AI, computation and scientific discovery: A Q&A with Karthik Duraisamy
The director of the Michigan Institute for Computational Discovery and Engineering discusses the institute’s past and future.
Michigan Engineering News
Experts
Karthik Duraisamy
Professor of Aerospace Engineering, Mechanical Engineering an Nuclear Engineering
Director of Michigan Institute for Computational Discovery and Engineering (MICDE)
In a little more than a decade, the Michigan Institute for Computational Discovery and Engineering (MICDE) has grown into a nationally important hub for exploring how computations and AI can be harnessed to answer bigger and bolder questions that can unlock major advances in science and engineering.
While large language models have gained attention for their ability to process text and images to respond to prompts, they represent just one type of foundation model. AI systems can be trained on diverse scientific data—from molecular structures to genomic sequences to experimental measurements and simulation outputs, and MICDE is leading national conversations in this space.
The Michigan Engineer sat down with MICDE’s director, Karthik Duraisamy, to learn more about how MICDE got to this point—and what’s next.
Duraisamy is a professor of aerospace engineering, mechanical engineering and nuclear engineering. He is the Samir and Puja Kaul Director of the Michigan Institute for Computational Discovery and Engineering. He also led U-M’s Generative Artificial Intelligence Advisory Committee and played an important role in recent campus-wide activities, including major collaborations with Los Alamos National Laboratory (LANL) and discussions around high performance computing (HPC) and AI.
Let’s start with the big picture. What is MICDE trying to accomplish?
At the heart of our vision is advancing the foundational science of computation. Over the past half century, computation has been playing an increasingly significant role in science and engineering. Our mission is to pioneer new paradigms and applications of computational science and artificial intelligence to enable scientific discoveries and engineering innovations to solve societal challenges. We envision a future in which computations provide direct answers to the most pressing questions of our time, transforming computations into both a source of new information and a generator of knowledge.
How is MICDE advancing computational science and AI toward that future?
Fundamentally, our job as an institute is to broaden the impact of computations on science, engineering and society, and ensure that the University of Michigan—with its unparalleled combination of depth and breadth—is at the forefront of these advances.
One of our key strengths is in identifying emerging trends and bringing together faculty from diverse disciplines to launch new initiatives, often before these areas become mainstream. MICDE achieves this by nurturing an environment for interdisciplinary research and education, while ensuring our researchers have the resources and connections they need to make breakthrough advances. We advance these goals through strategic initiatives, resources and our educational programs.
As an example, we provide seed funding through catalyst grants to prime promising ideas and organize seminars, workshops and networking events that foster collaboration. A key focus is connecting domain experts in fields like biology, materials science, and medicine with computational scientists to tackle complex challenges. We create innovation hubs where we match significant scientific questions with emerging computational tools and methods that could help answer them.
With the emergence of generative AI, for instance, we started an initiative to build expertise in foundation models and AI agents. Last year, we held the first ever conference on scientific foundation models and brought together leading experts in the field to broaden the horizons of generative AI in scientific applications. We organized a summer school on the same topic. We are also playing a role in national discussions on AI for science, particularly in collaboration with Los Alamos National Laboratory.
And, of course, education is central to MICDE’s mission. We’ve significantly expanded the PhD program in scientific computing. The program isn’t new—it started in 1988 as the first such program in the nation, initiated by the college of engineering’s Laboratory for Scientific Computation, which was the predecessor to MICDE. We also run certificate programs in computational neuroscience and computational discovery and engineering, and we are fostering a vibrant community of students via the scientific computing student club.
You played a role in MICDE’s launch, as a new assistant professor. How has the institute evolved?
I would like to emphasize that we had a good starting point—scientific computing has had a rich history at the University of Michigan, going as far back as the 1940s. We have always had illustrious faculty members and exceptional graduates in this field, and for a time, U-M played a leading role in the evolution of high performance computing in our country.
The first few years of MICDE under the leadership of Krishna Garikipati were about getting organized and building a community—growing a big tent so that faculty and students specializing in computational science and engineering could come together, collaborate and access shared resources.
In this period, the scope of our research and education grew rapidly. We started several new centers anchored around federally funded research, and in just a few years, our PhD program grew from roughly 15 students—mostly in engineering—to the present 160, spanning 37 departments from 8 schools across the university.
As we built critical mass and got better organized and connected, we have become more strategic in research directions and in deployment of resources. For our next chapter, we are intensely focused on new developments in areas such as scientific machine learning, quantum computing, formal verification, and expanding to disciplines such as biology and medicine. We are also strengthening our relationships with our national lab partners, as highlighted by a profound new partnership with Los Alamos National Laboratory.
MICDE has been around for a little more than twelve years now. What are some of its greatest accomplishments?
The heart of our success lies in our people and the collaborative culture we’ve built. Our most significant achievement has been assembling an exceptional community of researchers, catalyzing innovative partnerships across disciplines, and developing the next generation of computational leaders. It has been particularly rewarding to witness the remarkable growth and success of our junior faculty members and in seeing our students develop into accomplished researchers.
One of our strengths over the past decade has been at the intersection of traditional modeling and machine learning. As far back as 2015, with support from the National Science Foundation, we established a hardware and software ecosystem called Conflux to support data-enabled modeling of complex physical problems. More than 100 researchers from different parts of the campus as well as from other universities used this resource and great research came out of it. A number of students have been exposed to the interface between HPC and machine learning, and it contributed to our leadership in AI for science. Our faculty and students continue to shape the frontiers of this most important sub-discipline. We have also been increasing our entrepreneurial footprint.
One of our recent highlights was when Vikram Gavini won the Gordon Bell Prize in 2023, awarded by the Association for Computing Machinery for bringing quantum mechanical accuracy to large-scale materials simulations. This is the most prestigious prize in high performance computing, and a well-deserved reward for his rigorous and intensely focused research.
MICDE-affiliated researchers have been awarded several research centers and infrastructure grants. Just in the past two years, examples include a Space Weather Center of Excellence funded by NASA; a Science and Technology Center for nanotechnology funded by the National Science Foundation; an advanced algorithms and co-design center funded by the Department of Energy (DOE) and no less than three Multidisciplinary University Research Initiatives funded by the Department of Defense.
Recruiting talent and retaining leaders in the field is another of our strengths, and an area in which we’d like to do even better, as there is intense competition for high-end talent. At present, three of the College of Engineering’s department chairs are leaders in computational engineering, and that speaks to both the excellence of our computational talent and the way that our research is valued at U-M. We are also thrilled to have attracted a number of highly successful mid-career faculty members who are emerging as leaders in their domains.
MICDE has also been a big part of defining U-M’s generative AI strategy from its advent. On the science side, we are developing foundation models and AI agents that could help us develop better medicines, clean energy technologies, structural materials and more.
Looking ahead, what are you most excited about in MICDE’s future?
MICDE’s core competency is in the development of methods, frameworks, approaches, and models to enable scientific inquiry, and so our work in fundamental computational science will continue to define MICDE’s trajectory and impact in the years ahead.
That said, we stand at an incredibly exciting time for the scientific enterprise: Within the next few years, appropriately designed systems powered by recent advances in computational science and AI have the potential to significantly accelerate progress in science and technology. Realizing this potential is one of our main focus areas, and this requires strong partnerships with national laboratories, industry and academia.
Closely—yet not exclusively—related to this aspiration is the Los Alamos collaboration. The Strategic Partnership and Accelerated Research Collaboration (SPARC) is forging research collaborations between co-located LANL and University of Michigan scientists to enable transformative advances in science, engineering, and national security.
Along these lines, we are planning our second conference on foundation models and AI agents for science, to be held May 28-30 and we have invited a stellar cast of thought leaders and practitioners in this field, including from all the major AI companies.
We have been really fortunate to have robust support from our college and the university, and from Samir and Puja Kaul’s generous endowment. Looking ahead, I am always inspired by our exceptional community of scholars as we continue to expand the depth and breadth of our activities. MICDE is strategically strengthening our collaborations with colleagues across the biological sciences, medical school, school of information, and school of public health. These partnerships will begin to bear fruit in the coming years, and I can’t be more excited to see the resulting impact.