In the rapidly evolving landscape of artificial intelligence, a fundamental question has emerged: Can machines transcend the role of automated assistants to become active participants in the creative scientific process? This inquiry sits at the heart of an upcoming dual-session presentation featuring two of the field’s most prominent figures: Professor George Karniadakis and PhD candidate Nazanin Ahmadi.
Together, they will explore the convergence of Scientific Machine Learning (SciML) and agentic AI—a multidisciplinary frontier that promises to reshape how we formulate hypotheses, validate models, and uncover the fundamental laws governing complex systems.
The Symposium Schedule: A Morning of Innovation
The event is structured to provide both a theoretical foundation and a practical demonstration of these cutting-edge methodologies.
- 9:00 a.m. – 9:45 a.m.: George Karniadakis (Brown University) will deliver an opening talk exploring the mathematical and physical underpinnings of agentic scientific discovery.
- 10:00 a.m. – 10:50 a.m.: Nazanin Ahmadi (Brown University) will follow with a session detailing the practical implementation of these frameworks in systems biology and pharmacometrics.
The Convergence of Physics and Machine Learning
The core thesis of the presentation centers on the limitation of current AI models. While large language models and standard neural networks excel at pattern recognition, they often lack the "grounding" required for rigorous scientific discovery. Karniadakis and Ahmadi argue that true scientific creativity does not emerge in a vacuum; it is the product of a delicate interplay between data, mathematical structures, and physical laws.
Bridging the Gap: The SciML Framework
Scientific Machine Learning (SciML) serves as the engine for this new paradigm. By integrating physics-informed neural networks (PINNs) and neural operators, researchers can ensure that AI-generated models respect the conservation laws and physical constraints inherent in the natural world. Unlike "black-box" AI models that might produce statistically accurate but physically impossible results, SciML frameworks provide a bridge, ensuring that machine learning outputs are interpretable and scientifically valid.
The Agentic Turn
The speakers will introduce an "agentic framework" designed to mirror the collaborative nature of human research teams. Rather than relying on a single, monolithic model, this framework deploys a network of specialized agents. These agents are tasked with specific roles—proposing hypotheses, critiquing assumptions, running computational validations, and self-correcting through iterative feedback. This multi-agent approach mimics the social structure of a laboratory, where diverse perspectives are required to refine complex theories and identify genuine anomalies in data.
Biography: George Karniadakis
George Karniadakis is a titan in the field of computational mathematics. An elected member of the National Academy of Engineering and the National Academy of Arts and Sciences, his career reflects a lifelong dedication to pushing the boundaries of numerical analysis and fluid mechanics.
His academic journey, which includes advanced degrees from MIT and appointments at Princeton, Caltech, and Brown, is marked by an unparalleled citation record—over 184,000 citations and an h-index of 171. In 2025, he was recognized by Research.com as the preeminent mathematical scientist in the world. His accolades include the William Benter Prize, the SIAM/ACM Prize on Computational Science & Engineering, and the Alexander von Humboldt award, among others. As a Vannevar Bush Faculty Fellow, his work continues to set the standard for how physics-informed AI is applied to complex real-world challenges.
Biography: Nazanin Ahmadi
Representing the next generation of scientific leadership, Nazanin Ahmadi is a PhD candidate in Biomedical Engineering at Brown University. Her research is at the vanguard of applying agentic AI to systems biology and pharmacometric modeling.
Ahmadi has gained significant recognition for her pioneering work in utilizing PINNs to solve inverse problems in pharmacology. Her research bridges the gap between theoretical machine learning and actionable biomedical insights, focusing on the creation of "mechanistically grounded" AI. Her contributions have been acknowledged through several prestigious awards, including a win at the Machine Learning in Drug Discovery Symposium at the Broad Institute and a top-three finish in the ASCPT Student & Trainee Research Pitch Competition. As an active organizer of SciML symposia, she is not only contributing to the technical literature but is also actively shaping the community of researchers working at the intersection of AI and biology.
Supporting Data and Real-World Applications
The presentation will provide concrete evidence of these systems in action, moving beyond abstract theory to demonstrate utility in high-stakes environments.
Decoding Complex Systems
One of the key examples provided in the abstract is the application of these agentic frameworks to gene-expression trajectories. By integrating prior biological knowledge with high-dimensional data, the agents can identify non-obvious patterns in how genes regulate one another—patterns that traditional statistical methods often overlook.
Advancing Precision Medicine
In the realm of cancer-treatment dynamics, the ability to formulate and test hypotheses autonomously is revolutionary. Ahmadi’s work suggests that by using agentic AI, researchers can simulate the impact of various therapeutic interventions on patient-specific models. These agents can effectively "stress-test" treatment protocols, predicting potential resistance mechanisms before they appear in clinical settings. This capability represents a massive shift from the trial-and-error approach currently dominating many areas of pharmacology toward a model-driven, predictive future.
Implications for the Future of Scientific Research
The implications of this research are profound. If we can successfully offload the formulation and iterative testing of hypotheses to agentic AI, we may be on the cusp of a "discovery explosion."
Toward Autonomous Discovery
The ultimate goal of this research is not to replace human scientists, but to augment their capabilities. By automating the "drudgery" of hypothesis testing and validation, AI can free human researchers to focus on the high-level conceptual questions that define scientific breakthroughs. The framework proposed by Karniadakis and Ahmadi suggests that future scientific laboratories will function as hybrid environments, where human researchers define the goals and physical constraints, and a fleet of specialized agents executes the rigorous search for novel models.
Identifying the Unexplained
Perhaps most importantly, the speakers will address the potential for these systems to identify "unexplained patterns." In many fields, data is abundant, but the underlying mechanisms remain elusive. Agentic AI is uniquely positioned to traverse massive datasets, identifying anomalies that do not fit existing models and suggesting new, testable hypotheses to explain them. This could provide the necessary impetus to move scientific fields forward that have historically stalled due to the complexity of the data involved.
Conclusion: A New Era of Collaboration
The upcoming talks by George Karniadakis and Nazanin Ahmadi offer a rare glimpse into the future of scientific inquiry. By synthesizing physics-informed machine learning with multi-agent reasoning, they are providing the blueprint for a new generation of computational research.
As these technologies mature, the line between "computational tool" and "research partner" will continue to blur. The scientific community is currently at a turning point, moving away from purely descriptive data analysis toward a future of autonomous, interpretable, and truly creative discovery. Whether exploring the intricacies of gene expression or designing the next generation of life-saving therapeutics, the integration of agentic AI into the scientific workflow stands as one of the most promising frontiers of the 21st century.
For researchers, students, and practitioners interested in the convergence of AI and the natural sciences, these sessions offer an essential look at the methodologies that will define the next decade of discovery.
About the Author
