Can ChatGPT Pasken? Cheftza vs. Gavra (full article)
We start Horayot today, so I will put my Jewish Link article up a bit early.
What makes for a valid halachic decision? If I turn to ChatGPT or Claude and ask a halachic question, am I allowed to follow its advice? More than that, does its advice count as a formal pesak halacha? Indeed, within halacha, is there a construct called a pesak halacha, with defining features so that we can say that X is or is not a pesak halacha? If so, what are the halachic ramifications of something being a pesak halacha?
Relatedly, within halacha, is there a construct called a posek, or halachic decisor? If so, what are his defining features? What are the ramifications of someone having the status of posek? Finally, can a pesak halacha only arise from someone who is a posek? These are fascinating questions in general, but I think that they are critical if we are to answer whether a modern AI can be a posek, or whether its answers are pesak halacha.
Rav Schachter vs. Rav Wiederblank
Last year, for the Chanukah volume of the YU Torah To-Go, I wrote a brief article titled “And Among the Nations are not Computed: Should We Use ChatGPT to Pasken?” There, channeling Rav Herschel Schachter, I asserted that, at least in halachic theory, an AI could pasken. I will expand on this idea here. An AI can pasken because there is only a need for correctness in the halachic guidance being offered, if we are to follow it. However, there is no concomitant need for that guidance to be a formal pesak halacha, or for that guidance to be offered by a formal posek. Thus, a yo’etzet halachah is a halachically knowledgeable female Jewish person. Assuming she’s sufficiently knowledgeable to give accurate answers (see next week’s article), we should listen to her halachic guidance.
This is not an issue of female rabbis being valid or invalid, or a pesak only coming from a rabbi or posek. Indeed, a knowledgeable and trustworthy gentile can pasken. And, if someone designed a computer program which took in a question as input, performed halachic calculations or performed a database lookup, and output an accurate answer, then that also works. We don’t care that the computer program isn’t Jewish, male, adult, human, or even intelligent. We are concerned about accuracy.
In my Torah To-Go article, I also raised several concerns about ChatGPT-based approaches, such as hallucinations (that is, making things up), semantic shift (words changing their meaning across time), and ethical alignment (guiding AI to provide answers confirming to modern secular American values). Indeed, when talmidim showed Rav Schachter how ChatGPT incorrectly described his positions on techeilet, he clarified that it could not presently be used for pesak, because we require a valid and trustworthy pesak. Still, an AI can pasken in theory, and perhaps advances in the technology could allow for such pesak.
In that same volume, Rabbi Netanel Wiederblank wrote a longer article, “Ten Reasons Why We Cannot Turn to AI for Psak: Understanding the Nature and Philosophy of the Halachic Process.” He takes a contrary position, claiming that even if AI is more accurate than humans in paskening, we should not turn to AI. I’ll let him speak for himself:
“If psak resembles a solution to a math problem in which there is one correct answer, then it follows that if a computer is more likely to come up with the right answer, it should be followed. However, this approach fails to appreciate what psak is. Psak is not a preexisting thing. It is created by the posek. It is not the answer to the question; it is a human answer to the question given by a qualified posek, someone higiya l-hora’ah. It only becomes halacha once it is issued.”
To frame this dispute in Brisker terms, we might approach the issue as cheftza vs. gavra. For Rabbi Wiederblank, what’s necessary is a gavra, a human posek, from whom a legitimate pesak can arise. For Rav Schachter, a pesak halacha is an observable artifact, a cheftza, namely a correctly reasoned conclusion based on halachic sources. Though perhaps Rav Schachter doesn’t even frame the pesak as a cheftza in its own right – just that in order to act, you need to have obtained correct information.
Both Positions Grounded
Rabbi Wiederblank doesn’t just blankly assert this thesis. He provides many sources and interprets those sources to prove his point. Still, I’ve encountered most of those sources before, including many in Rav Schachter’s shiur. I would interpret them in a somewhat different manner than Rabbi Wiederblank did, and I’d guess that Rav Schachter would as well – he would put them into a different sophisticated and nuanced framework. I don’t have the space here to explain or rebut each source and respond point by point, but I’ll explore just one idea.
Even as Rav Schachter requires an accurate result, he doesn’t discount halachic pluralism. In discussing why each shevet has their own Sanhedrin, he has explained it as something Hashem desires, invoking “variety is the spice of life”. There are many disputes in halacha, and in each dispute, someone wins. The aggregate of those resolved halachot form our halachic system. Still, both these and these are the words of the Living God. It could be that in shevet Reuven they decided one way in the dispute between Rabbi Meir and Rabbi Shimon about noten ta’am lifgam, and in shevet Zevulun, they decided another way. And so for many other disputes, and those disputes might well interact in interesting ways. These each aggregate to different equally valid halachic systems, for their own shevet. The same goes for Sefardic and Ashkenazic halacha, both equally valid. Relatedly, this is why a religious American Jew shouldn’t automatically follow a pesak from a Gadol in Eretz Yisrael, like Rav Chaim Kanievsky zatza”l. It might be predicated on a different set of accepted halachic axioms.
Even so, there was a formal process of centralizing halacha in the Sanhedrin and an imperative for people to follow that now centralized law. And even so, within any valid halachic system, it is possible for a posek to err, and for a halachic decision to be erroneous. Much of our present masechta, Horayot, is about what to do when the Sanhedrin erred.
Yes, the people in Lud, Rabbi Eliezer ben Hyrcanus’ hometown, followed his unique position and pesak, and performed preparatory acts for circumcision on Shabbat that could have been performed beforehand, and were rewarded for it. And, in the case of tanur shel achnai, the oven of coils, the halacha accorded with the Sages over Rabbi Eliezer despite Rabbi Eliezer’s position perhaps making more sense and had Hashem and thus truth on its side. Does that mean that a formal pesak has legitimacy, even if it is wrong? To oversimplify a complex topic, can we say that “we are doing the right thing when we follow directions”? I don’t think so – certainly not if we know it’s wrong. That would accord with what Horayot 2b labels טָעֵי בְּמִצְוָה לִשְׁמוֹעַ דִּבְרֵי חֲכָמִים, “erring with regard to the mitzva of heeding the statements of the Sages”.
Rabbi Eliezer (ben Hyrcanus) was a third-generation Tanna. We often don’t heed him because (Shabbat 130b) רַבִּי אֱלִיעֶזֶר שַׁמּוּתִי הוּא, he was either placed under the ban (shamta) or because he was a member of Beit Shammai. We would more likely follow his contemporary Rabbi Yehoshua (ben Chananya), aligned with Beit Hillel. Each represents a different legitimate halachic system, especially before the centralization, so the Luddites could follow Rabbi Eliezer’s position; and we can avoid the Torah developing into multiple Torot and select a single system aligned with the Sages for the coiled oven, even if Rabbi Eliezer’s position had its own logic and legitimacy.
That is not the same as saying that, within a Beit Hillel system, within a Sefardic or Ashkenazic system, any ruling issued by a posek is legitimate, even if there were errors in understanding reality or in interpreting the sources. There might be a single correct halachic conclusion within that system, or with some wiggle room, admittedly a few possible correct halachic conclusions. If a rabbi or AI gets it right, congratulations! If he / it erred, then that’s bad.
Gemir and Sevir
At the start of Horayot, the Mishnah declared that if a present student, who was fit to decide, knew the Sanhedrin erred but stayed silent, and then X (usually understood as that student, but who might also be the yachid of the Mishnah’s first clause) acted on their ruling, then X cannot rely on a korban that the Sanhedrin bring for this error (because he should have known better, or because it strips the ruling of hora’ah status). What makes for a valid student regarding hora’ah? In Horayot 2b, Rava and Abaye discuss the qualities of gemir, knowing the sources, and sevir, reasoning ability. We see these qualities mentioned in Sanhedrin 5a as well, about valid judges in monetary cases, for making valid court rulings.
Perhaps this means that a posek is a gavra with these qualities; only then is the pesak a cheftza. To defend Rav Shachter’s position and allow AI to pasken: a ruling is valid if it’s correct; knowledge and reasoning ability lead to correct pesak. The question is whether ChatGPT has knowledge and can reason, in the way we require, even with RAG (retrieval augmented generation) and its reasoning model. In the next two articles, I’ll explore these two qualities. Stay tuned!