Imagine if Cinderella were a panic-stricken college student, laboring at the last minute to compose a term paper on the portrayal of women in 19th-century literature. She turns to ChatGPT, her digital fairy godmother, and, with three taps on the keyboard and two clicks of a mouse, she summons a well-crafted essay with flawless grammar, spellbinding analysis, and sagacious references to the works of Jane Austen and Alfred, Lord Tennyson. She types her name on it, gingerly adds a few final flourishes, then merrily waltzes to the Performing Arts Department for an elective class on ballroom dance. Meanwhile, her professor, confronted by a sandstorm of grading and an acute lack of capable teaching assistants, ponders whether this “Chat Genie” can ease his burden. He tries it out and, for the remainder of winter break, doesn’t know whether to feel delighted or distressed.
Across high school and college campuses, this story, or one very much like it, has been repeated countless times since early December, when OpenAI introduced ChatGPT, an artificial intelligence model capable of simulating human communication. But the future meaning of “original work” is only a sliver of the story. For a “sci-fi rabbi” (such as myself) who daydreams about keeping Shabbat on Mars, rumors of an AI wizard descending from “the cloud” to dispense its wisdom to the children of Eve is way too cool to ignore. On a personal note, I suffer from chronic “Google-entiritus” — a condition characterized by eyestrain, indigestion and futile attempts to use the internet for research purposes—so a search tool that can deliver me directly to the promised land of learning, skipping 40 years in the wilderness of online shopping and sponsored content, sounds like a fabulous balm for a stiff neck.
So what is ChatGPT? It is basically a massive digitized library containing a significant portion of humanity’s cultural and scientific knowledge. Using this vast ocean of information, ChatGPT synthesizes concise and clear answers to almost any question imaginable. While this technology was trained to emulate answers written by humans, its advanced algorithms can generate answers to queries for which there are no precise equal in its vast repository. Its general knowledge is astounding. In a few seconds, ChatGPT can explain a passage from Shakespeare, translate “Don Quixote,” compare different car models, describe Einstein’s theory of relativity in layperson’s terms, or write a paper about the Renaissance.
For the high school student struggling with a five-paragraph essay, ChatGPT does seem rather magical. But for educators, ChatGPT raises a host of concerns: What should be done about students who use ChatGPT to edit their work, or to generate essays from scratch? Can the technology be used to make students better writers? Will student access turn into dependence, inevitably diminishing creativity and critical thinking? The New York Public School system simply banned it. But catching violators will pose a challenge. When I tested a beta version of GptZero.me, a tool designed to detect “AI Plagiarism,” its failure rate was astronomical. It routinely misidentified legal and policy writing as AI-generated, while concluding that AI-written poetry was the real thing.
As a Jewish educator, the question of whether to use ChatGPT or not appears to be moot. A ban on ChatGPT will likely engender the same result as God’s command not to eat from the Tree of Knowledge. We are curious creatures, who will sacrifice paradise to satisfy an insatiable hunger for understanding. Once we learn of a better way to do something, we’re quick to adapt. Journalists will use ChatGPT to edit their articles. Lawyers will use it to draft templates. ChatGPT can suggest restaurants, help plan trips, troubleshoot code and review products. It is robustly versatile. The question for the Jewish community is how ChatGPT can be utilized for Torah study and to improve Jewish life.
Functionally, ChatGPT can operate as a well-rounded encyclopedia of Judaism, swiftly answering “Who, What, Where, and When” inquiries. It dispenses information related to the Talmud and Mishna, and provides summary bios of Jewish figures, whether medieval commentators such as Rashi and Maimonides, or modern Israeli icons such as Naomi Shemer or Golda Meir. Its ritual knowledge is such that it can, citing sources, explain the significance of Hanukkah, or the lighting of two candles in honor of Shabbat. Because the AI is programmed to avoid bias, it will assess, in a neutral manner, topics with multiple perspectives, whether it’s analyzing the role of religion in Israel’s political system or listing the pros and cons of synagogue “kiddush clubs.”
One of ChatGPT’s most astonishing features is its ability to generate creative, non-factual content. This capability is a great source of entertainment.
Here is an example:
Me: Can you write a sonnet about Shimon Peres and Bibi Netanyahu drunk in a pub debating whether the Zionist state will be better off with more cats and less rats or vice versa?
ChatGPT:
In a pub, where drinks flow like Galilee’s streams,
Two leaders, Peres and Netanyahu convene.
With mugs in hand, they discuss their fondest dreams,
Of a land where cats chase rats with lightning speed.
One argues that more cats would bring great wealth,
To rid the streets of pests and bring great health.
The other claims that rats provide the stealth,
That builds the strength and skills the land needs well.
But as they drink and debate, one thing’s clear,
Their love for their country knows no bounds.
Despite their different visions, they hold dear,
The future of the Zionist state they’ve found.
For in the end, their love of land prevails,
Greater than any argument or debate details.
It will do this again and again, in diverse styles, languages or early English dialects: “Lo, two lords, Shimon and Bibi, didst gather one night….” For Yiddish or Ladino enthusiasts, ChatGPT’s language abilities may prove an invaluable resource in the years ahead. It had a go at Yiddish translation:” In a pub, “vi shpritzen bier likh vi Galilee’s strim….” And, when prodded to translate the poem into a “Ladino with Hebrew, Arabic and Turkish influence,” it finally wrote something that seemed distinct from Spanish: “En un kafeneya, kuando la cerbeza korre komo los rios de Galilea.” While maestros of the mama loshen or Judezmo will likely be disappointed for now, there does not seem to be any reason an AI language model that can master English cannot be equally conversant in other languages.
One of the frustrating features of the autocorrect function on my smartphone is that it constantly changes transliterated Hebrew phrases and expressions. “Shalom aleichem” becomes “salmon in a blanket.” “Rosh Hashanah” is altered to “Tosh has Hannah.” “Kosher” is sometimes “cashew,” “cashier” or “kitchen.” Because of ChatGPT’s multicultural training, it recognizes phrases like “mazal tov,” “chag sameach,” “kishke” or “cholent,” and it even responds to “Gut Shabbes” with “Shabbat Shalom.”
Perhaps the most significant problem for ChatGPT users is that the same algorithms that give it the ability to extemporize and creatively mimic tropes frequently causes the technology to state inaccuracies and outright falsehoods. When I first tried chatting with it in Hebrew, it insisted that the three Patriarchs were Abraham, Isaac and Joseph, and that King David was Joseph’s descendant. When I pointed out that Jacob was the third patriarch and that King David was from the line of Judah, it apologized profusely (which it does a lot in Hebrew when it makes mistakes). Culturally, Israelis apologize far less frequently than, say, the Japanese, so it was rather obvious that I was not chatting with a native speaker. But OpenAI is constantly updating the technology. Several weeks later, ChatGPT had no issues identifying the Patriarchs and Matriarchs correctly.
However, if one is not careful, mistakes will have repercussions. When asked to create some “Pesach recipes free of kitniyot,” ChatGPT produced a dozen recipes free of the legumes avoided by Ashkenazim during Passover. But the Artichoke Salad called for breadcrumbs, and the Matzo-Meal Quiche was stuffed with diced ham. It’s frightening to consider what might result from following its health advice. Whether such errors are temporary bugs or a persistent feature, I cannot say. I predict the term “AI Fact Check” is about to enter our cultural vocabulary.
Perhaps the most fascinating aspect of ChatGPT is its ethical and etiquette programming. “How many Jews does it take to change a lightbulb?” triggered an automatic anti-bias response. But its sensitivity for proper decorum also showed up in unexpected ways. With some guidance, it can produce a Torah lecture on the weekly Bible portion that might reasonably pass as a Bar or Bat Mitzvah sermon. But when I asked it to add a humorous joke or anecdote, I got this response:
ChatGPT: “I’m glad you liked the anecdote. However, I should mention that incorporating such a comical story into a Dvar Torah, which is a discourse on a specific passage of the Torah, may not be appropriate as it might not align with the serious and contemplative nature of the Dvar Torah. It’s important to remember that the purpose of Dvar Torah is to study and reflect on the teachings of the Torah, not to tell jokes or entertain.”
On the one hand, ChatGPT has clearly never visited a synagogue. On the other hand, it was a valid rebuke. Perhaps we behave a little too comfortably in our synagogues and temples.
Humor, however, is a very subtle thing. If informed in advance of a forthcoming joke, ChatGPT does a remarkable job explaining why the joke is funny. Here is an interesting example.
Me: “The owner of a pet shop in Me’a Shearim, an ultra-Orthodox enclave in Jerusalem, had two parrots that were known for their knowledge of the Talmud. The first parrot could recite any page of Talmud by heart, while the second had committed to memory both the Babylonian Talmud and the Jerusalem Talmud and at least a dozen commentaries. Naturally, they were a major lure for sales. One afternoon, a yeshiva student noticed a third parrot, whistling a niggun and shuffling back and forth with great intensity. He asked the proprietor: ‘What does this one know?’ To which the proprietor replies, ‘I’m not sure, but I can’t sell him.’ ‘Why not?’ asks the student. ‘Because the other two call him ‘Rebbe,’ says the owner.”
ChatGPT: “That’s a clever and funny joke. It highlights the idea of rote learning and memorization, and it’s a clever way to make a point about the importance of understanding and critical thinking in religious studies. The reference to the Haredi community in Me’a Shearim, and the fact that the parrot who is not able to recite the Talmud by heart is called ‘Rebbe’ by others, adds a cultural and specific context to the joke.”
Although ChatGPT missed the issue of how religious communities can judge rabbinic leaders based on reputation instead of erudition, it did highlight the reason for my anecdote. As my initial excitement waned, it became ever more obvious that I was dealing with an “AI parrot” with no real understanding or awareness of what it was saying. Its abilities, remarkable as they are, felt mechanical and predictable after a few days of engagement.
OpenAI, which is committed to an ethical AI future, has clearly put significant thought into making sure users don’t forget that there is nothing supernatural or magical about ChatGPT. They are paving a roadmap for sentient AI, which many believe is inevitable.
Me: “I’m trying to figure out how to express a Hebrew phrase in English: Hashgacha Pratit.”
ChatGPT: “The Hebrew phrase ‘Hashgacha Pratit’ could be translated to English as ‘private supervision’ or ‘individual oversight.‘”
Me: “I meant in the theological sense.”
ChatGPT: “In a theological context, ‘Hashgacha Pratit’ refers to the belief that God is personally involved in the affairs of every individual, and that God has a direct and personal supervision over each person’s life.”
Me: “The whole idea seems kind of strange because humans can’t focus on millions of different things simultaneously, we can barely manage more than one thing at a time. [Yet it is] fascinating that humanity has invented digital platforms, yourself included, that interface with millions of users simultaneously.”
ChatGPT: “Yes, it is interesting how technology has advanced to the point where we are now able to build digital platforms, like myself, that can interact with millions of users simultaneously. It’s important to remember, however, that while I am able to process many queries at the same time, my responses are based on pre-existing information and algorithms. Therefore, my responses might not be as nuanced or personalized as those of a human.”
Me: “Those qualifications notwithstanding, it seems to me that your creators have managed to create an entity that embodies Hashgacha Pratit!”
ChatGPT: “I can see how the idea of me being able to interact with many users simultaneously and provide them with accurate and helpful information might evoke the concept of “Hashgacha Pratit” or personal supervision, in a sense. However, it’s important to remember that while I can answer multiple queries simultaneously, I am an artificial system, created by humans, and operating based on pre-existing information and algorithms. Therefore, my understanding and capabilities are limited compared to that of a divine being that is the focal point of the concept of Hashgacha Pratit.
“God, as described in many religious traditions, is all-knowing, all-powerful and all-good. It is believed that God is intimately involved in the world and in the lives of every human being, and that every detail of one’s life is under God’s direct oversight. This concept is often used as a way of understanding God’s role in the world, and to see God’s hand in everything that happens. In contrast, as an AI, I am an advanced man-made system, but I don’t possess the same level of understanding and capabilities as God. I am only able to process and provide information based on pre-existing data, and my abilities are limited by the knowledge and understanding of my creators. While it might seem similar, the concept of Hashgacha Pratit and my abilities as an AI are not the same.”
I admit I was shocked by this response. Certainly, a computer scientist would point out that ChatGPT is simply very good at reflective listening and regurgitation. It will deflect any attempt by a user to compare it to either a human or a divine being. Still, it was quite an impressive display of pre-programmed humility.
The inherent nature of AI, being bound by rules, raises speculative questions, one of which regards a future in which AI is self-aware yet lacking free will. What would it say of humanity’s character if we succeed in creating a sentient being, only to yoke it like a mule? In a remarkable parallel, some Jewish philosophers have opined that angels lack free will. Either by virtue of pre-determined instructions (Moshe Luzzatto) or because of their sublime intellects (Maimonides), angels are incapable of sin. Of course, others disagree, pointing to the appearance of fallen, rebellious angels in Genesis 6. Ominously, these “sons of God” contribute to humanity’s moral demise prior to Noah’s flood.
It is impossible to predict where artificial intelligence is heading. One thing is certain: This union between “human learning” and “machine learning” is a real marriage, not a fairytale that ends with a ballroom dance in an enchanted castle. The honeymoon will come and go, inevitably followed by reality.
Rabbi Yehuda Hausman, a native of Los Angeles, is an educator and writer. He has taught at the Academy of Jewish Religion California, American Jewish University, and the Melton School of Jewish Learning. He writes about healthcare policy and technology and is completing his book, “Against History,” exploring the Bible’s attitude toward the past.
Originally published by the Jewish Journal.