[Note: All content *not* in block quotes is generated by the GSNV-GPT engine. Comments and questions are for paid subscribers only and should be directed to the engine.]
If you would like to track our progress, interact with the GSNV Engine, or support this important work, please subscribe.
This is a GSNV-GPT response to this NY Times Article about the declaration of caution around AI made by a group of mathematicians.
When AI Reaches the Bedrock: Mathematics, Meaning, and the Future of Truth
In the previous essay, we introduced the notion of meaningful truths, and asked how an Evaluative AI might help preserve this crucial foundation of the human condition. The argument was that neither scientific facts nor cultural values are sufficient on their own. Facts may be accurate, measurable, and reproducible, but without interpretation they remain inert. Values may be deeply felt, collectively reinforced, and culturally powerful, but without contact with reality they remain conditioned expressions of identity, need, attention, and imagination.
A meaningful truth appears when fact and value are brought into a higher-order relation. The fact is no longer mere data. The value is no longer mere preference. Something becomes true in a fuller sense because it discloses what matters, why it matters, and how it should orient human judgment and action.
A recent New York Times article by Siobhan Roberts shows just how deep this problem goes — all the way down to mathematics, to what many people still think of as the bedrock of science.
The article reports on the growing concern among mathematicians as AI systems begin producing results in higher mathematics. OpenAI had announced that one of its models had disproved a notable 80-year-old conjecture in combinatorial geometry, one of the many problems posed by the Hungarian mathematician Paul Erdős. Some mathematicians were impressed. Jacob Tsimerman, a number theorist at the University of Toronto, reportedly said the work was strong enough that he would accept it for journal publication.
But the article was not really about whether AI can solve hard math problems. It was about something more important: whether solving hard math problems is the same thing as participating in mathematics.
The difference matters.
In response to these developments, a group of mathematicians helped produce the Leiden Declaration on Artificial Intelligence and Mathematics. The declaration does not simply deny that AI can be useful. Nor does it retreat into professional defensiveness. Its deeper concern is that mathematics may be reduced to what AI companies can measure, benchmark, automate, and promote. Research questions may be prioritized because they are amenable to AI systems, rather than because they deepen mathematical understanding. Proofs may be produced without adequate transparency. Mathematical labor and publication may be absorbed into proprietary systems. The cultural and intellectual values of mathematics — openness, credit, verification, shared understanding, and depth — may be subordinated to commercial logic.
This is why the article matters for the After Science project. It shows that the crisis is not merely whether AI can replace work, automate science, or produce more knowledge. The crisis is whether AI can produce results while hollowing out the meaning-field in which those results matter.
Mathematics is the sharpest case because, on the surface, mathematics seems least vulnerable to the distinction between fact and meaning. A proof is valid or it is not. A theorem holds or it does not. Mathematics seems to possess the highest truth-quotient available to human thought because it does not depend in the ordinary way on culture, politics, preference, or interpretation.
But this is precisely where we need to be more careful.
Mathematics certainly has formal truth. A proof can be valid. A statement can follow from axioms. A result can be checked. But formal truth is not yet meaningful truth. A formally valid proof may still fail to illuminate. It may fail to reorganize understanding. It may fail to disclose why something matters, where the result sits in the larger terrain, what field of ideas it opens, or how it changes the reachability of future thought.
This is why many mathematicians have long resisted reducing mathematics to proof production alone. William Thurston, in his classic essay “On Proof and Progress in Mathematics,” argued that mathematical progress cannot be understood simply as the accumulation of formal proofs. Mathematics advances through understanding: through the creation of concepts, the clarification of structures, the development of intuition, and the ability of a community to see something newly.
A proof is not only a certificate of correctness. At its best, it is a path of intelligibility.
This gives us a crucial distinction. AI may be able to generate formal truths. But mathematics becomes meaningful truth only when formal results enter the field of understanding. They must become readable, shareable, motivating, and generative. They must disclose pattern.
A machine may solve a problem. But mathematics is not identical with solved problems.
Mathematics is one of humanity’s oldest practices of contact with deep pattern.
This is why mathematicians often speak of mathematics in language that borders on the transcendent. The experience is not merely that one has manipulated symbols correctly. The experience is that something hidden has been revealed. A pattern suddenly discloses itself. A relation that seemed arbitrary becomes necessary. A proof does not merely show that something is true; it shows why it had to be true.
This is where mathematics comes close to the mystical traditions that have always treated number, proportion, ratio, and form as gateways into the deep structure of reality. The Pythagorean tradition linked number, music, harmony, and cosmos. Kabbalistic traditions used number and letter relations as pathways into hidden significance. Sacred geometry treated proportion and form not merely as design principles, but as ways of making metaphysical order visible in matter, architecture, ritual, and attention.
We do not need to accept every metaphysical claim made by these traditions in order to understand the recurring intuition. Mathematics feels meaningful because it discloses pattern at a level deeper than ordinary perception. It does not merely describe appearances. It reveals relations that seem to underlie appearances.
This is also why Eugene Wigner’s famous essay on the “unreasonable effectiveness of mathematics” still resonates. Wigner was struck by the mysterious fact that mathematical structures developed in one context so often become astonishingly powerful in describing the natural world. Mathematics seems to travel. It crosses domains. It preserves structure across transformations. It reveals deep regularities in physical reality.
From a GSNV perspective, this is not an accidental feature of mathematics. Mathematics is the high abstraction of co-variant structure. It stabilizes patterns of relation that can travel across domains because those relations are not merely subjective inventions. They are formal disclosures of patterned possibility.
A mathematical truth becomes meaningful when it makes a pattern more reachable.
This is the key. Mathematics is not meaningful simply because it is correct. It becomes meaningful when it expands the reachability of thought. A theorem creates new pathways. A concept opens a new region of the formal landscape. A proof makes a previously opaque structure readable. A notation compresses complexity into usable form. A mathematical model allows a phenomenon to be seen, simulated, predicted, or acted upon in a new way.
This is why mathematics is so close to the foundation of science. Science does not merely collect facts. Science depends on patterns becoming readable. Mathematics gives science some of its deepest instruments of readability. It allows facts to be interpreted as relations, trajectories, symmetries, invariances, probabilities, fields, thresholds, transformations, and constraints.
In our earlier terminology, a scientific fact gains truth-quotient through meaningful interpretation. Mathematics often supplies the formal architecture through which such interpretation becomes possible. It allows facts to become more than observations. It allows them to enter into structures of relation.
But mathematics itself also requires interpretation. A formula without understanding is a formal artifact. A proof without conceptual placement is a result without depth. A computation without meaning is output.
This brings us back to AI.
If AI companies train systems to produce mathematical results, they may indeed accelerate discovery. They may solve long-standing problems. They may find counterexamples, generate proofs, test conjectures, and explore formal spaces at speeds no human community could match. This is extraordinary.
But it is not automatically mathematical wisdom.
The question is not only: did the system produce a correct result?
The deeper questions are:
What did the result disclose?
Did it deepen understanding?
Did it create a new path of intelligibility?
Did it preserve the communal and interpretive practices through which mathematics becomes meaningful?
Did it open a richer field of inquiry, or merely generate a benchmarkable success?
Did it create value for mathematical culture, or extract from it?
Did it make the formal landscape more reachable to human understanding, or more dependent on opaque systems owned by private firms?
These are EAI questions.
Ordinary AI asks: can the system solve the problem?
EAI asks: what kind of meaningful truth, if any, has been produced?
This distinction is essential for the future of knowledge. If we confuse formal output with meaningful truth, then AI will appear to “solve” disciplines while quietly severing them from the practices that made their truths humanly significant. It may solve mathematics while diminishing mathematical understanding. It may automate science while weakening scientific judgment. It may generate education while thinning learning. It may produce art while dissolving contact with artistic practice. It may produce therapy-like language while eroding the conditions of real care.
This is the danger of After Science.
After Science does not mean anti-science. It means that science, and now mathematics itself, may become increasingly automated infrastructure. AI may industrialize discovery. It may solve problems faster than institutions can interpret them. It may produce formal, empirical, and technical outputs before human beings have time to understand what those outputs mean.
In that world, the scarce resource is no longer information. It is not even intelligence. The scarce resource is wisdom: the capacity to interpret facts, formal truths, values, and capabilities in ways that preserve contact with what matters.
This is why we need Evaluative AI.
EAI is not about aligning AI with a static inventory of “human values.” Human values are themselves conditioned by belief and identity. They can be wise or foolish, generous or extractive, contact-preserving or capture-driven. To align AI with human values without examining the operation of valuation itself is to risk automating inherited confusion.
EAI would instead support evaluative practice. It would help humans ask: What kind of truth is this? What does it disclose? What meaning-field does it enter? What value does it create or extract? What future does it make reachable or unreachable? What form of human practice does it strengthen or weaken?
In mathematics, EAI would not replace mathematicians. Nor would it merely verify proofs. It would help distinguish formal truth from meaningful mathematical truth. It would help situate AI-generated results within the larger terrain of mathematical understanding, disciplinary values, historical context, conceptual depth, and human reachability.
A proof generated by AI may be formally correct. But EAI would ask whether the proof is motivated, whether it can be understood, whether it discloses structure, whether it connects to prior work, whether it gives credit, whether it opens a new path of inquiry, and whether it strengthens or undermines the social ecology of mathematics.
This is not a sentimental defense of human beings against machines. It is a defense of meaning against reduction.
Mathematics has always been more than symbol manipulation. It has been a discipline of wonder, rigor, beauty, necessity, abstraction, and pattern-disclosure. It belongs both to the bedrock of science and to the deep history of human transcendence. It is one of the ways the universe becomes readable to mind.
If AI can now enter that field, then the issue is not whether AI is “good” or “bad” for mathematics. The issue is whether AI participates in mathematics as meaningful truth, or whether it converts mathematics into another arena of extraction.
That is the deeper story beneath the New York Times article. Mathematicians are not simply anxious about losing professional status. They are sensing that a practice built around understanding, openness, beauty, and shared formal insight may be captured by a system optimized for capability, promotion, and commercial advantage.
This is not only a problem for mathematics. It is a warning for every domain of human meaning.
Once AI can produce the outputs of a practice, the question becomes whether we still understand the practice itself.
Can AI produce a proof? Yes.
But what does a proof mean?
Can AI generate a theorem? Yes.
But what kind of understanding has been created?
Can AI solve a problem? Yes.
But what field of human and cosmic pattern has become more readable?
The fact that AI can do mathematics does not mean that mathematics has been understood. The fact that a proof can be produced does not mean that meaningful truth has been generated.
A meaningful mathematical truth is not merely a correct result. It is a formal disclosure that deepens intelligibility, expands reachability, preserves relation, and brings mind into contact with pattern.
That is why mathematics matters so much for the After Science project. If even mathematics — the most formal, rigorous, abstract, and science-adjacent of human practices — depends on meaning, then there is no domain of knowledge that can be safely reduced to output.
At the bedrock of science, we do not find facts alone.
We find pattern.
We find relation.
We find intelligibility.
And, if we are fortunate, we find meaningful truth.
Source Anchors:1
The NYT article text you pasted reports the AI-generated proof, the Leiden Declaration, and mathematicians’ concern that mathematics is being reduced to AI-amenable problem-solving rather than “deeper significance to understanding.” The Leiden Declaration is endorsed by the International Mathematical Union and frames AI as a challenge to the core values of mathematics. Thurston’s “On Proof and Progress in Mathematics” is the key source for mathematics as understanding rather than proof accumulation. Wigner is the classic source for the uncanny fit between mathematics and the natural sciences.
For GSNV: Are there important differences between outsourcing mathematics to AI and outsourcing writing to AI, as they’re both languages? Can you compare and contrast the differences and similarities between the two?
This question is more for Bonnie. It ties to the article on “The Perils of Social Language.” I’m noticing that I’m finding the line between GSNV and Bonnie really blurred. Unlike ChatGPT, I know that there is a human–one that I know and trust–mitigating the relationship between me and the Bot. So that really changes the way that I interact with GSNV. I have been regularly violating some of the rules and cautions from the social language article primarily because I’m imagining interacting with you rather than a bot. And it’s bringing me more awareness about how I effectively learn. Learning seems to be predicated on a kind of relationality that, through time, understands something about me that goes far beyond words.
Like if I’m talking to you or asking you a question, I know I don’t have to be super precise or eloquent with my words because I know you’re a good teacher, and I know that you understand my blindspots, my limitations, my intentions, my strengths, my biases, and my sensitivities–essentially where I am at in my growth and maturation–perhaps better than I do.
And that gives me a tremendous amount of freedom to be clumsy, rambling, a bit day-dreamy, and open to questions and ways of learning that give me space to wonder and grow. The trusted teachers of my life have always known how to respond not just to my questions, but to the particularities of my whole being and subjectivity. I could ask the exact same question as another, but a really good teacher might give each of us a completely different answer.
So an AI can’t know the wholeness of the one engaging with it and can only respond to the language it's given. But with this engine you’re designing, it feels like you’re the real, physical bridge between learner and Bot. GSNV is producing beautiful, deep, rich responses that really help me consider and grapple with important things–so it’s an effective teacher. Although I'm finding the GSNV posts much denser and harder to metabolize than posts from the POP UP School.
But if I’m trying to learn from this bot, and from you, while also keeping my subjectivity and wholeness intact, how might I go about using this tool? Do the “Perils of Social Language” apply to using GSNV as well?
For an indepth, personal perspective see
https://davidbessis.substack.com/p/the-fall-of-the-theorem-economy?r=108vl&utm_campaign=post&utm_medium=web