Daniel Litt: The Mathematician's Guide to AI

The rapid advancements in artificial intelligence are redefining how we approach complex problem-solving, opening doors to efficiencies and insights previously unattainable across various fields. This discussion with Daniel Litt, a mathematician at the University of Toronto, highlights that while current AI models excel at executing calculations, synthesizing known techniques, and exploring vast solution spaces, they still fall short in areas like developing intuition, formulating novel theories, or discerning the most impactful questions. The core argument is that AI can autonomously solve intricate mathematical problems, such as a recent breakthrough with the Erdős unit distance problem, yet it struggles with the broader, more abstract reasoning that underpins human mathematical progress. For working professionals, this distinction is critical for strategically leveraging AI. Consider a small e-commerce shop in Austin, Texas, struggling to optimize its supply chain: instead of trying to make AI invent entirely new logistical paradigms, they could feed it historical sales data, shipping routes, and inventory levels. The AI could then rapidly simulate and identify the most efficient combinations of existing suppliers and delivery services, saving thousands in operational costs by optimizing known variables. Similarly, an indie SaaS founder in Seattle building a niche analytical tool could use AI to automatically generate and test variations of statistical models against a given dataset, pinpointing the most robust existing approach without needing the AI to *invent* a new statistical method from scratch. Even a hospital administration team in Chicago, aiming to streamline patient flow, could deploy AI to analyze existing appointment schedules, staff allocations, and facility layouts, identifying bottlenecks and suggesting optimal reconfigurations based on established hospital management principles, rather than expecting it to redesign healthcare on a fundamental level. The practical impact is clear: AI is a powerful enhancer of existing methodologies and a formidable tool for exploring permutations within defined constraints, but it isn't yet a replacement for human creativity, intuition, or the ability to identify truly novel problems. To capitalize on this, developers can focus on building systems where AI augments human expertise by handling the intensive computational and combinatorial aspects of a problem. Founders and operators should direct AI towards optimizing existing processes and extracting insights from vast datasets that align with established operational frameworks. This week, try this experiment: pick a routine, complex decision in your work that involves combining multiple known factors or analyzing large sets of existing data. Instead of aiming for a revolutionary AI solution, build a small prototype where AI helps you systematically explore the *known* solution space or identify the optimal combination of *existing* resources. For example, if you manage project schedules, use an AI tool to suggest the best sequence of tasks given known dependencies and resource availability, rather than expecting it to invent a new project management methodology.