Machine learning: science and technology.
Q: What are the main reporting topics for last year’s meeting of leaders in mathematics and physical sciences?
AND: It was enlightening to gather so many researchers at the forefront of artificial intelligence and science in one room. Although workshop participants came from five different scientific backgrounds—astronomy, chemistry, materials science, mathematics, and physics—we found many similarities in the way each of us engages with AI. From our lively discussions, a real consensus emerged: coordinated investments in computing and data infrastructure, interdisciplinary research techniques, and strict training can significantly advance both AI and science.
One of the main observations was that it had to be a two-way street. It’s not just about using AI to improve learning; science can also improve artificial intelligence. Scientists excel at extracting insights from complicated systems, including neural networks, by discovering underlying principles and emergent behaviors. We call this “AI science,” and it comes in three flavors: science-driven AI, in which scientific reasoning shapes the fundamental approaches to AI; science-inspired artificial intelligence, where scientific challenges drive the development of recent algorithms; and the science of explaining artificial intelligence, where scientific tools lend a hand explain how machine intelligence actually works.
For example, in my field of particle physics, researchers are developing real-time artificial intelligence algorithms to cope with the deluge of data from collider experiments. This work has direct implications for discovering recent physics, but the algorithms themselves are proving valuable outside our field as well. The workshop explained that AI science should be a priority for the community – it has the potential to change the way we understand, develop and control AI systems.
Of course, combining science and artificial intelligence requires people who can work in both worlds. Participants consistently emphasized the need for “centaur scientists” – researchers with true interdisciplinary knowledge. Supporting these policymakers at every stage of their careers, from integrated bachelor’s degrees to interdisciplinary PhD programs and collaborative faculty hiring, has proven imperative.
Q: How do MIT’s efforts on AI and science align with the workshop’s recommendations?
AND: During the workshop, recommendations were based on three pillars: research, talent and community. As director NSF Institute for Artificial Intelligence and Fundamental Interactions (IAIFI) — a collaborative effort in artificial intelligence and physics between MIT, Harvard, Northeastern, and Tufts universities — I have seen firsthand how effective this framework can be. By scaling this to the MIT level, we can see where progress has been made and where opportunities lie.
On the research side, MIT already enables work on artificial intelligence and science in both fields. Even a quick scroll shows how individual researchers in the Faculty of Science are implementing AI-based projects, building a pipeline of knowledge and discovering recent opportunities. At the same time, collaborative efforts such as IAIFI and Institute for Accelerated AI Algorithms for Data-Driven Discovery (A3D3). focus interdisciplinary energy for greater impact. The MIT Generative Impact AI Consortium also supports work on application-based artificial intelligence at university scale.
To support newborn talent in artificial intelligence and science, several initiatives are training the next generation of centaur scientists. MIT Schwarzman College of Computing’s Common Ground for Computing Education helps students become “bilingual” in computer science and home disciplines. Interdisciplinary PhD tracks are also growing in popularity; IAIFI collaborated with MIT Institute for Data, Systems and Society create one in physics, statistics, and data science, and currently about 10 percent of physics Ph.D. students choose to do so – and that number is likely to raise. Dedicated postdoctoral roles such as IAIFI Scholarship AND Tayebati Association provide early-career researchers with the freedom to conduct interdisciplinary work. Funding centaur scientists and providing them with space to build connections across fields, universities and career stages has been transformative.
Finally, community building brings it all together. From specialized workshops to gigantic symposia, organizing interdisciplinary events signals that artificial intelligence and science is not a work in isolation – it is an emerging field. MIT has the talent and resources to make a significant impact, and hosting events like this at various scales helps solidify that leadership.
Q: What lessons can MIT learn about continuing to improve its AI and science efforts?
AND: Something vital crystallized during the workshop: the institutions that lead in AI and science will be the ones that think systematically, not piecemeal. Resources are confined, so priorities matter. Workshop participants clearly identified what becomes possible when an institution coordinates staffing, research and training based on a coherent strategy.
MIT is well-positioned to build on more structural initiatives already underway – shared department lines in computer science and science, expanded interdisciplinary degree pathways, and targeted funding for “artificial intelligence science.” We are already seeing moves in this direction; this year, the MIT Schwarzman College of Computing and the Department of Physics are conducting their first-ever joint faculty search, which is invigorating.
The virtuous cycle of AI and science can truly become one another – offering deeper insights into AI, accelerating scientific discovery, and creating hearty tools for both. By developing an intentional strategy, MIT will be well-positioned to lead and benefit from the coming waves of artificial intelligence.