Yi Ma at ICML: Intelligence as Science and Information Theory's Missed Opportunity

Recently, prominent scholar Yi Ma delivered a keynote at the ICML Foundation of Deep Generative Models Workshop, proposing that "intelligence should be treated as a scientific research subject" and might be even more important than physics. This sparked discussions on the role of information theory and the future of researchers in the AI era.

The Scientization of Intelligence and Information Theory's Missed Opportunity

Ma emphasized that studying intelligence requires physics-level rigor. He pointed out that the information theory community has long排斥ed engineering practices in intelligence, missing a huge opportunity to serve as the theoretical foundation for intelligence research. Related discussions noted that the academic community has largely ignored the connection between AI engineering and information theory over the past decade.

The Impact of AI Automation on Traditional Research

As AI technology evolves, future systems will be capable of autonomously running experiments, checking mathematical proofs, and synthesizing literature. This raises a profound question: what will happen to the work of traditional researchers? Ma believes that current engineering practices and scientific education systems face unprecedented challenges, and the role of researchers needs to be redefined.

The Criteria for AI to Become a True Science

Expanding on the idea of "intelligence as a science," researcher Kamilov pointed out that physics became a science because it makes bold predictions about reality. Similarly, the field of AI will only truly become a science when theories about "intelligence" can predict phenomena *before* models are even built.

2026-07-11 ~ 2026-07-12 · 5 related posts

1 near-duplicate retellings: YiMaTweets