The Impending Phase Change: AI's Unpredictable Remaking of Mathematics

Ravi Vakil, a distinguished algebraic geometer at Stanford and president of the American Mathematical Society, posits that artificial intelligence will not merely augment mathematics but instigate a profound "phase change." Speaking with Daria Ivanova, Vakil shared his nuanced perspective on AI's current capabilities, its potential for true creativity, and the inevitable, yet unpredictable, transformation it promises for the field of mathematics. His insights, rooted in a deep understanding of mathematical discovery and historical technological shifts, offer a compelling vision for founders, VCs, and AI professionals grappling with the future of intelligent systems.

Vakil views AI today as akin to "a bright but confused undergraduate"—eager, capable, and on the precipice of genuine insight, yet still lacking the profound leaps of understanding that define advanced mathematical thought. He awaits a "wow" moment, an instance where AI produces something mathematically surprising and cool, an unexpected discovery that transcends mere computation or the systematic exploration of known possibilities. This echoes the "AlphaGo moment," where DeepMind's AI, given only the rules of Go, developed strategies that astonished human grandmasters, demonstrating a creativity previously thought exclusive to humans.

The core of mathematical value, Vakil contends, is not merely finding proofs but achieving "understanding." Mathematics, perhaps more than any other discipline, thrives on epiphanies—those sudden bursts of insight that illuminate complex structures and reveal elegant solutions. This emphasis on understanding highlights a current frontier for AI: while machines excel at logical deduction and processing vast datasets, the intuitive leap, the recognition of deep connections, and the formulation of novel theories remain largely human domains. The challenge, then, lies in crafting problems that test this very human ingenuity, distinguishing between what is merely computationally complex and what demands genuine conceptual innovation.

Vakil emphasizes the historical unpredictability of technological advancements. "Whenever in the past people have tried to predict the effect of technological advances, they’ve always been wrong," he states, suggesting that AI's impact will be "even more wrong" than previous predictions. The internet, for instance, introduced unforeseen shifts like social media, reshaping human interaction in ways no one could have fully anticipated. Mathematics, too, has consistently accelerated with new tools, from LaTeX to personal computing. AI, Vakil believes, will continue this trend, not by diminishing human mathematicians but by providing new lenses and capabilities, fundamentally altering how research is conducted.

This transformation will not necessarily be a direct competition but a new "game with imperfect information" between humans and AI. Mathematicians will design problems specifically to probe the limits of AI, and in doing so, they will gain deeper insights into the nature of human intelligence itself. Certain types of problems, particularly those requiring "advanced mathematical creative thinking," currently pose significant obstacles for AI. These are the areas where human intuition and the ability to generate entirely new theoretical frameworks still hold an advantage.

Related Reading

• Engineering AI Prompts: Google's Framework for Benchmarking and Automation
• English is the New Machine Code: Replit's Vision for AI-Driven Software Development
• Karpathy's Brutal LLM Assessment Ignites AI Progress Debate

The integration of AI into mathematical practice will likely evolve over generations. Current researchers, like Vakil, may learn to utilize AI as a "second language," a powerful tool to enhance their existing methodologies. However, future mathematicians, born into a world where AI is ubiquitous, will likely speak it as a "first language," seamlessly integrating AI-driven insights and methods into their fundamental approach to problem-solving. This native fluency could unlock entirely new avenues of mathematical exploration and discovery.

Vakil cautions against the short-sightedness of redirecting funding from general scientific research to narrowly focused AI development. Such a move, he argues, is "short-sighted and stupid," as the broader scientific ecosystem ultimately feeds the very innovations that drive AI. Instead, he advocates for a balanced approach that recognizes the symbiotic relationship between foundational research and technological advancement. The true measure of AI's success in mathematics won't be its ability to solve existing problems faster, but its capacity to generate surprising, cool, and genuinely novel mathematical ideas—connections that humans would not have conceived on their own. This represents the next frontier, a phase change where AI becomes a source of true mathematical wonder.