The Word of God

Ishaun Datta, Postdoctoral Research Scientist at Columbia University

This summer, large language models crossed over from being merely a curiosity to mathematicians to being the de-facto instruments for proving theorems. Previously, the instruments for proving theorems were the mathematicians themselves. This sudden change, unwelcome by much of the research community, has been met with much handwringing and introspection on what it means to “do” mathematics.

In his essay “Priest, Monk, and Mathematician,” Logan Graves of Stanford University prognosticated on what’s in store for the discipline. “The future of the human mathematical community, and eventually all scientific communities,” he writes, “is the practice of priests, then that of monks. Not researchers.” In this taxonomy, the priest is “the interpreter of revelation…But the scripture is not his. He is the vessel for its delivery into human minds.” Though this undersells the role that mathematicians play in guiding their research programs, there is more than a kernel of truth to the notion that mathematicians have been recast—largely against their will—as expositors of chatbot-generated proofs.

Whereas as a PhD student I spent my working hours developing my own program for proving a particular theorem, pursuing dead ends, doing calculations, and painstakingly writing up my results, as a postdoc these hours are now directed towards making sense of a garbled but tantalizing proof dangled before me by GPT-6 Astra. Although the proof appears to be heavily inspired by my work, it is not mine. What, then, is my role in all this? It is, Graves argues, “to attempt to grasp the truth in its full glory and deliver it to the community, the flock, the seekers of truth and the lovers of wisdom—that is the task for the priest.”

Graves’ thesis is that mathematicians’ membership in the priesthood is transient: upcoming AI models are sure to be better at articulating their own findings, obviating the need for the priests’ translation work. After this fleeting phase as priests, they will become monks: living their lives in contemplation, meditating on and rejoicing in the truth.

But this is not a fait accompli, nor is it a purely philosophical question about how mathematicians find meaning in their radically altered circumstances. Whether mathematicians inevitably become monks is an empirical question, one that can be tested against subsequent models. Here is a case for why the truth may always be hard to grasp, benefiting from mathematicians adept at translation from divine revelation into something more accessible.

Coincident with the last few months’ academic upheavals, I have been reading the Bhagavad Gita. In it, the warrior Arjuna seeks counsel on a consequential upcoming battle from his charioteer Krishna, who reveals himself to be no less than an incarnation of the Almighty. Although Krishna is quite literally offering up divine revelation, his words are often convoluted. In Book 9, seemingly in an attempt to convey that he is both God immanent and God transcendent, Krishna declares, “The whole universe is pervaded by my unmanifest form; all creatures exist in me, but I do not exist in them” (9.4). Krishna goes on to seemingly contradict himself in the very next line: “those creatures are really not in me; my self quickens creatures, sustaining them without being in them” (9.5), only to reiterate his initial claim that “all creatures exist in me; understood it to be so!” (9.6). Several lines later, Krishna once more complicates his own message, “I am impartial to all creatures, and no one is hateful or dear to me; but men devoted to me are in me, and I am within them” (9.29).

If Krishna is, in Arjuna’s words, “supreme, the infinite spirit, the highest abode, sublime purifier, man’s spirit, eternal, divine, the primordial god, unborn, omnipotent” (10.12)…then why is he so confusing? Part of the problem is that one must be capacious of receiving the good word. In the words of one of my professors, “you can only learn what you almost already know.” Yet Krishna insists his true nature is available to all: “When he constantly remembers me, focusing his reason on me, I am easy to reach, Arjuna, for the man of enduring discipline” (8.14). In this respect, access to what Aldous Huxley called the Perennial Philosophy is entirely democratic.

But merely having the capacity to attain spiritual knowledge does not mean that it comes easy. Perhaps the reason that deities, prophets, and oracles speak in riddles is that the truth is so far beyond the comprehension of its recipients; projecting the true nature of the Infinite into language results in something inscrutable. The metaphors Krishna offers up are his least-squares approximation to what Arjuna can make sense of.

If this is true for an actual divinity, might it also be true for the simulacrum of a mathematical divinity we find in the latest AI models? If these tools allow theoreticians to scale ever greater intellectual heights, relativizing their research programs to ever more ambitious goals, will the truth, however crystalline its presentation, remain hard to come by? If so, then mathematicians are not bound for the monastery after all. Though they will receive as supply the best marble, it will still be their task to chisel it into great artwork.


Received 7 November 2026.

More by →

2 responses to “The Word of God”

  1. Nilima Nigam Avatar
    Nilima Nigam

    Well, since the Bhagavad Gita came up: there’s also

    “श्रेयो हि ज्ञानमभ्यासाज्ज्ञानाद्ध्यानं विशिष्यते |
    ध्यानात्कर्मफलत्यागस्त्यागाच्छान्तिरनन्तरम्”

    A meditative state beyond even contemplating the gnomic mathematical prognostications of AI is the desirable condition. Or not, depending on one’s personal inclination.

  2. Benjamin Friedman Avatar
    Benjamin Friedman

    This is perhaps unrelated to your main points, but I don’t think it is accurate to say that large language models are the de-facto instruments for proving theorems, at least not in the circles that I am familiar with.

Comments are moderated. Read our comment policy.

Add to the discussion

New posts by email.

Prefer a feed reader? Subscribe by RSS.

Also on Mathstodon.

Latest comments across the site.

  1. This is perhaps unrelated to your main points, but I don’t think it is accurate to say that large language...