A lot has changed in the 15 years that Kaiming has been a doctoral student.
“When you’re in a PhD, there’s a high wall between different fields and subjects, and there’s an even higher wall in computer science,” he says. “The guy sitting next to me might be doing things I don’t fully understand.”
Seven months into his tenure at MIT’s Schwartzman College of Computing as the Douglas Ross (1954) Professor of Professional Development in Software Technologies in the Department of Electrical Engineering and Computer Science, he says he is experiencing something he considers “very rare in the history of human science”—the breaking down of the walls that span across disciplines.
“There is no way to understand high-energy physics, chemistry, or the frontiers of biological research,” he says, “but now we see something that can help us break down those walls, and that is the creation of a common language, which is found in artificial intelligence.”
Building the artificial intelligence bridge
According to him, this shift began in 2012 as a result of the “deep learning revolution,” the point at which this set of machine learning methods based on neural networks was found to be so powerful that it could be more widely used.
“At this point, computer vision—helping computers see and understand the world as if they were humans—was growing rapidly because it turned out you could apply the same method to many different problems and in many different domains,” he says. So the computer vision community really grew quickly because these different sub-topics could now speak a common language and share a common toolset.
From there, he says, the trend began to spread to other areas of computer science, including natural language processing, speech recognition, and robotics, laying the foundation for ChatGPT and other advances toward artificial general intelligence (AGI).
“All of this has happened in the last decade, and it has led us to an emerging trend that I’m really excited about, which is seeing AI methodology driving other scientific disciplines,” he says.
One of the most famous examples, he says, is AlphaFold, an artificial intelligence program developed by Google DeepMind that predicts the structure of proteins.
“It’s a very different scientific discipline, a very different problem, but people are also using the same AI toolset, the same methodology to solve these problems, and I think this is just the beginning,” he says.
The future of artificial intelligence in science
Since arriving at MIT in February 2024, he says he has spoken to professors in almost every department. Sometimes he finds himself in a conversation with two or more professors from very different backgrounds.
“I certainly don’t fully understand their research area, but they just provide context and then we can talk about deep learning, machine learning [and] neural network models in their problems,” he says. In that sense, these AI tools are like a common language between these scientific disciplines: the machine learning tools translate their terms and concepts into terms I understand, and then I can learn from their problems and share my experiences and sometimes suggest solutions or opportunities for them to explore.”
Expansion into various scientific fields has significant potential, from using video analytics to predict weather and climate trends to accelerating research cycles and reducing costs associated with discovering new drugs.
While AI tools provide obvious benefits to the work of his scientific colleagues, he also points to the mutual influence they can have on the creation and development of AI.
“Scientists present new problems and challenges that help us continue to develop these tools,” he says. But it’s also important to remember that many of today’s AI tools come from earlier scientific disciplines—for example, artificial neural networks were inspired by biological observations. Diffusion models were motivated to create a picture of a physical concept.
“Science and AI are not separate subjects. We approached the same goal from different perspectives, and now we are coming together.”
And where better to meet than at MIT?
“It’s no surprise that MIT was able to see this change earlier than many other places,” he says. [The MIT Schwarzman College of Computing] has created an environment that brings together different people and allows them to sit together, talk together, collaborate, exchange ideas, and all speak the same language—and I see that happening.
As for when the walls will completely come down, he notes that it’s a long-term investment that won’t happen overnight.
“Decades ago, computers were considered high-tech and we needed specialized knowledge to understand them, but now everyone uses a computer,” he says. “I expect that in about 10 years, everyone will be using some form of AI for their research – it’s just basic tools, their basic language, and they can use AI to solve their problems.”
