Can a 400-year-old dream solve the modern AI hallucination crisis?
Generative AI faces a fundamental truth problem: its tendency to hallucinate makes it unreliable for science. This video explores how computer scientist Tudor Achim proposes using a centuries-old logical framework to ground AI in rigorous, error-free verification.
At TEDAI San Francisco 2025, Tudor Achim presented a vision for mathematical superintelligence. He argues that the current issue of AI hallucinations creates a truth problem that scientific progress cannot afford. To solve this, Achim looks back to the 17th-century dream of Gottfried Wilhelm Leibniz: a logical framework designed to make errors impossible.
By integrating formal verification into AI, we can move beyond unreliable chatbots. The goal is to transform these models into rigorous partners capable of driving genuine scientific discovery through a system where mistakes are structurally prevented.
Source: The Path to Mathematical Superintelligence | Tudor Achim | TED
