Geoffrey Hinton, a leading figure in artificial intelligence, has made a significant prediction: machines will surpass human capabilities in mathematics within the next decade. This is an audacious claim regarding a discipline fundamental to human achievement for centuries, but Hinton tends to speak with conviction.
In his recent address at the Sana AI Summit, Hinton presented an argument that appears straightforward at first glance. Mathematics is a closed system, akin to strategic games like chess and Go, which also have strict rules and clear objectives. The successes of AI in these games suggest that mathematics may operate on similar principles.
When algorithms are provided with a set of axioms, they can autonomously explore conjectures and proofs. Unlike human mathematicians, AI systems can enhance their skills by engaging in self-directed learning. This is analogous to AlphaZero mastering chess through countless self-play games; AI can navigate complex mathematical proof landscapes through pattern recognition and brute-force exploration.
This prediction is all the more striking considering Hinton's earlier views. Previously, he estimated that artificial general intelligence, where AI matches or exceeds human cognitive abilities across various fields, was far off. However, he now assesses a 50% probability of AGI manifesting within the next two decades. Such odds concerning a breakthrough that could transform human civilization understandably induce a degree of apprehension.
Hinton's concerns over AI safety led him to resign from Google in 2023, a choice that stood out because of his profound contributions to the field, including the neural network technologies that underlie current AI systems. His departure from the tech giant served as both a personal and public declaration regarding the potential existential risks posed by advanced AI.
In recognition of his groundbreaking work, Hinton received the Nobel Prize in Physics in 2024 alongside John Hopfield, reiterating the foundational nature of his contributions. During the summit, he expressed the belief that superintelligent systems could emerge during his lifetime. This marks a significant shift from his past positions and highlights a growing trend among leading AI researchers as the timeline for achieving AGI continues to draw closer.
The rapid advancements in technology challenge researchers, not because they are becoming overly optimistic, but rather due to the unexpected pace of AI development.