Harnessing Human-in-the-Loop AI: Insights from Terence Tao on Mathematics

An insightful interview featuring Terence Tao, the Australian mathematician and Fields Medal laureate, conducted by Matteo Wong of The Atlantic, has been published in The Atlantic. You can read the full interview here.

In the interview, Professor Tao shared his perspective on the role of artificial intelligence in mathematics. He noted that while AI may not be a creative collaborator in its own right, it has the potential to act as a catalyst for mathematicians’ hypotheses and methodologies, paving the way for a new era of ‘industrial-scale mathematics’ that was previously unattainable.

While not AI, tools such as proof assistants have already transformed how mathematicians approach proofs, allowing for collaborative theorem proving at an unprecedented scale. However, these proof assistants require specialized language to express proofs effectively. Professor Tao proposes leveraging AI to translate these proofs into everyday language, which can then be converted back into the specific language required by proof systems. This approach could serve as an interface, effectively bridging the gap between human mathematicians and technological systems.

As Tao articulates:

“I am not very interested in replicating what humans already do well because it seems inefficient. I believe that at the forefront, both humans and AI are always needed due to their complementary strengths. AI is very skilled at converting billions of data points into a single answer. Humans excel at making truly inspired guesses from ten observations.” (Please note that this is a translated excerpt from a Korean article and may not be a word-for-word translation.)

In this light, I believe Professor Tao recognizes a parallel between mathematics and advancements in artificial intelligence, which aligns with my previous observations regarding the intersection of chemistry and AI. It highlights what experts across different fields must do to facilitate further progress in AI. Central to this discussion is the importance of language. It involves translating the specialized terminology used by leading experts, which is often obscure and difficult for most to grasp, into a more accessible language that ordinary individuals can understand. This accessibility is crucial for integrating current AI technologies into automated workflows.

In scenarios where training on vast amounts of data is impractical, including drug discovery, chemical reactions, and natural product chemistry, it becomes essential to connect the unique language crafted by top experts who often draw inspiration from a limited number of cases with established AI technologies.

This connection underscores my thoughts on human-in-the-loop AI, emphasizing the need for collaboration between human expertise and artificial intelligence.