The landscape of modern science is undergoing a tectonic shift. For centuries, the engine of discovery has been the human intellect—a solitary process of hypothesis, experimentation, and rigorous verification. However, as the volume of data grows exponentially and the complexity of biological and physical systems transcends human cognitive bandwidth, a new paradigm is emerging.
On the horizon is a powerful synthesis: the marriage of Scientific Machine Learning (SciML) and Agentic AI. This collaborative evolution aims to move beyond simple automation, transforming computers from passive tools into active, creative partners in the scientific method. This transition will be the focus of an upcoming high-level symposium featuring two of the field’s leading voices: National Academy of Engineering member George Karniadakis and rising researcher Nazanin Ahmadi.
The Core Thesis: Moving Beyond Automation
At the heart of the upcoming discussion is a fundamental question: Can AI move past the role of a data-processor and become a generator of genuinely novel hypotheses?
The prevailing view in computational science is that creativity does not happen in a vacuum. It is the product of iterative feedback loops—interactions between ideas, empirical data, and mathematical models. While current AI systems excel at pattern recognition, the speakers argue that the next frontier is "Agentic AI." Unlike static models, these agents function as specialized, multi-disciplinary teams capable of memory retention, information sharing, and self-correction.
The proposed framework utilizes SciML as a foundation, embedding physical laws and mathematical structures directly into neural networks. By constraining AI systems with the immutable laws of physics—a process known as Physics-Informed Neural Networks (PINNs)—researchers can ensure that the "reasoning" performed by these agents remains grounded in reality, even when exploring vast, uncharted theoretical spaces.
Symposium Chronology: A Morning of Innovation
The symposium is structured to provide both a macro-level overview of the theoretical foundations and a granular look at the practical applications of these systems.
9:00 a.m. – 9:45 a.m.: George Karniadakis
Professor Karniadakis will open the session by framing the broader scientific challenge. Drawing from his decades of expertise in computational mathematics, he will explore how neural operators and SciML frameworks act as the engine for a new era of model discovery. His presentation will establish the theoretical necessity of combining data-driven learning with physical constraints, providing the "rules of the road" for building reliable, autonomous scientific agents.
10:00 a.m. – 10:50 a.m.: Nazanin Ahmadi
Following the theoretical groundwork, PhD candidate Nazanin Ahmadi will shift the focus toward the practical, agentic implementation. Her presentation will highlight the application of these frameworks in systems biology and pharmacology. By showcasing how specialized agents can propose, critique, and validate candidate explanations, she will demonstrate the power of autonomous workflows in solving complex inverse problems in medicine.
Supporting Data and Technical Context
The shift toward Agentic AI is driven by the limitations of traditional, monolithic AI approaches. A single, large language model or neural network often lacks the specialized depth required for rigorous scientific validation.
The Multi-Agent Advantage
The framework presented by the speakers relies on a distributed architecture. In this system:
- Specialized Agents: Separate modules are tasked with specific functions—one might focus on hypothesis generation, while another acts as a "skeptic" responsible for mathematical and computational validation.
- Iterative Refinement: Through a process of self-improvement and cross-agent critique, the system filters out spurious correlations, ensuring that only scientifically plausible models remain.
- Integration of Prior Knowledge: By leveraging SciML, these agents do not start from scratch. They ingest existing literature, physical equations, and historical data, significantly reducing the "hallucination" risk common in non-scientific AI.
Real-World Efficacy
The research draws on compelling use cases, most notably in gene-expression trajectories and cancer-treatment dynamics. In these high-stakes environments, the ability of AI to integrate vast datasets—often noisy or incomplete—with known biological mechanisms provides a level of predictive power that purely data-driven models cannot match.
Biographies: The Minds Behind the Method
George Karniadakis: A Pioneer of Computational Mathematics
George Karniadakis is widely recognized as one of the most influential figures in modern applied mathematics. With an h-index of 171 and over 184,000 citations, he has spent his career bridging the gap between fluid mechanics, machine learning, and physical modeling.
His academic journey is a tapestry of prestige, spanning from MIT and Princeton to his current tenure at Brown University. As an elected member of the National Academy of Engineering and the National Academy of Arts and Sciences, his work has consistently pushed the boundaries of what is computationally possible. In 2025, he was formally recognized by Research.com as the number one math scientist in the world, a testament to his sustained impact on the global research community. His involvement in the symposium brings a level of institutional and historical depth that anchors the discussion in the rigors of classical computational science.
Nazanin Ahmadi: Leading the Next Generation
Nazanin Ahmadi represents the vanguard of the next generation of scientific researchers. As a PhD candidate in Biomedical Engineering at Brown, her work is defined by a commitment to "mechanistically grounded AI."
Ahmadi is recognized for her pioneering work in applying physics-informed neural networks to the field of pharmacometrics. Her research seeks to make AI models interpretable—a critical requirement for the adoption of automated discovery in clinical settings. With multiple awards for her lightning talks and research pitches at prestigious institutions like the Broad Institute, she has quickly established herself as a leading voice in the intersection of autonomous workflows and systems pharmacology.
Implications: The Future of the Scientific Method
The implications of integrating Agentic AI into the laboratory are profound. We are moving toward a future where the "bottleneck" of scientific discovery—the human capacity to process information—is significantly widened.
Toward Autonomous Discovery
The goal of this research is not to replace the scientist, but to provide them with a "super-powered" assistant capable of navigating high-dimensional data spaces that would take a human lifetime to map. By automating the routine aspects of hypothesis testing and validation, these systems allow human researchers to focus on the higher-order tasks of research strategy, ethics, and philosophical direction.
The Problem of "Unexplained Patterns"
Perhaps the most ambitious aspect of the upcoming talks is the focus on "identifying unexplained patterns." Current AI is largely designed to optimize known objectives. However, the next generation of SciML agents is being designed to identify anomalies—patterns in data that defy current physical understanding. This is the very definition of scientific innovation: the ability to recognize when the data contradicts the theory, and the initiative to propose a new, testable hypothesis to reconcile the difference.
Ethical and Practical Considerations
As with any technological leap, the adoption of Agentic AI brings challenges. Questions regarding the transparency of "black-box" models, the reproducibility of AI-generated hypotheses, and the potential for bias in training data are central to the work of both Karniadakis and Ahmadi. By focusing on interpretable AI and physically constrained models, they are directly addressing the trust gap that often prevents AI from being adopted in critical sectors like drug discovery and medicine.
Conclusion
The symposium represents a critical junction in the evolution of scientific research. By moving from isolated computation to a collaborative, multi-agent framework, science is on the verge of a new "Age of Discovery." Whether it is accelerating the development of life-saving drugs or uncovering the hidden dynamics of complex physical systems, the integration of Scientific Machine Learning and Agentic AI offers a path forward that is both powerful and grounded in the fundamental laws of nature.
For those in the fields of mathematics, engineering, biology, or AI development, the session promises not just a glimpse of the future, but a blueprint for how that future will be built. As George Karniadakis and Nazanin Ahmadi prepare to share their findings, the scientific community awaits a vision that balances the raw power of machine learning with the enduring wisdom of physical theory.
About the Author
