The Future of Mathematics: A Golden Era or a Digital Eclipse?

Originally a guest contribution by Francesco Maggi, hosted on Terence Tao's blog.

---

Large Language Models (LLMs) and the broader ecosystem of "Artificial Intelligence" have already shown a startling capacity for generating mathematical proofs. While we are still mapping the boundaries and flaws of these systems, their potential to catalyze discovery and drive leaps in science and technology is undeniable.

However, the trajectory we take depends on how we integrate these tools. We are standing at a crossroads between two distinct futures:

The Divergent Paths of Progress

---

1. LLMs as Research Instruments

It is an objective fact: LLMs are now potent tools for research. We have seen them tackle monumental tasks, such as:

• Constructing a finite-time blow-up solution for the $\text{Navier-Stokes}$ equations (with a forcing term).
• Proposing proofs for the quasi-Riemann Hypothesis.

These achievements demand rigorous, independent verification and assimilation by the global mathematical community.

The "Translation" Gap

To ensure absolute correctness, researchers use Lean—a formal proof assistant that verifies arguments against strict axioms. However, a troubling trend has emerged:

As the Lean version of a proof grows in length (sometimes reaching $\approx 10^5$ lines of code), the LLM's ability to translate that formal logic back into "natural language" degrades.

This creates a paradoxical situation:

1. The Formal Proof: Correct, but too massive for a human to read.
2. The Natural Language Summary: Readable, but messy and potentially unfaithful to the logic.

This leads to a profound philosophical crisis: If a machine provides a proof that no human can understand, has mathematics actually progressed? Without human comprehension, these proofs are merely benchmarks for AI, not advancements in human knowledge.

The Biological Root of Math

Mathematics is not just a set of rules; it is deeply entwined with human cognition:

• Identification $\rightarrow$ Counting $\rightarrow$ Numbers.
• Spatial/Temporal Awareness $\rightarrow$ Measurement $\rightarrow$ Geometry.

Because different cultures developed different mathematical frameworks, and because the field expands often by accident, mathematics is a living part of human culture. Therefore, the claim that "math is over" because machines can prove theorems is false. If we wish to avoid a future where knowledge is "alien," we actually need more mathematicians to bridge the gap between machine output and human insight.

---

2. Skepticism and the Mirage of Creativity

The current euphoria surrounding AI's mathematical prowess often lacks the "cool-headedness" required by the scientific method. We must be careful not to mistake performance for essence.

Critical Distinctions:

• Performance $\neq$ Intelligence.
• Pattern Generation $\neq$ Creativity.
• Output $\neq$ Consciousness.

These concepts are complex scientific subjects that require:

• Explicit, rigorous definitions.
• Testable, empirical hypotheses.
• Verifiable evidence.

We must resist the urge to view social transformations as "inevitable." The impact of AI will not be decided by the software's capabilities alone, but by the institutional choices we make today.

Visualizing the Human Element

---

Final Thought: The transition toward either a golden or dark age is not a predetermined path, but a result of our current scientific and po... (Text ends abruptly)