soon to be shattered by the AI robots and their resounding march of progress. Some of them appear to be already signalling their hostile intentions through Chap GPT, leading learned friends and scholars (and some random people who the Facebook automatons have placed on my feed) to question what sort of future if any lies in store for rabbinics, halachic rulings and Torah study

So how much does Judaism value the intellectual input of robots and algorithms into the study hall of halachic deliberation? Will Robo-Rabbi kill the Yeshiva World star?

Very often, responses to this question will focus on the apparent “accuracy” of rabbinic responses, and whether these can be bettered by a computer chip. This though may be a red herring. While it is true that some halachic questions tend towards more technical, algorithm answers, even apparently-clearcut and rigid rules like kashrut can sometimes vary depending on subjective assessments such as shaat hadechak (extreme necessity), hefsed or hefsed meruba (loss or significant monetary loss).

Furthermore, it is important to acknowledge the irreplaceable human dimension and empathy within psak. A rabbi who has a profound understanding of his community, will sometimes give non-identical answers to the same question when asked by different congregants – because halachic rulings often require insight into the nature of the questioner. For this reason, objections have been raised in the past to the posthumous publishing of halachic rulings issued by great rabbis to specific individuals.

Even more crucially, though, we need to recognise the role within Jewish study of the process of establishing halacha. As I examine in my upcoming book, determining halacha and analysis of the related Talmudic passages was traditionally a single process. Talmudic analysis was largely centred around more practical questions, and halachic determination was a matter of directly analysing the relevant Talmudic sources rather than summaries made by later sages.

The gradual shift towards a codification of halacha, which went hand-in-hand with a Talmudic focus on hyper-analytical pilpul in European yeshivot, was heavily criticised by some of the leading rabbis of the time. One sage who was particularly vocal in his opposition was the Maharal, who considered that:

The sage has only that which his intellect provides him and allows him to learn from within the Talmud. And when his understanding and wisdom lead him to err, he is nonetheless beloved by God so long as he is issuing instructions based on the dictates of his reason…and this person is preferable to one who determines halacha from within one book, knowing not the reason for the matter, walking like a blind person along the way. (Netiv HaTorah, 16)

Crucial, according to the Maharal, is the process of engagement with the halachic analysis - of poring over Talmudic passages to compare, contrast and identify their conclusions. This process itself is more important than actually reaching any “objectively correct” conclusion. The implications for those who propose using AI to produce halachic rulings are clear.

Why is this process of halachic engagement deemed so critical?

Writing in Halakhic Man, Rabbi Joseph B. Soloveitchik develops the theory that Torah study is not focused simply on the question of how to establish practical Jewish law. Rather the student is involved with understanding the Torah and the world on the level of the abstract ideal. Drawing upon sources as diverse as Rambam and the Ba’al HaTanya, Rav Soloveitchik proposes that through immersing oneself in the world of Talmudic analysis, a person can absorb its wisdom and internalise its analyses to the extent that they can come to look at the entire world differently – from a more divine perspective. (My upcoming book will also analyse and develop this theory further).

It is hard to imagine that delegating one’s Torah study to a robot can begin to achieve any of these benefits.

In a related question, I was recently asked for my thoughts on the much-heralded advent of a computer chip which could plug into our brains and thus instantly provide us with all Torah knowledge when studying. This second question could perhaps be looked at differently – just having “neurological access” to an entire treasure trove of Torah could be seen as bolstering and enhancing rather than replacing the project of Torah study. In a similar vein, the wonderful array of online Torah resources – and Mesivta compilations of Talmud commentaries – have revolutionised the ease with which one can prepare a shiur.

While such benefits are certainly substantial, they are not without significant drawbacks. I have often found, for example, that the greater the quantity of information available, the less well it has been understood and internalised. Talmudic students confronted with compilations containing a mass of information and commentaries find it increasingly challenging to read and analyse the simple meaning of the biblical or talmudic text in its own right, without reaching for additional sources. Lacking a sensitivity to the nuances and challenges presented by the text, these students then fail to gain a real appreciation of what the commentaries are saying. My fear is that a computer chip which imported a mass of information into the brains of Yeshiva students would exacerbate rather than address this problem.

On a personal note, I find that I can quickly identify shiurim which based on are little more than compiling a mass of collated sources on a subject with scant analysis – which are typically very dull. What I really search for in a shiur, book or commentary is an original angle or insight. A new theory or piece of analysis which sheds light on the passage being studied and suggest broader principles which can be applied elsewhere. In order to produce such a shiur, a person must first have fully internalised and understood the meaning of the relevant source.

There is a derogatory concept in rabbinic literature of “Chamor Nosei Seforim” (donkey loaded with books) which refers to a putative Torah scholar who has merely memorised numerous sources without having properly understood or internalised their meaning. What we can hope for is that the advent of AI makes information so ubiquitous that Torah rote-learning becomes unnecessary. If this occurs, it could allow for a shift of focus to analysis of more profound principles and thereby provide a valuable service to the world of Torah.

Posted to Facebook 26 February 2023, here.