Oden Institute Faculty Lead Panels on Climate Science, Cosmos and Medicine at UT's AI LIVE

While artificial intelligence (AI) has been around since the 1950’s, it has become the ‘it’ topic, exploding even further with the 2022 arrival of ChatGPT. AI is infiltrating all aspects of technology, and our lives. While it offers the potential to improve everything from mundane tasks to cutting-edge improvements and new discoveries in healthcare, climate science, and the cosmos, many questions remain, such as building trust, and maintaining transparency and accuracy in this evolving relationship with AI.

Designating 2024 the Year of AI, The University of Texas at Austin showcased groundbreaking research and highlighted the interdisciplinary approach to AI, and provided discourse for the unanswered questions that lay ahead. The year-long initiative was capped off with AI LIVE, a three-day showcase, held Nov. 13-15, with events featuring expert panels, innovations and inspiration in the world of AI. 

Faculty and researchers from the Oden Institute for Computational Engineering and Sciences led three panels covering Earth sciences, astronomy and medicine. Graduate students representing the Oden Institute’s Center for Autonomy participated in the robotics parade, marching robots of all shapes and sizes down Speedway Plaza, bringing an energetic conclusion on the final day of AI LIVE.

AI for Climate Science and Natural Hazards, a panel led by Oden faculty Thorsten Becker, Omar Ghattas, and Patrick Heimbach, discussed the adoption of the digital twin concept: a computational replica of a physical system, capable of simulating predictive responses with usage in the areas of ocean climate science, tsunami forecasting and earthquake dynamics. AI has shown potential to reduce computational costs, both in time and resources.

“For natural hazards, the goal is to discover how can risk be mitigated quickly,” said Ghattas, Director of the OPTIMUS Center at the Oden Institute. “Imagine it takes a day to get the data for a prediction – a computational model might have to run hundreds of thousands of times. How can AI come to the rescue? Combining machine learning in combination with neural networks, predictions can be executed in milliseconds instead of hours. A successful digital twin of a tsunami would give an evacuation order well before a city would be impacted, with machine learning and AI, we can do predictions much quicker.”

Thorsten Becker, affiliated faculty at the Oden Institute and professor in the Department of Earth and Planetary Sciences, is using AI with an end goal of creating earthquake forecasts. Becker said the goal is to utilize AI algorithms to improve prediction accuracy. “Using digital twins to predict the Earth’s behavior in terms of earthquakes - turns out AI systems can do quite well in predicting behaviors. AI can help us figure out the right questions and detect how systems change, so that we can understand what we should be measuring to improve what will eventually be some sort of weather forecast for earthquakes.”

“In oceanography, continuous measuring in the greatest of ocean depths isn’t always feasible, and this is when the power of data assimilation comes,” said Patrick Heimbach, lead of the Computational Research in Ice and Ocean Systems Group (CRIOS) at the Oden Institute, “As Dr. Ghattas stated, we bring in neural network into a physical model, to solve equations, taking portions of the model that aren’t well known to train the model on the data. This is one application of machine learning, and a stepping-stone to realizing a computational digital twin of the ocean.” 

At the mid-morning panel discussion, Exploring the Universe with Artificial Intelligence, Stella Offner, Director of the new NSF-Simons AI Institute for Cosmic Origins (CosmicAI) housed within the Oden Institute, talked about the role of AI in cosmology. 

“Astronomy is a really amazing space for AI, in part because our data is publicly available,” said Offner. “We are in a Big Data era because we have these very large telescopes producing huge data sets, and we need methods that can sort data quickly and efficiently.” Offner explained that AI can be trained on simulations by giving it real data and asking it to define features that can’t otherwise be easily seen by the eye, adding, “we're really just beginning.”

Offner shared the panel with Matthew Lease, Co-Director of CosmicAI and professor at UT’s School of Information, and Arya Farahi, assistant professor in the Department of Statistics and Data Sciences at UT. The panel was moderated by post-doctoral fellow Josh Taylor, in UT’s Department of Astronomy.

Lease, who is also with UT’s Good Systems, emphasized the importance of building trust with AI output. “Responsible AI can’t be designed in a vacuum – it needs to be grounded in societal challenges that are consequential. Taking a problem, such as cosmology and learning where the gaps in AI are such that we can fix those gaps and advance AI in general, and then drive that back to advance cosmology itself.”

“These collaborations between computer scientists, and math statisticians and astronomers - working together, we have a good sense of the big picture, which gives us confidence that when we apply these AI models that we're able to do good science with them.”

Arya emphasized the importance of providing education at all grade levels, “All the issues and methods we are looking at can very easily translate to other domains. There is this triangle of education, computing and data. As a part of the Institute, we are focusing on bringing these three resources together so the larger community can use them.” 

Transforming Medicine with Math and AI, facilitated by Tom Yankeelov, Director of the Center for Computational Oncology at the Oden Institute, featured presentations by two students in the Computational Science, Engineering and Mathematics (CSEM) graduate program at the Institute: Shruti Motiwale and Casey Stowers.

“Every 30 seconds someone dies from heart disease,” said Motiwale, a CSEM student with the Willerson Center for Cardiovascular Modeling and Simulation. Motiwale, who recently defended her dissertation, presented collaborative research with Hao Liu, Wenbo Zhang, and Michael Sacks, Director of the Willerson Center.

While comprehensive virtual models of the heart are being developed for treatment planning for cardiac diseases, their practical applications in the clinic are limited because they take too long to provide results. 

“We have developed a novel approach using machine learning that retains the same level of accuracy as well as established traditional simulation methods but provides results within seconds. Our approach overcomes the need for training data by replacing the cost function for data-driven training with a cost function derived from physics, and in turn allows us to retain high accuracy. We implemented the approach for a geometrically-idealized model of the left ventricle to predict mechanical effects of heart attack in any location on the ventricle within 3 seconds compared to 7 hours for traditional methods and within 0.9% accuracy.”

Stowers, a fifth year Ph.D candidate in the CSEM program, presented on methods on integrating mechanism-based and data-driven modeling to predict breast cancer response to therapy. Goals in the treatment include developing models with early treatment data that can predict a patient’s response and determine if the treatment will be relevant to the patient. 

While the multitude of uses of AI are still unfolding, UT and Oden Institute faculty, students and researchers are seeking innovative ways to incorporate the many benefits that AI can offer to solve societal grand challenges across disciplines. AI can improve efficiency, both in time and resources, and researchers are working to ensure artificial intelligence is ethical, transparent and trustworthy to navigate the ever-evolving landscape brought about by AI.