Will it be a golden age or a dark age for mathematics (and science)?
Rapid advances in large language models are forcing mathematicians and scientists to ask whether AI will usher in a golden age of discovery or a dark age of dependence and exclusion. Commenters weigh fears that machines will flood the field with opaque proofs, deskill human experts, and concentrate power in private labs against hopes that AI will accelerate research, democratize access to advanced ideas, and free humans to focus on understanding rather than brute-force problem solving. Many argue the outcome will hinge less on raw capability than on governance: whether AI tools and infrastructure remain open, distributed, and aligned with broad human education, or become proprietary engines driving a narrow, profit-driven research agenda.
Meta: Blog and Authorship
- Several comments focus on the blog itself: people miss original posts by the host and feel guest essays repeat similar themes.
- Some would prefer fewer guest posts or a separate venue for opinion pieces.
- Others defend the blog’s recent AI-focused leadership role, arguing the field needs coordination and someone to catalyze discussion.
Golden Age vs Dark Age for Mathematics
- Optimistic view: AI will trigger a golden age by accelerating calculations, proofs, and cross-referencing, enabling new theories and deeper understanding.
- Pessimistic view: AI may produce an opaque flood of results that few humans can understand, concentrating knowledge and power, and potentially leading to a “dark age” for human mathematics even if formal progress accelerates.
- Many expect a mixed outcome: pockets of golden age and dark age depending on institutions, incentives, and openness.
Role and Limits of LLMs in Math
- Key worry: AI can prove but not teach. Machine-generated proofs (especially Lean formalizations) are often huge and hard to understand; natural-language explanations are viewed as low-quality or messy.
- Some argue math is fundamentally about human understanding, not just solved conjectures; if only machines understand, “human mathematics” shrinks.
- Others counter that faster theorem-solving still increases the need for humans to interpret, unify, and turn “slop” into concepts and textbooks.
Control, Openness, and Infrastructure
- Strong concern over private control of frontier models: big labs could effectively become arbiters of which problems get solved, hoard capabilities, and use researchers’ prompts as training data.
- Thread debates whether building independent, academic LLM infrastructure is essential or just self-interested rent-seeking.
- Open vs closed models is a recurring theme; distributed local models and open science are seen as more democratic.
Careers, Incentives, and Gatekeeping
- Disagreement over whether AI increases or decreases demand for mathematicians: some foresee more roles focused on understanding, others see straightforward displacement and fewer jobs.
- Comparisons to software engineering: AI reduces need for juniors, risks “black box” systems no one fully understands, but also enables individuals to do work that once took large teams.
- Some vent frustration at perceived past elitism in math; others argue math has been unusually meritocratic with objective standards.
Broader Societal & Economic Concerns
- AI is framed as part of a larger shift affecting all knowledge work, not just math.
- Concerns include permanent underclasses, elite capture of AI-driven advances, IP and patent regimes around AI-generated discoveries, and erosion of human expertise if people rely too heavily on automated tools.