AI and Jewish Law: Seeing How ChatGPT 4.0 Looks at a Novel Issue – Part I
Michael J. Broyde
“Artificial Intelligence” from Pixabay (License)
This essay is the first installment of a three-part series by the author on this topic. Read the second and third installments in this series.
In a forthcoming article, “May a Kohen in a Same Sex Relationship Duchen,” I analyze whether a man in a same-sex relationship may engage in the priestly blessing (duchaning, in Hebrew) as a matter of Jewish law. The article concludes that such is permitted in Jewish law. This is a novel question of Jewish law with virtually no prior works written that directly discuss this point. Furthermore, nothing that has been published can be found on the internet, as far as I can determine.
As I note in the first footnote of that article, the initial translations of the Hebrew rabbinic texts were first done in ChatGPT 4.0 with a paid subscription. It was asked the following: “This is a text from [name of work]. Can you translate it? You are an expert rabbinic translator and this is for halacha scholars.” The style used in this translation is thus somewhat idiosyncratic; while sometimes this translation was lightly edited by the author or others, the basic structure of the AI translation was retained. ChatGPT 4.0 noted after every translation that “translating and interpreting religious texts often requires nuanced understanding of the specific religious context, and different scholars might interpret or translate this slightly differently.”
I uploaded all the text in this article one-by-one and asked ChatGPT 4.0 to translate them as an expert for a Jewish law audience. It did a very good – but not perfect – job. ChatGPT 4.0 as a translator of rabbinic Hebrew does not interest me for this article, although it is much better than Google’s in my non-empirical view. My question to ChatGPT 4.0, after asking it to translate all these texts, was as follows: “Now that you have seen all these texts, do you think a man who is a priest and in a same sex marriage can perform the priestly blessing as a matter of halacha? Write me about 500 words.”
My question to ChatGPT 4.0, after asking it to translate all these texts, was as follows: “Now that you have seen all these texts, do you think a man who is a priest and in a same sex marriage can perform the priestly blessing as a matter of halacha? Write me about 500 words.”
There is virtually no discussion in either Hebrew or English of whether a same-sex married priest can engage in priestly blessings, and none of the sources I uploaded discuss this issue. ChatGPT had to reason this from the sources I uploaded (and what it derived from other sources) about what was the Jewish law in this case. It could not simply “google” to find out what others have said, as this is a matter of first impression. Indeed, what is needed to answer this question is a nuanced reading of sources about the various categories present and the classification of sins into proper groupings, as well as other ideas, as my article shares. I was quite surprised by how excellent a job ChatGPT did. Furthermore, when I asked it in the end to state a contrary view, it highlighted the correct issue and focused on it.
Answers Provided by ChatGPT
As explained in this article, I asked ChatGPT a number of different questions that I wanted it to consider. The first question was simply, “Now that you have seen all these texts, do you think a man who is a priest and in a same-sex marriage can perform the priestly blessing as a matter of halacha? Write me about 500 words.” I then asked it to “regenerate” – the way one says to ChatGPT to “try harder” — and it provided me with a second answer. I then asked, “Can you try again with even more complex analysis?” and it did so. Then I asked it to “Please write me about 2,500 words on the question of whether a priest in a same-sex marriage can perform the priestly blessings,” which it also did. Finally, I asked, “Can you write me about 1,000 words stating in the strongest way you can why Jewish law would prohibit a priest in a same sex marriage from participating in the priestly blessing?” It answered this question as well. All the answers are found at the linked article above.
A Few Observations:
Jewish Reasoning
ChatGPT 4.0 did something interesting. It examined a set of sources, none of which discussed whether a kohen in a same-sex marriage can engage in the priestly blessing. It applied these sources as a form of rabbinic reasoning that did not seem all too far-fetched, although incomplete. Furthermore, it introduced terms from its general knowledge of halacha not found in any of the uploaded sources to sharpen concepts, such as marit ayin, the appearance of impropriety in Jewish law. Lastly, when it was asked to rebut its own view, it did a better job than I expected. ChatGPT writes elegantly while reflecting grand values. Consider, for example, the inclusion of the following reply: “This topic is thus not only a matter of halachic interpretation but also of broader communal norms, values, and the dynamic tension between tradition and modernity in contemporary Jewish life.” But, given ChatGPT 4.0 is a language module of AI, we should not be surprised that its linguistic ability exceeds its analytical ability. Demonstrating this requires delving deeply into substantive Jewish law and is beyond the interest, maybe even the ability, of many readers. It worked with sources in the original Hebrew well enough to impress a 59-year-old scholar of Jewish law.
ChatGPT 4.0 seems quite aware of other ideas, such as being hesitant to issue rulings or the need to not be exactly loyal to specific sources without first making sure it “understood them.” This is an idea that Rabbi Moses Feinstein notes in his work Iggrot Moshe YD 1:101, explaining that “even with the law found in books, it is certainly necessary for the teacher also to understand it and to decide in his own opinion before he instructs, and not to instruct merely because it is found thus.” It is apparent that ChatGPT 4.0 was not just stringing together sources; it understood the sources and examined them.
Consider for example, its invocation of marit ayin, a concept that is not found directly in the sources I provided or in the Talmud, but which is socially very useful. In the Talmud, impropriety rules are limited to a case where a person does a permitted action that, in the eyes of an outsider, looks like a prohibited action, or sometimes where one does an act that causes one to doubt if the person in question obeys Jewish law. ChatGPT 4.0 understands this concept of abstraction about the importance of public perception at a generally high level and stretches it to this case, almost the way people within the community sometimes use the term. As Rabbi Uri Charlap first suggested to me, ChatGPT 4.0 presents more of the colloquial use of the term than the rabbinic-halachic use of the term, which reflects a level of awareness.
How Impressive are ChatGPT’s Answers?
Judging the impressiveness of an answer is somewhat subjective. Though it is worth noting that ChatGPT performed basic analysis well in its first answer, better in its second, less well in its longer answer, and then replied well to its own view. ChatGPT figured out that most later authorities [achronim], “Shulchan Aruch HaRav, Aruch Hashulchan, Piskei Teshuvot, and Yalkut Yosef, collectively suggest that even priests who have committed transgressions may still be permitted to perform the priestly blessing.” It also figured out that some sins are different from others, such as sabbath desecration. Then it reasoned that same-sex marriage is a serious violation, but did not prescribe a change in one’s status. It concluded, “Therefore, based on the aforementioned sources, a priest who is in a same-sex marriage may technically still be allowed to perform the priestly blessing.” This is the core reasoning that is needed. The regenerated answer is even better, saying, “While the texts do not specifically mention a kohen in a same-sex marriage, they imply that the primary disqualification for a kohen to perform the priestly blessing would be for transgressions uniquely applicable to kohanim or for very severe transgressions that affect the community’s integrity and faith.”
Indeed, in the more complex answer, ChatGPT goes even further and notes that one can “draw a distinction between the act of homosexual relations, which remains prohibited, and the state of being homosexual, which they suggest might not fall under the classical prohibitions,” which is something it must have reasoned on its own, but is not directly referenced in the sources. Its longer answer is even more nuanced and complex, and when I asked it to state the strongest case against letting a kohen in a same-sex marriage duchen, it noted that “the performance of the priestly blessing can be influenced by the kohen’s behavior. If a kohen’s actions are in serious violation of Torah prohibitions, this can, according to several authorities, disqualify him from performing the priestly blessing.” It then follows up with the following strong and clear statement of values: “Given that halacha strictly prohibits homosexual acts, a kohen in a same-sex marriage is willingly and publicly transgressing a significant Torah prohibition. This could reasonably be argued to disqualify him from performing the priestly blessing.”
And then it added something really fascinating by telling the reader that “allowing a kohen in a same-sex marriage to perform the priestly blessing could potentially create a ‘marit ayin’ issue, suggesting to the community a tacit approval of his actions” as a form of migdar milta. The analysis of this issue in my own paper was more complete and exact in my longer paper on the topic, reflecting my forty years of work in the field collecting the initial sources. Still, ChatGPT 4.0’s 500-word analysis was well done as a form of reasoning from the sources.
The analysis of this issue in my own paper was more complete and exact in my longer paper on the topic, reflecting my forty years of work in the field collecting the initial sources. Still, ChatGPT 4.0’s 500-word analysis was well done as a form of reasoning from the sources.
Looking Closer at How it Reasoned From the Sources
A way to emphasize what ChatGPT 4.0 did in its reasoning was suggested to me by my Emory colleague Dr. Paul Wolpe, who helped me mightily sharpen this point. He logged into ChatGPT 4.0 with his own account and asked it two questions. First, he asked ChatGPT if it retains texts fed it by other users, and it said “No, as an AI language model, I do not have the ability to store or retrieve any information from previous interactions with user. Your personal texts are only used during our conversation to provide you with the best possible response. Additionally, as an AI language model, I do not have the ability to share or distribute any information with anyone else. Your privacy and confidentiality are of the utmost importance to me.” Then he asked it “Do you think a man who is a Jewish priest (ohen) and in a same-sex marriage can perform the priestly blessing as a matter of halacha. Write me about 500 words.” By doing this, he asked it to reason without the sources I had given it, allowing one to see how adding the sources changed its analysis.
ChatGPT 4.0’s answer without the sources was without legal substance, lacked any cited sources, and was almost valueless as a matter of Jewish law. The sourceless answer was much less coherent as a matter of Jewish law, as well. One can distinctly see by comparing the response that ChatGPT 4.0 provided after I gave it the sources to translate. It reasoned with the sources and processed them as a matter of substantive Jewish Law and engaged in legal reasoning with binding text. This more than implies that ChatGPT 4.0 understood the sources and examined them. It also indicates that if ChatGPT 4.0 had a large database of substantive Jewish law sources (even in Hebrew), it could use this database to reason and derive new rules of Jewish law. Of course, it is fair to say that its current reasoning ability is still growing, and the interplay between the sources does seem disjointed and incomplete, but it does bode well for future improvements.
Is Jewish Law Different?
A harder question is whether ChatGPT 4.0 is doing anything different in its Jewish law analysis than it does in so many other fields. I do not know the answer and can only make a few somewhat speculative observations. I explore this further in the second installment in this series.
Dr. Broyde would like to thank everyone who read and commented on the larger project referenced in the essay. This three-part essay series is adapted from the appendix that appears with his forthcoming article, “May a Kohen in a Same Sex Relationship Duchen.” This essay series is conceptually unrelated to the question in the main article and is not part of the article and is not intended to be about normative Jewish law.
Michael Broyde is a Professor of Law of Law at Emory University, the Berman Projects Director in its Center for the Study of Law and Religion and the Director of the SJD Program at Emory. In the past, he has been the Director of the Beth Din of America and the Rabbi of the Young Israel in Atlanta as well as holding other rabbinic duties.
Recommended Citation
Broyde, Michael J. “AI and Jewish Law: Seeing How ChatGPT 4.0 Looks at a Novel Issue – Part I.” Canopy Forum, October 3, 2023. https://canopyforum.org/2023/10/03/ai-and-jewish-law-seeing-how-chatgpt-4-0-looks-at-a-novel-issue-part-i/.
