Much has transformed over the past 15 years since Kaiming He embarked on his PhD journey. Reflecting on this period, he remarks, “During my PhD years, there were significant barriers separating various disciplines — even within computer science. The person sitting next to me could be engaged in work that was completely beyond my understanding.”
Upon joining the MIT Schwarzman College of Computing as the Douglas Ross (1954) Career Development Professor of Software Technology in the Department of Electrical Engineering and Computer Science, He has observed what he describes as “a historical reduction of barriers between scientific fields.” He believes that this phenomenon is catalyzed by the emergence of a shared language in the realm of artificial intelligence (AI).
“I might never grasp the intricacies of high-energy physics, chemistry, or cutting-edge biology, but we are now seeing a breakthrough — a common language emerging from advancements in AI,” he asserts.
Establishing Connections Through AI
This transformation can be traced back to 2012, marked by the onset of the “deep learning revolution.” It was during this period that researchers recognized the immense potential of machine-learning techniques grounded in neural networks.
He explains, “The field of computer vision — equipping machines to interpret the world as humans do — saw explosive growth. The realization that these neural methodologies could be widely applicable across diverse challenges led to the expansion of the computer vision community.”
As a result, the influence of these AI methods has permeated various domains within computer science, encompassing natural language processing, speech recognition, and robotics. This synergy has laid the groundwork for innovations like ChatGPT and advancements toward artificial general intelligence (AGI).
He highlights the impressive capabilities of AlphaFold, an AI tool crafted by Google DeepMind, which accurately predicts protein structures. “Though it addresses a vastly different scientific challenge, it utilizes the same AI frameworks. And this is just the tip of the iceberg,” he posits.
The Intersection of AI and Scientific Research
Since arriving at MIT in February 2024, He has engaged in discussions with professors across a broad spectrum of disciplines. He frequently finds himself collaborating with scholars who bring diverse knowledge to the table.
“While I may not entirely grasp the nuances of their specialization, the introductory context they provide enables us to delve into conversations about deep learning, machine learning, and neural networks as applicable to their research challenges,” He states. “These AI tools serve as a bridge, translating complex terminology into accessible concepts, facilitating an exchange of insights and collaborative problem-solving.”
Incorporating AI across various scientific arenas holds tremendous promise — from employing video analytics to foresee weather patterns to streamlining research processes and cutting the costs associated with drug discovery.
Not only do these AI technologies benefit He’s colleagues in science, but they also feedback into the AI evolution. “Researchers introduce fresh challenges that continue to propel the evolution of AI tools,” He explains. “Moreover, many contemporary AI techniques were inspired by earlier scientific studies — for instance, artificial neural networks drew from biological concepts, while diffusion models for image generation emerged from principles in physics.”
“There is an inherent connection between science and AI. We have approached similar goals from disparate angles, and now, those paths are converging,” he asserts.
And what better incubator for this collaboration than MIT? He notes, “It’s not surprising that MIT is at the forefront of this change. The MIT Schwarzman College of Computing fosters an environment where diverse minds come together to collaborate, share ideas, and communicate effectively in a common language.”
As for the timeline for fully dismantling existing barriers, He emphasizes this as an ongoing endeavor. “Decades back, computers were seen as advanced technology requiring specialized knowledge. Now, they are ubiquitous. I anticipate that in 10 years or more, everyone will be leveraging some form of AI in their research — it will become an integral tool in their arsenal,” he predicts.
Photo credit & article inspired by: Massachusetts Institute of Technology
