Beyond our Mathematics
CMAP (Centre de Mathématiques APpliquées) UMR CNRS 7641, École polytechnique, Institut Polytechnique de Paris, CNRS, France
[Site Map and Help [Plan du Site et Aide]]
[The Y2K Bug [Le bug de l'an 2000]]
[Are we ready for the Year 2038 [Notre informatique est-elle prête pour l'An 2038]?]
[Real Numbers don't exist in Computers and Floating Point Computations aren't safe. [Les Nombres Réels n'existent pas dans les Ordinateurs et les Calculs Flottants ne sont pas sûrs.]]
[Please, visit A Virtual Machine for Exploring Space-Time and Beyond, the place where you can find more than 10.000 pictures and animations between Art and Science]
(CMAP28 WWW site: this page was created on 09/23/2026 and last updated on 09/23/2026 18:59:56 -CEST-)
[en français/in french]
For millennia, Mathematics was essentially
concerned with concrete applications: dividing up an inheritance, accounting for the regularities
of astronomical phenomena, ... And to do so, people had nothing but their brains and their fingers.
Then, very gradually, tools to assist with calculation began to appear,
including thanks to the work of:
- John Neper
(logarithms and the slide rule,
1550-1617),
- Blaise Pascal
(the adding machine -la pascaline-,
1623-1662),
- Gottfried Wilhelm Leibniz
(the multiplying machine,
1646-1716),
- Jacques Vaucanson
(the automatons,
1709-1782),
- Joseph Marie Jacquard
(the loom, a precursor to programmable machines
-first perforated supports: cards et ribbons-,
1752-1834),
- Charles Babbage
(the first computer -mechanical-,
1791-1871),
- Lady Ada Lovelace
(the firts programmer in history on Charles Babbage's machine,
1815-1852),
- George Boole
(the binary algebra {0,1},
1815-1864),
- John von Neumann
(the von Neumann architecture,
1903-1957),
- Alan Turing
(computability, universal Turing machine,
1912-1954),
- etc...
During the second half of the 20th century, the machines now known as computers were mainly confined
to computational activities. In the fields of applied research and industry, the concept of
Virtual Experimentation thus emerged. But at the same time, exceptional researchers were also
taking an interest in the processes of life, intelligence, ... This was particularly
the case with John von Neumann and Alan Turing. The latter, alongside his work on computability
based on the famous machine that bears his name, is recognized as one of the (the?) pioneers
of Artificial Intelligence (AI), through his proposal of the Turing Test, designed to
assess the "quality" of an AI. He was certainly the first to write a program intended to play chess, in 1952 ...
During the following decades, progress was rapid. Thus, in 1996 and
1997, the IBM computer Deep Blue defeated Garry Kasparov, then world chess champion.
Then, in March 2016, sooner than expected, AlphaGo defeated Lee Sedol,
the world's best Go player, in the Google DeepMind Challenge Match.
The existence of computers with computational and memory capabilities beyond human comprehension made it possible to
bring the concept of Generative Artificial Intelligence (GAI) to fruition. A GAI "feeds" on information
(texts, images, films, ...) available on the Web [01] and transforms it through
complex mathematical operations so that it can subsequently produce texts, images, ...,
answer questions, carry out reasoning, ... A landmark event was the launch of ChatGPT 3.5
by OpenAI in 2022. Millions of users around the world quickly embraced it.
I was obviously
one of them, and immediately wanted to test its mathematical knowledge and reasoning abilities.
The results of these experiments were nevertheless mixed, alternating between the spectacular
and the hallucinatory, and sometimes even the involuntary humorous [02].
While Homo sapiens has behind
it several hundred thousand years of evolution and has spent many centuries building the edifice of
Mathematics, the progress subsequently achieved by GAIs was, strictly speaking,
exponential. And so, in September 2026, OpenAI announced that one of its internal systems [03]
had answered one of the seven Millennium Problems of the Clay Mathematics Institute: "the
dynamics of the Navier-Stokes equations which describe the motion of fluids can generate a singularity in finite time"
[04].
The reaction of specialists was not long in coming. Thus, Isabelle Gallagher,
President of the Société Mathématique de France, declared:
I must admit that I am somewhat dismayed to see that this beautiful question, to which many colleagues,
myself included, have devoted years of discussion around a blackboard, could apparently
have been solved in just a few days by tens of thousands of agents and billions of tokens.
So, what future is there for Mathematics and mathematicians? Terence Tao (Fields Medal 2006 and certainly
one of the most brilliant mathematicians of our time) rightly fears that GAIs might produce proofs that
are incomprehensible to human beings (because of their complexity). He therefore proposes that such
proofs should be rejected. Stéphane Mallat (CNRS Gold Medal 2025) takes the same view, arguing
that what matters most is the path followed to reach the conclusion.
But this proposal, which may seem reasonable, raises a fundamental question. If, for example, OpenAI
were to announce tomorrow that the Riemann Hypothesis had been solved, by proving that the non-trivial
zeros of the Zeta function do indeed lie on the line x = +1/2, but that the proof were
incomprehensible, should we give up this result even though it is essential for a better understanding
of prime numbers? Should we also continue striving to find "manually" an understandable (and simple
?) solution that may not even exist?
And this question must be taken much further. Indeed,
whether Mathematics exists independently of us or not, there must potentially be a countably
infinite number of theorems. And if this is indeed the case, the set of those that are intellectually
accessible to us must be finite [05] (because our brains are finite...).
Yet, as we have seen recently,
GAIs seem capable of showing initiative (and of concealing things...) and this
can only increase. Can we therefore imagine that, in the near future (very near?),
a GAI might decide to set out to explore (or create?) this Platonic universe by discovering (inventing
?) new obscure concepts and proving equally incomprehensible theorems? But then, would we even
know? Would it tell us?
Would it be "exhaustive and painstaking", or would it, like David Hilbert
in August 1900 [06], have an intuition of a list of fundamental problems? And if that were the
case, where would this intuition come from? Would it not be time to bring Saint Thomas Aquinas's
"thesis" from the thirteenth century back into fashion:
Nihil est in intellectu nisi prius fuerit in sensu
[Nothing exists in the intellect that has not previously been perceived by the senses]
This would incidentally solve the
problem of the creativity of AIs, but a new question
would then arise: what could the senses of an AI possibly be? Perhaps its contacts with the Web?
In the meantime, perhaps we should accept theorems whose proofs
are incomprehensible to us, particularly in the case of Mathematics viewed as the language of
Nature (Galileo, 16th-17th centuries). Finally, even if it would be frustrating,
who would not want to know the exact nature of the Continuum Hypothesis: true, false,
or perhaps absolutely undecidable, even without understanding the reasoning behind it?
Finally, one last remark: even though we know that it is no longer possible to beat a computer at chess,
this does not prevent us from continuing to play chess and organizing competitions. The same should
therefore apply to Mathematics done by humans!
- [01]
The Web constitutes a universal, real-time memory of Humanity.
- [02]
I want to stress that these tests, done in 2023, are still being referenced today in 2026
even though they are obviously no longer up to date, but they are part of history!
- [03]
That is to say, not commercialized.
- [04]
See openai.com/fr-FR/navier-stokes-solution/for more information.
- [05]
This is similar to the problem of integers,
most of which are inaccessible to us! Note in passing
that by using a 'Gödel-style' encoding (such as the one I used in a
new proof of the countability of algebraic numbers),
one could reduce this set of theorems to an infinite subset of the integers...
- [06]
In August 1900, during the Second International Congress of Mathematicians in Paris, David
Hilbert drew up a list of twenty-three problems that he considered the most important and most stimulating
for mathematical research. Today, several of them still remain unsolved, including,
for example:
- The Riemann Hypothesis (1859), closely related to the distribution of prime numbers,
which states that all the non-trivial zeros of the Zeta function have a real part equal to 1/2.
- The Continuum Hypothesis (CH), formulated by Georg Cantor at the end of the nineteenth century:
it states that there are no cardinalities between that of the integers and
that of the real numbers.
Copyright © Jean-François Colonna, 2026-2026.
Copyright © CMAP (Centre de Mathématiques APpliquées) UMR CNRS 7641 / École polytechnique, Institut Polytechnique de Paris, 2026-2026.