Keynote Speakers
Mario Krenn, Research Group Leader of the Artificial Scientists Lab, Max Planck Institute for the Science of Light
About the Speaker
Dr. Krenn is a research group leader of the Artificial Scientist Lab at the Max Planck Institute for the Science of Light (Theory Division). He is excited about the potential of artificial intelligence-inspired and -augmented science, and how we can use algorithms in a more “creative” way. To make progress, he believes it will be important to learn what humans mean by crucial scientific concepts such as surprising, creativity, understanding or interest. He has created AIs for designing quantum experiments and hardware (several actually built in labs) and inspiring novel ideas for quantum technologies. (See these Scientific American and National Academy of Science articles for recent coverage on part of this research). He also builds autonomously semantic networks from scientific publications, and use machine learning to predict and suggest personalized future research questions and ideas. In that sense, Krenn uses the machine as a source of inspiration to accelerate scientific progress. Ultimately, he wants to create algorithms that help us to uncover the secrets of the Universe. More about the research philosophy and ideas can be read in this Nature Review Physics perspective. Krenn’s work was awarded an ERC Starting Grant 2024.
[1] Krenn, Kottmann, Tischler, Aspuru-Guzik, Conceptual understanding through efficient automated design of quantum optical experiments. Physical Review X 11(3), 031044 (2021).
[2] Krenn, Pollice, Guo, Aldeghi, Cervera-Lierta, Friederich, Gomes, Häse, Jinich, Nigam, Yao, Aspuru-Guzik, On scientific understanding with artificial intelligence. Nature Reviews Physics 4, 761–769 (2022).
[3] Krenn et al., Forecasting the future of artificial intelligence with machine learning-based link prediction in an exponentially growing knowledge network, Nature Machine Intelligence 5, 1326 (2023)
[4] Gu, Krenn, Interesting Scientific Idea Generation Using Knowledge Graphs and LLMs: Evaluations with 100 Research Group Leaders. arXiv:2405.17044 (2024)
Towards an Artificial Muse for New Ideas in Science
Abstract
Artificial intelligence (AI) is a potentially disruptive tool for physics and science in general. One crucial question is how this technology can contribute at a conceptual level to help acquire new scientific understanding or inspire new surprising ideas. I will talk about how AI can be used as an artificial muse in physics, which suggests surprising and unconventional ideas and techniques that the human scientist can interpret, understand and generalize to its fullest potential.
[1] Krenn, Kottmann, Tischler, Aspuru-Guzik, Conceptual understanding through efficient automated design of quantum optical experiments. Physical Review X 11(3), 031044 (2021).
[2] Krenn, Pollice, Guo, Aldeghi, Cervera-Lierta, Friederich, Gomes, Häse, Jinich, Nigam, Yao, Aspuru-Guzik, On scientific understanding with artificial intelligence. Nature Reviews Physics 4, 761–769 (2022).
[3] Krenn et al., Forecasting the future of artificial intelligence with machine learning-based link prediction in an exponentially growing knowledge network, Nature Machine Intelligence 5, 1326 (2023)
[4] Gu, Krenn, Interesting Scientific Idea Generation Using Knowledge Graphs and LLMs: Evaluations with 100 Research Group Leaders. arXiv:2405.17044 (2024)
Pat Langley, Principal Research Scientist, Georgia Tech Research Institute
About the Speaker
Dr. Pat Langley is a Principal Research Scientist at Georgia Tech
Research Institute and Director of the Institute for the Study of
Learning and Expertise. He has contributed to AI and cognitive science for more than 40 years, publishing over 300 papers and five books on these topics. Dr. Langley developed some of the first computational approaches to scientific knowledge discovery, and he was an early champion of experimental studies of machine learning and its application to real-world problems. He is the founding editor of two journals, Machine Learning in 1986 and Advances in Cognitive Systems in 2012, and he is a Fellow of both AAAI and the Cognitive Science Society. Dr. Langley’s current research focuses on architectures for embodied agents, explainable, normative, and justified agency, and induction of dynamic process models from time series and background knowledge.
.
Integrated Systems for Computational Scientific Discovery: Progress, Challenges, and Implications
Abstract
There has been a steady stream of AI work on scientific discovery
since the 1970s, much of it leading to published results in fields
like astronomy, biology, chemistry, and physics. However, most efforts have focused on isolated tasks rather than addressing their interaction. In this talk, I challenge the research community to develop and adopt integrated discovery systems. I note distinguishing features of scientific discovery and examine five component abilities, in each case specifying the problem and reviewing results in the area. After this, I note some successes at partial integration and consider some remaining hurdles that we must leap to transform the vision for integrated discovery into reality. I also discuss promising domains, natural and synthetic, in which to test such computational artifacts. In closing, I consider ways that integrate discovery can accelerate scientific progress and factors that influence whether results are trustworthy.
Nyalleng Moorosi, Senior Researcher, Distributed AI Research Institute (DAIR)
About the Speaker
Nyalleng Moorosi is a senior researcher at DAIR, and her research interests are in understanding how we can build models which center populations often regarded as peripheral. Before DAIR, she was a research software engineer at Google, where she was one of the first employees at the Google Africa research lab. She has also been a senior researcher at the South African Council for Scientific and Industrial research, where she worked closely with government and academic institutions to develop a diversity of products for private and public institutions. Outside of formal work she is involved in efforts to democratize AI; she is a founding member of the Deep Learning Indaba, the largest machine learning consortium of AI/ML practitioners in Africa, a member of A+ Alliance an international coalition that seeks to not only detect, but correct, gender bias in Artificial Intelligence.
Moving Away From Imaginary Worlds for People in the Global South
Abstract
While there is significant excitement surrounding generative AI and the development of increasingly large models, smaller, localized, and specialized AI tools continue to demonstrate impressive performance, even when compared to larger AI models. And whereas large tech companies continue to concentrate on developing AI-for-everything models, smaller community-based developers leverage their understanding of local communities, cultures, and languages to create tools optimized for specific problems within those communities. This hyperlocalization influences decisions ranging from dataset construction and model architecture selection to product evaluation metrics. It also impacts prioritization, which directly relates to discussions about bias, safety, and alignment.
Over the last year and a half, my colleagues and I have engaged with developers who build for their communities to understand how being part of the target audience influences priorities. Our conversations have included developers from developing nations in Africa, Latin America, and Asia, as well as those from marginalized Indigenous populations in developed nations. In each interaction, we have sought to understand how the design choices of a researcher deeply familiar with a community’s political, cultural, and historical context differ from those of a researcher with limited understanding of this context.
In this presentation, I will share our findings from the past year and discuss how these insights continue to influence our efforts as we develop solutions for our communities.
Lav R Varshney, Associate Professor, University of Illinois Urbana-Champaign
About the Speaker
Lav Varshney is a professor of electrical and computer engineering at the University of Illinois Urbana-Champaign. He leads the AI and hybrid cloud thrust of the IBM-Illinois Discovery Accelerator Institute, a $200M investment. He is also co-founder and CEO of Kocree, Inc., a startup company using AI in social music co-creativity platforms to enhance human wellbeing across society, and chief scientist of Ensaras, Inc., a startup company focused on AI to optimize wastewater treatment. He further holds affiliations with RAND Corporation and with Brookhaven National Laboratory. He is a former White House staffer, having served on the National Security Council staff as a White House Fellow, where he contributed to national/international AI and wireless communications policy. Previously at IBM Research, he led the development and deployment of the Chef Watson system for culinary creativity, the first commercially successful generative AI technology, which also received worldwide acclaim. At Salesforce Research, he was part of the team that open-weight released the largest and most capable large language model at the time, jumpstarting the open-weight movement. His work and public scholarship has been featured in media ranging from Fox News and the Wall Street Journal to the New York Times, NPR, Slate, and The New Yorker. He appeared in the Robert Downey, Jr. documentary series, Age of AI. He holds a B.S. degree in electrical and computer engineering from Cornell University and S.M. and Ph.D. degrees in electrical engineering and computer science from the Massachusetts Institute of Technology. His current research interests include information theory; AI foundations, explainability, and governance; agent-based policymaking; creativity and innovation; and AI applications in sustainability, science, and wellbeing.
AI and Scientific Creativity
Abstract
Artificial Intelligence (AI) is increasingly contributing to scientific discovery, yet its potential for enhancing scientific creativity remains largely unexplored. Drawing insight from the philosophy of science and creativity research, we describe a novel classification of AI’s contributions to scientific creativity based on the concreteness and autonomy of generated ideas. These include: (1) AI Muse, where AI systems suggest surprising combinations of concepts that researchers can develop into concrete research plans; (2) AI Designer, where systems propose executable ideas in the form of specific experiments, equations, or reaction pathways; and (3) AI Scientist, where AI systems autonomously execute discovered ideas through automated experimentation, simulation, or mathematical proofs. For each class, we analyze how to increase both novelty and functionality—the twin components of creativity. We further discuss novelty and functionality in the context of our mathematical theory of creativity, which provides an information-geometric tradeoff between the two dimensions, as well as formal mathematical relationships between statistical and structural notions of creativity. The broad framework provides a foundation for developing AI systems that can contribute more meaningfully to scientific creativity and accelerate the pace of scientific discovery.
AI Journeys Speakers
“AI Journeys” are research stories presented by U-M and partners who have started research projects with AI implementation. These presentations will focus on the researchers’ experience on determining a research question, identifying an AI method, enlisting collaborators, and recommendations of dos and don’ts.
Peter Adriaens, Professor, U-M College of Engineering and U-M School for Environment and Sustainability
Designing an AI Model to Predict Financial Operational and Supply Chain Risk from Water Scarcity in Global Corporate Facilities
AI Summary
An advanced AI model is being developed to predict financial risks that climate-induced water scarcity poses to corporations, particularly in water-intensive industries. Collaboration was established through the U-M Center for Digital Asset Finance, partnering with LimnoTech for water resource modeling, and Equarius for financial risk modeling. Key challenges in the project include defining system boundaries, curating diverse data sources, developing causal relationships, and extracting meaningful insights through natural language processing, all of which are being addressed through interdisciplinary collaboration and leveraging computational infrastructure.
Harkirat Singh Arora, Ph.D. Student, Biomedical Engineering, Michigan Medicine
A Mechanistic Neural Network Model Optimizes both the Potency and Toxicity of Combination Therapies
AI Summary
This presentation will look at examining antibiotic resistance through AI-driven optimization of drug combination therapies. Challenges like bridging biological insights with AI and validating predictions were overcome through interdisciplinary collaboration with the Center for Chemical Genomics.
Barbara Glover, Program Manager, African Union Development Agency-NEPAD (AUDA-NEPAD)
Bridging Innovation and Equity: How the African Union is driving AI for Sustainable Development in Africa
AI Summary
Artificial Intelligence is a transformative force that holds immense potential to accelerate Africa’s development agenda. The African Union (AU) has taken a proactive role in shaping AI policies, fostering innovation, and ensuring that AI is leveraged as a tool for sustainable development. This talk will explore the AU’s strategic initiatives, including the Continental AI Strategy, governance frameworks, and partnerships that are driving AI adoption across key sectors such as healthcare, education, agriculture, and digital governance. Additionally, it will highlight ongoing efforts to address challenges related to data sovereignty, ethical AI, and capacity building to ensure that AI benefits all Africans equitably. Through policy leadership, collaboration, and infrastructure investment, the AU is positioning Africa not just as a consumer of AI but as an active contributor to the global AI landscape.
Lubomir M Hadjiyski, Professor of Radiology, U-M Medical School
Large Language Models, Deep learning and Radiomics for Survival Prediction of Bladder Cancer Patients
AI Summary
The project team used AI, including large language models, radiomics, and deep learning, for predicting five-year survival rates in bladder cancer patients by integrating clinical reports and imaging data. A multidisciplinary team collaborated closely with clinical experts to define key tasks, annotate data, and evaluate results, addressing challenges in data collection, model selection, and robust validation through pilot experiments, transfer learning, and iterative refinement of AI methods.
Dani Jones, Associate Research Scientist, Data Science and Modeling, U-M School of Environment and Sustainability
Navigating Environmental Data: Unsupervised Classification in Oceanic and Atmospheric Research
AI Summary
The team analyzed extensive oceanographic data from autonomous profiling floats was analyzed by applying unsupervised learning techniques, specifically Gaussian Mixture Modeling (GMM), to objectively identify large-scale ocean structures like gyres without relying on geographic information. To foster collaboration, an “AI Lab” was established to facilitate interactions between machine learning experts and domain scientists, addressing such as data complexity and interpretability.
Nkem Khumbah, Head of STI Policy Systems, Governance and Partnerships, African Academy of Sciences
AI Journey Title TBA
AI Summary
Summary TBA
Fraser King, Research Fellow, Climate and Space Sciences and Engineering, U-M College of Engineering
Towards Interpretable Machine Learning Models Across the Geosciences
AI Summary
This work addressed the challenge of ensuring trust in machine learning (ML) models for geosciences by focusing on explainability and interpretability to align model decisions with physical processes. Through international collaboration on projects using generative AI, sparse autoencoders, and dimensionality reduction, the team navigated challenges of model complexity, leveraging interdisciplinary expertise, robust validation, and transparent communication to foster trust in AI-driven approaches for high-stakes environmental applications.
To learn more about what Dr. King is working on, please visit: https://frasertheking.com
Nicholas Kotov, Irving Langmuir Distinguished University Professor of Chemical Sciences and Engineering, Joseph B and Florence V Cejka Professor of Engineering, Professor of Chemical Engineering and Professor of Macromolecular Science and Engineering, U-M College of Engineering
Graph Theory Toolbox for AI/ML Design of Complex Biomimetic Nanostructures
AI Summary
Summary TBA
Max Li, Assistant Professor of Aerospace Engineering, Civil and Environmental Engineering and Industrial and Operations Engineering, U-M College of Engineering
CHATAC: An Air Traffic Flow Management LLM-powered System
AI Summary
The project team deploys LLMs in air traffic flow management by developing CHATATC, a question-answering system trained on a comprehensive dataset of Ground Delay Programs. The presentation will focus on addressing challenges in query accuracy and information and share experience of working with collaborators in a highly regulated industry.
Xiaofan Liang, Assistant Professor of Urban and Regional Planning at U-M Taubman College of Architecture & Urban Planning
Lessons from a Human-in-the-loop Machine Learning Approach for Identifying Vacant, Abandoned, and Deteriorated Properties in Savannah, Georgia
AI Summary
The study used a human-in-the-loop machine learning (HITLML) approach to identify vacant, abandoned, and deteriorated (VAD) properties in Savannah, Georgia, addressing the challenge of limited municipal data and the need for transparency in AI processes. Collaboration between academic researchers and city planners revealed that while the AI system broadened property identification beyond traditional field surveys, the project faced challenges with expert involvement costs, data maintenance, and scaling the system, highlighting the importance of sustained human-AI interaction and infrastructure investment over technical model details.
Christopher Miller, Professor of Astronomy, U-M College of Literature, Science, and the Arts
Untangling Galaxy Evolution with AI
AI Summary
The researchers aimed to improve astrophysical inference by applying Convolutional Neural Networks (CNNs) to complex astronomical imaging, addressing the limitations of traditional models in capturing the intricate physical relationships between stars and their environments. Collaboration required expertise across astrophysics, instrumentation, and AI, overcoming challenges related to interpretability and scientific tradition, ultimately leading to a restructured, NSF-funded research program focused on trustworthy AI-driven discoveries.
Abiodun Modupe, Research Fellow and Lecturer of Computer Science, University of Pretoria
All You Need Is Community!
AI Summary
The Data Science for Social Impact research group at the University of Pretoria defines its scientific question around expanding AI and data science education beyond academia, leveraging collaborations with industry and grassroots initiatives like Masakhane and Deep Learning Indaba to strengthen African NLP research. By fostering inclusivity and innovation, the team navigates challenges in representation and sustainability through strategic partnerships, cutting-edge technology, and community-driven initiatives to build a thriving AI ecosystem in Africa.
Grite Nelson Mwaijengoa, Lecturer in the School of Materials, Energy, Water and Environmental Sciences, Nelson Mandela African Institution of Science and Technology
Revolutionizing African Great Lakes Management: AI for Smart Monitoring, and Data-Driven Decision Making
AI Summary
This presentation defines the scientific question as the challenge of efficiently monitoring the African Great Lakes (AGL) amidst climate change, overexploitation, and regulatory constraints, proposing AI-powered tools such as machine learning, satellite imagery, and remote sensing analytics to automate environmental assessments and enhance decision-making. By integrating AI-driven predictive modeling and standardized data-sharing frameworks, the project fosters cross-border collaboration, improves monitoring accuracy, reduces costs, and supports sustainable resource management, addressing key implementation challenges through technological innovation and coordinated governance strategies.
Qing Qu, Assistant Professor of Electrical Engineering and Computer Science, U-M College of Engineering
Learning Deep Low-dimensional Models from High-Dimensional Data: From Theory to Practice
AI Summary
This presentation addresses the integration of low-dimensional models and deep learning in science and engineering, emphasizing the importance of identifying inherent low-dimensional structures in data to enhance the formulation and understanding of deep learning models. They propose that principles from low-dimensional models can guide the design of more parameter-efficient, robust, and interpretable deep learning models, offering new perspectives on learned representations, generalizability, and transferability.
Geoffrey Siwo, Research Assistant Professor, Learning Health Sciences, U-M Medical School; Co-director, Ecosystems, Finance & Health
Can LLMs and AI Agents Accelerate Real-world Adoption of Scientific Evidence?
AI Summary
This presentation defines the scientific question as how to accelerate the translation of scientific evidence into real-world solutions, particularly in addressing the interconnected impacts of climate change, biodiversity loss, and food systems on human health. Through the Ecosystems, Finance & Health (EFH) initiative’s innovation sprint in Nairobi, they explore leveraging generative AI to enhance access to high-quality scientific evidence, emphasizing collaboration among researchers, policymakers, and funders while addressing challenges in integrating AI-driven tools into decision-making processes.
Jon Zelner, Associate Professor of Epidemiology, U-M School of Public Health
Confronting the Challenges of AI-assisted Infectious Disease Transmission Modeling Outside the ML ‘black box’
AI Summary
This presentaiton will demonstrate the importance of deep, interdisciplinary collaboration in calibrating complex infectious disease models using AI-driven Bayesian techniques, moving beyond a traditional “consultative” approach to foster meaningful integration across Epidemiology, Biostatistics, and Statistics. By blurring the boundaries between methodological and substantive expertise and prioritizing cross-disciplinary mentorship, this project highlights the need for mutual understanding and transparency to navigate the growing complexity of data and model calibration.
AI Tutorial Instructors
At lunchtime on both days of the symposium, MIDAS postdoctoral researchers will offer hands-on tutorials on AI methods and their applications in research. No registration needed. Open to all symposium attendees.
James Boyko, Assistant Professor/Postdoc Scholar, Ecology and Evolutionary Biology, U-M College of Literature, Science, and the Arts
An Introduction to Generative AI Tools for Research
Tutorial Description
This tutorial introduces using commercially available Generative AI for research. Assuming no background knowledge, we will discuss which tools are currently available, how to best use them, and address common misconceptions. This tutorial will not dive into the technical details underlying the models themselves, instead focusing on how to use these tools to enhance your research workflow. Most of the tutorial will be focused on Large Language models (LLMs) such as ChatGPT and the skill of prompting to receive high quality replies
Nathan Fox, Schmidt AI in Science Fellow and MIDAS AI Scientist
Segment Anything Model: A Beginner’s Guide to AI-Powered Image Segmentation
Tutorial Description
This tutorial introduces the Segment Anything Model (SAM), a cutting-edge AI tool for universal image segmentation. Image segmentation is the process of dividing an image into meaningful parts, such as separating objects, regions, or structures within the image. It’s a fundamental technique for tasks like identifying animals in photos, analyzing satellite imagery, or segmenting cells in microscope imagery. Participants will learn how SAM works, explore its key features, and apply it to diverse research scenarios, including ecology, microbiology, and geography. Through hands-on exercises, attendees will gain practical experience in leveraging SAM for their projects.
Questions? Contact Us.
Message the MIDAS team: [email protected]