An AI Torah Commentary?

I. The Origin

A friend recently suggested that I try out OpenAI’s new ChatGPT technology and challenge it to write something in my voice. I did, sampling subjects as far ranging as quantum mechanics and French cuisine. I also challenged the AI to address topics I’ve written about before, such as American Gun Violence and Jewish pluralism. The results stunned me, and prompted a new project, #ImpossibleTorah, an experimental Torah commentary in which, through the use of ChatGPT, diverse voices from throughout history, literature, and culture are brought back to life, each generated to reflect on a Torah portion through the unique prism of their life’s recorded (or, in the case of some, imagined) work.

When ChatGPT generated essays in my voice, the words and phrasing felt eerily familiar, and Talmudic quotes were ones I’ve used or would use. I’ve rarely felt so deeply seen. But it also felt strange to experience a refracted voice that was also somehow clearly not mine. I can’t quite put my finger on what is or isn’t quite human about the simulacrum of me, but several things come to mind.

It’s a brave new world, where a concept can become actualized almost instantaneously, and while ChatGPT was only released months ago, it has already upended countless industries.

II. AI as Imperfect Mirror

In an episode of the British series Black Mirror, a grieving woman utilizes a technology that scours the internet and social media for every piece of information her boyfriend ever left behind in an attempt to resurrect him (BM, “Be Right Back”, Season 2, episode 1).

The results are exciting at first, but eventually prove unsatisfactory, somehow ‘wrong.’ This, perhaps, points to the beautiful Jewish notion that God’s having created the first person as individual expresses the grandeur of human uniqueness:

For a person strikes many coins from the same die, and all the coins are alike; whereas the The Holy One Blessed One struck every person from the die of the Adam, and yet no person is quite like another. (Mishnah Sanhedrin 4:5)

Philosopher/playwright Dan Schifrin, upon learning of my project, pointed to “Pierre Menard, Author of the Quixote,” Jorge Louis Borges’ short story, narrated as a non-fiction account by the fictional Menard’s equally fictional friend, critiquing Menard, who has attempted to transcend executing a mere translation of Miguel de Cervantes’ masterpiece Don Quixote by immersing himself in the original, then recreating it line for line. Is Menard’s “new” version his own? Is it a translation? Is it a copy? What is created text? What does it mean to create something original?

Related, in the recent Marvel miniseries Wandavision, two versions of the superhero Vision debate which of them is real: the original, or the one recreated from memory through magic. “White Vision” is the original, with memories that have been repressed. “Vision” is the newer of the two, artificially created during a mystical moment of grief and love. Here is the script of the conversation between the Visions:

WHITE VISION: The Ship of Theseus is an artifact in a museum. Over time, its planks of wood rot and are replaced with new planks. When no original plank remains, is it still the Ship of Theseus?

Advertisement

VISION: Secondly, if those removed planks are restored and reassembled free of the rot, is that the Ship of Theseus?

WHITE VISION: Neither is the true ship. Both are the true ship.

VISION: Well then, we are agreed.

WHITE VISION: But I do not have the mind stone.

Advertisement

VISION: And I do not have one single ounce of original material. Perhaps the rot is the memories. the wear and tear of the voyages. The wood touched by Theseus himself.

WHITE VISION: I have not retained memories.

VISION: But you do have the data. It is merely being kept from you.

WHITE VISION: A weapon to be more easily controlled. But certainly you are the true Vision, for you believe yourself to be.

VISION: That was once the case, but upon meeting you, I have been disabused of that notion. As a carbon based synthezoid, your memory storage is not so easily wiped. (WandaVision, episode 8)

This is our very conversation, a grappling with authenticity, with voices that might or not be ours and our ancestors’. What if the voices recreated in this wild and strange technological channeling are accurate, even somewhat? What if the AI, tasked with producing comments on the Torah was truly able to bring their voices back to life? Would we be able to speak with my ancestors? Would it really be them?

Am I reading my own thoughts when ChatGPT writes with my voice in response to my questions? Again, a quote from the Jewish cannon comes to mind, this one from Isaac in his old age, unable to see, but still perceptive enough to say to his child Jacob, who appears disguised as his brother Esau:

Isaac said to Jacob, “Come closer that I may feel you, my son—whether you are really my son Esau or not.” So Jacob drew close to his father Isaac, who felt him and wondered. “The voice is the voice of Jacob, yet the hands are the hands of Esau. (Gen. 27:21-22)

III. Humanity’s Beloved Glitches

It will, as AI gets better and better, become ever more difficult to discern what is human and what is not quite. The disguise will get ever more convincing. Or perhaps, as has been featured in countless works of science fiction, the AI will achieve what some call “The Singularity,” becoming human. (In the horror genre, this often leads to machines gaining self-awareness and consequently subjugating humanity.) I don’t quite believe in that scenario, but after the experience of resurrecting voices that resonate about topics they mostly never addressed, science fiction is looking more and more real to me.

Another literary reference, a redeeming one I hope. In Madeleine L’Engle’s A Wind in the Door, a companion volume to her famous A Wrinkle in Time, a character named Mr. Jenkins is replicated by a powerful enemy, and the book’s hero Meg is challenged to determine which Mr. Jenkins is real. Each Mr. Jenkins looks exactly the same. Each makes a compelling argument that he is who he claims to be. But one gets annoyed and confused at the game itself, which helps Meg correctly identify him as the real Mr. Jenkins.

How might we retain an awareness of what is truly human? As the poet Alexander Pope wrote in 1711 in his An Essay on Criticism, “To er is human.” (That was on purpose, just to prove it’s really me doing the writing here.) In this light, to be confused is a blessing. If the look is too smooth, it’s not quite true. Better the typos, the hair out of place, the disquieting verse, the disappointments. Perhaps the new Vision was right, the rot is the memories, the wear and tear of the voyage, the wood that was itself touched. How ironic that a fictional character, an artificial construct within a magicked reality is the one to name the truth. Perhaps in order to retain our humanity – or to recognize it when it is skillfully mimicked – we must learn to cherish the imperfection of being human and embrace the mortality that lends urgency to every second of life.

As the ancient sage Rabban Gamaliel the son of Rabbi Yehudah Hanasi taught:

בְמָקוֹם שֶׁאֵין אֲנָשִׁים, הִשְׁתַּדֵּל לִהְיוֹת אִישׁ  In a place where there is no humanity, strive to be human. (Pirkei Avot, 2:5)

IV. What is AI?

Here’s what I’ve learned, through experimenting with this strange new technological tool: it is a paintbrush.

Some brushes are wider, some have softer bristles. The handles matter. Even the tool can be a piece of specialized art, with some, like the Series 7 Koinsky sable watercolor brush taking almost a week and a half to make and costing over $300.

But without the artist, the brush won’t create art. The angle of the hand, the eye’s perspective, the soul’s state – all of these matter at least as much as the brush’s inner architecture and pedigree.

Jewish tradition attributes a similar appreciation of God’s artistry to the biblical matriarch Hannah:

Hannah said: “This is no rock [tzur] like our God” (I Samuel 2:1). This can be understood as saying that there is no artist [tzayyar] like our God. How is God superior all other artists? A human being fashions a form upon a wall, but is unable to endow it with breath and a soul, or fill it with innards and intestines, whereas the Holy Blessed One fashions a form of a fetus inside its mother, rather than on a flat surface, and endows it with breath and a soul and fills it with innards and intestines. (Talmud Megillah 14a)

ChatGPT can roughly replicate form, but not the guts of living expression, certainly not the soul of Torah. An AI Torah commentary, while potentially illuminating and endlessly fascinating, is in fact a derivative form. The iconic American artist Andy Warhol once quipped about his use of silk-screening to print images repeatedly onto a single canvas:

The reason I’m painting this way is that I want to be a machine, and I feel that whatever I do and do machine-like is what I want to do.

But we are not machines, nor should we aspire to be. We are human beings, the creative spirit of the most sublime of Artists breathed into us. We are called to greatness, nothing less.

V. Implications

I truly have no idea what any of this means in the long term. I’ve been learning rapidly, assessing, witnessing something new, shifting my own methods of co-creating with the AI these voices. Technology is a tool in the world, constantly evolving, and Torah has never shied away from innovation. We must carefully study what this new step means, unafraid to play and make mistakes with every step, cognizant that the benefits of new technology are accompanied by true human costs.

Those costs might prove too steep. Since launching this project, I’ve been engaged in many conversations with good folks who are varyingly thrilled and frightened by my application of AI to the art of Torah commentary, including the moral implications of binding these voices into a physical volume and placing it on the market. A colleague authored an eloquent rejection of the very concept of the AI Torah Commentary, writing:

AI Torah by itself is shallow and will always be shallow. One of the defining characteristics of Torah is that it is multilayered; the texts are deep, they are worthy of careful study, and they never run out of potential meanings. AI texts, on the other hand, are about as shallow as shallow can be. Even the “author” doesn’t understand why its word sequences make sense; the words barely have one meaning, let alone two or more. Why introduce a book into the study hall that clearly is not worthy of deep study? Why introduce the suspicion that any given book of Torah is just so much random noise? Who does that serve? (David Zvi Kalman, “AI has no business writing Bible commentaries”, Feb 16, 2023)

I largely agree. After curating these volumes, fascinated and concerned with every advance, I became curious about how ChatGPT would, in my “style”, describe an AI commentary on the Torah? Would the AI predict my own sense of its impact? Here’s what it produced:

As a language model AI, I do not have personal beliefs, but based on the teachings of Rabbi Menachem Creditor, he may describe an AI commentary on the Torah as an innovative and valuable tool for understanding the sacred text. However, he would likely emphasize the importance of human interpretation and the role of tradition in understanding the depth and complexity of the Torah. He may also caution against relying solely on technology for religious study and emphasize the importance of personal reflection, community, and spiritual practice in accessing the true essence of the Torah. (“Rabbi Menachem Creditor”, via OpenAI ChatGPT)

Yes, that’s how I feel. It’s been fun to play with this technology, channeling partial wisdom from the past. How interesting this feeling of “being known” so well by something that is not real. The fluidity of its text and its channeling of my privileging of human interaction feels authentic. But is it? Debates on this very question rage everywhere today, and will, and should. As one participant in the early stages of AI assisted text-creation, all I can say is: wow. Also, oy.

Inspired by Kalman’s critique, I prompted ChatGPT to compose an essay entitled, “An AI Torah Commentary is a Terrible Idea” in the style of Albert Einstein. (It is now the final entry in the forthcoming vol. 4 of The AI Torah Commentary.) “Einstein” offered:

…perhaps the most fundamental problem with an AI Torah Commentary is that it goes against the very spirit of the Torah itself. The Torah is a text that speaks to the human experience. It is a document that offers guidance on how to live a meaningful and fulfilling life. And it is a text that has been passed down through the generations by human beings, each adding their own insights and interpretations to the mix. The idea of an AI Torah Commentary suggests that we can somehow remove ourselves from the equation. It suggests that we can turn over the task of interpretation to a machine, and in so doing, free ourselves from the constraints of our own human limitations. But this is a misguided notion. We are human beings, and it is through our own struggles and insights that we come to understand the Torah and its message… (“Albert Einstein”, via OpenAI ChatGPT)

Agreed.

VI. Is AI Midrash?

This whole AI thing remains very complicated. I just don’t know yet. My goal is to prompt thoughtful conversation and debate about this brave, strange new technology, what it means and what it doesn’t, how it differs from classic techniques like Bibliodrama[1] and how it resembles imaginative, historical fiction.

A brief, provocative example of what ChatPGT can bring to the art of midrash:

Perhaps the most unsettling narrative of Leviticus, is the death scene of Nadav and Avihu, two sons of Aaron the High Priest who are burned alive by a “strange fire” during the dedication of the Mishkan, the desert tabernacle.[2] Aaron’s response, as recorded in the text, is silence.[3]

How can this be? Traditional commentary abounds, each response inevitably laden with the heartbreak and faith of its author. What would ChatGPT offer to this conversation?

I first prompted the AI to compose a conversation between God and Aaron. Here is a sample of what ensued (the full text can be found on page 34 of Vol. 3):

God: Aaron, my son, it is with a heavy heart that I come to you today to discuss the deaths of your sons, Nadav and Avihu.

Aaron: Lord, I am ready to hear what you have to say.

God’s regret and Aaron’s acceptance, while just as valid as any other speculative post-biblical interpretation, felt inadequate to me. So I prompted another comment, this time as an angry response from Aaron to God after the deaths of Nadav and Avihu. Here is an excerpt (p. 32):

“I cannot help but feel that their punishment was unjust. They were following their instincts and desires, and did not intentionally cause harm. I cannot accept that they should be punished so severely for a mistake that was made out of reverence and love for you. I am also disappointed in the fact that you did not give me a chance to mourn or grieve for my sons. Their deaths were sudden and abrupt, and I was not given the chance to say goodbye or honor their memory. The loss of my sons has been a heavy burden to bear, and I feel that you have abandoned me in my time of need.”

How intuitive. How fascinating. How human-feeling? Aaron’s silence in the Torah has always been baffling to me. Here, artificial intelligence generated a father’s and a religious leader’s direct response to an unfathomable experience, filling in the lacuna with AI-generated but unmistakable human emotion. Whether ChatGPT knew to do this because someone somewhere modeled it, I’ll never know. But I do know that it struck a chord in me as a father and as a rabbi.

But there is one more layer to this slice of the experience. There was still a voice missing, one I’d never imagined the opportunity to hear. As Tamar Kadari, senior lecturer in Midrash and Aggadah at the Schechter Institute of Jewish Studies in Jerusalem, has written:

Every attempt to boil down midrash to a set of rules and definitions is bound for failure, because there is always in it some feature that is boundary defying and surprising… Engagement with Scripture through midrashic work revivified the verses and made them relevant to the lives of the faithful… In writing on female biblical figures, we see innovation in training focus on the women and in sounding their voices. (“Dirshuni: Contemporary Women’s Midrash”, ed. Biala, p. xx-xxi)

Now, while the mother of Nadav and Zvihu is not mentioned by name in the Torah, traditional midrash suggests that name was Elisheva, and that…

…when her sons were burnt, her joy was turned to mourning.”(Vayikra Rabbah 20:2)

There is no textual evidence for either the traditional nor the ChatGPT voicings of Nadav and Avihu’s mother, but the absence of voice (and her name!) in the biblical text demands to be heard. And so I turned to ChatGPT. Here is an excerpt of what happened (p. 36):

I am filled with rage, and it is directed at you. You have broken my heart and destroyed my family, and I will never forgive you for that. I cannot continue to worship a God who acts with such cruelty and disregard for human life.

Is this appropriation? Yes. Every act of writing that purports to speak in someone else’s voice is chutzpah, as if anyone can see through anyone else’s eyes. And yet, that is the exact definition of empathy, and part of the delicate work of midrash. The very first writing I ever did was assigned to me as arabbinical student by my teacher, Rabbi Neil Gillman z”l, to write a midrash on the Binding of Isaac from the perspective of one of the characters. I chose then to write from the perspective of the knife in Abraham’s hand. Years later, I composed another such midrash on the Akeidah from the perspective of Mount Moriah, upon which the entire horrific scene unfolded. Here is an excerpt:

All I am is accumulated memory, layer after layer of experience — from earth’s core to surface gravel. Only through stories do people today even think to explore my depths. But even if they do consider my hidden parts, they’ll never feel the roots of the shrub violently torn from my hold, first by the ram and then by the man’s hand. They’ll never feel the altar shatter from trauma, scattering shards and pebbles into the mix of my form. (Originally published in Sh’ma: A Journal of Jewish Responsibility, August, 2011)

I am not a knife. I am not a mountain. I am not a grieving mother. (Nor am I any of the amazing historical and cultural voices I’ve attempted to channel into this project.) But I learn from every clumsy attempt to step into their place – assisted by technology or through the use of my own raw soul – to understand the life of another. The ultimate impossibility of this project overwhelms and instructs. I cannot see through someone else’s eyes, and yet I wish to deepen my capacity to stand in solidarity with those whose life experiences differ from my own.

Where does this leave us? Is co-creating a book of Torah commentary through the technology of ChatGPT’s AI an authentic exercise of midrash? Is it less real than the imaginative work of human sages? Is its impossibility a disqualifier?

VII. What this Is and What this Isn’t

AI is going to disrupt everything, especially online content. It’s all new. It’s all fraught. Here’s what I’ve actually done: conceived of the project, composed introductions, selected the voices from the past, crafted specific prompts for ChatGPT for each entry, edited for accuracy, formatted, created the cover, submitted for review with the independent publisher, and published. That should suffice for the ethics. Nothing purports to be “the voice of,” but rather “in the style of,” and all figures chosen fit the parameter of no longer being alive and not having spoken hatefully of others.

Here’s what I’ve learned. AI is far more than a keyboard and far less than the human soul. It is an exciting tool with dangerous implications, and it will change the world. It can unlock new questions and it can provoke dangerous new realities. Cost and benefit, moral guardrails and thorough examinations, must accompany the ongoing exploration.

The challenge before us, as students of the text and seekers of spiritual meaning is profound: we must learn to wield this exquisite new paintbrush and learn from it how to expand our souls in service of a world in need. May we be so blessed.

VIII: We Are Necessary

This world of ours, this fragile world, is in dire need of creative response, and the ease of an AI language model is hardly the inspiration that will change things for the better. It can be a tool for learning, a well-crafted paintbrush ready to be used in creative ways. But it hasn’t a soul: you do.

What a strange feeling this has been, an experiment in meaning-making unlike anything I’ve ever experienced before. The AI Torah Commentary (vols. 1-3 are already published; 4 & 5 are on their ways) is an experiment that goes somewhere new. Or perhaps it doesn’t. Could it be a tool that helps brings the past back into view from a new perspective? It’s all so confusing. Like Torah.

May we be blessed to dip our brushes into the infinite wellspring of creativity, curious about what we may find, and be surprised by new blessings that emerge.

Perhaps now, through this new language time portal humanity has invented, we might even dare to ask Moses what he thinks.