In the rapidly evolving landscape of artificial intelligence, the transition from simple automation to genuine scientific creativity represents the next great technological hurdle. While current AI systems excel at processing vast datasets and optimizing existing workflows, a fundamental question persists: Can machines move beyond the role of a tool to become an active, collaborative partner in the generation of novel scientific hypotheses?
On the horizon is a pivotal session featuring two of the field’s most prominent voices—George Karniadakis and Nazanin Ahmadi—who will explore this exact inquiry. Through the lens of Scientific Machine Learning (SciML), the upcoming talks will bridge the divide between computational power and human-like scientific intuition, presenting a new paradigm for how we approach discovery in physics, biology, and pharmacology.
The Schedule: A Deep Dive into AI-Driven Science
The upcoming event promises a comprehensive exploration of the intersection between physics-informed computation and autonomous agentic workflows.
- 9:00 a.m. – 9:45 a.m.: George Karniadakis. As a global leader in applied mathematics and computational science, Karniadakis will set the stage by defining the foundational architecture of SciML. His presentation will address how we can fuse the raw predictive power of neural networks with the immutable laws of physics to create "computational engines" capable of genuine scientific reasoning.
- 10:00 a.m. – 10:50 a.m.: Nazanin Ahmadi. Building on these foundations, Ahmadi will move the discussion toward implementation. Focusing on systems biology and pharmacology, she will demonstrate how agentic frameworks—systems where multiple AI agents debate, test, and refine hypotheses—can be applied to complex, real-world biological challenges.
The Abstract: Moving Beyond Automation
The core thesis of the joint presentation challenges the current trajectory of AI development. Scientific creativity is rarely a solitary endeavor; it is a social, iterative process of interaction between ideas, models, data, and researchers. The speakers argue that if AI is to contribute meaningfully to science, it must mirror this collaborative structure.
The Role of Physics-Informed Neural Networks (PINNs)
The speakers posit that the traditional "black box" approach of deep learning is insufficient for science. Instead, they advocate for Scientific Machine Learning, where models are constrained by mathematical structure and physical knowledge. By integrating physics-informed neural networks and neural operators, researchers can ensure that AI-generated models are not only accurate but also interpretable and physically consistent.
The Agentic Framework
Perhaps the most revolutionary aspect of their research is the proposal of an "agentic framework" for scientific discovery. Rather than relying on a single, monolithic model, the framework employs a multi-agent system. In this environment:
- Specialized agents propose candidate hypotheses.
- Critique agents challenge these hypotheses against prior knowledge and physical laws.
- Validation agents perform mathematical and computational testing.
- Self-improvement cycles allow the system to learn from its own failures and successes.
This methodology aims to transform the machine from a passive calculator into an active participant capable of identifying unexplained patterns and proposing entirely new directions for research.
Meet the Speakers
George Karniadakis: A Titan of Applied Mathematics
George Karniadakis, a member of both the National Academy of Engineering and the National Academy of Arts and Sciences, brings decades of experience to the table. Currently serving as a professor at Brown University and a Vannevar Bush Faculty Fellow, his influence on the field is unparalleled.
With an h-index of 171 and over 184,000 citations, Karniadakis has been recognized by Research.com as the world’s leading math scientist for 2025. His academic pedigree—spanning MIT, Stanford, Princeton, and Caltech—is matched only by his long list of accolades, including the William Benter Prize, the SIAM/ACM Prize on Computational Science & Engineering, and the Alexander von Humboldt award. His career has been defined by his ability to bridge the gap between abstract mathematics and real-world mechanical and ocean engineering, a trajectory that now finds its zenith in his work on SciML.
Nazanin Ahmadi: A Rising Star in Biomedical AI
Nazanin Ahmadi represents the next generation of researchers pioneering the practical application of agentic AI. A PhD candidate in Biomedical Engineering at Brown University, Ahmadi has established herself as a leader in the intersection of systems pharmacology and machine learning.
Her work focuses on the "inverse problem"—using AI to infer the underlying mechanistic models that drive biological phenomena. By applying physics-informed neural networks to drug discovery and pharmacometric modeling, Ahmadi is pushing the boundaries of what is possible in precision medicine. Her recognition as a winner at the Broad Institute’s Machine Learning in Drug Discovery Symposium and the ASCPT Student & Trainee Research Pitch Competition underscores her role as a vital contributor to the future of autonomous scientific workflows.
Supporting Data and Evidence
The efficacy of the proposed framework is grounded in its application to complex, high-stakes biological domains. The speakers will present case studies ranging from gene-expression trajectories to cancer-treatment dynamics.
These examples serve to prove that when AI is allowed to "think" through a process—by checking its own work against physical constraints and mathematical proofs—it can uncover interpretable models that human researchers might overlook. For instance, in pharmacometric modeling, the ability to generate a testable hypothesis regarding how a drug interacts with a specific protein pathway can save years of clinical trial time. By automating the "discovery" portion of the scientific method, these systems allow scientists to focus on the high-level interpretation of results rather than the tedious labor of data fitting.
Implications for the Scientific Community
The shift toward Agentic AI in science has profound implications for how research institutions will function in the coming decade.
1. Accelerating Discovery Cycles
By offloading hypothesis generation and initial validation to multi-agent AI systems, the speed of discovery could increase exponentially. What currently takes years of manual data analysis and hypothesis testing could be compressed into weeks of agentic cycles.
2. Enhancing Interpretability
One of the primary criticisms of AI in science has been the lack of transparency in "black box" models. The speakers’ emphasis on "physically grounded AI" directly addresses this, ensuring that the discoveries made by these machines are explainable to the human experts who must ultimately approve them for scientific publication or clinical use.
3. A New Paradigm of Collaboration
The future of science, as envisioned by Karniadakis and Ahmadi, is not "Human vs. AI," but "Human-in-the-loop AI." In this model, the AI acts as an tireless research assistant that never sleeps, constantly proposing new avenues of inquiry, while the human researcher acts as the architect, defining the scope and validating the ethical and strategic direction of the work.
Conclusion: The Path Ahead
The upcoming presentations by George Karniadakis and Nazanin Ahmadi serve as a call to action for the scientific community. We stand at a threshold where the tools of computation are finally catching up to the complexity of the scientific method. By shifting our focus from simple automation to the development of agentic frameworks that incorporate memory, reasoning, and self-criticism, we are entering an era where machines will not just process data—they will help us understand the very laws that govern our universe.
Whether one is a practitioner in the field of artificial intelligence, a biologist seeking new tools, or an engineer interested in the future of computational reasoning, this session provides an essential roadmap for the integration of machine intelligence into the creative process of discovery. As the speakers prepare to share their findings, the message is clear: the future of science is autonomous, it is physics-informed, and it is collaborative.
About the Author
