Are LLM chatbots a form of omniscient super-intelligence — or just BS? That was the question posed by distinguished UC Santa Barbara materials professor Simon Billinge during his like-titled talk, presented on May 18 as part of the UCSB Library’s AI in Action speaker series. While the title got a laugh from the audience, Billinge, director of the California NanoSystems Institute (CNSI), followed it with an overview of his research and a thoughtful exploration of what intelligence means in LLMs. He spoke alongside Nina Miolane, assistant professor of electrical and computer engineering (ECE) and co-director of REAL AI (Reliable, Efficient, and ALigned AI) for Science, who discussed how her group employs AI in brain science.

Cooking Up New Materials with AI

Billinge, who came to UCSB in January 2026, explained how he and his colleagues use AI in their materials research at CSNI, where he studies how atoms are arranged in materials, arrangements that determine materials’ properties and their possible applications “As materials scientists,” he said, “we can play around with rearranging atoms to get all kinds of different materials.” 

Researchers often try to determine the relationship between a material’s properties and the arrangement of its atoms. Companies such as Meta, Google, and Microsoft are using AI to accelerate these efforts, which can lead to discovering new materials more quickly. AI, Billinge said, “is powering a revolution in materials science.”

Billinge compared the discovery process to developing a recipe in the kitchen. What is called the forward problem — using a material’s atomic structure to determine its properties — could be thought of, he said, as akin to asking “‘Given this recipe [or atomic arrangement], what dish [material] do I get?’” The inverse problem, determining the atomic structure that underlies a particular material and its properties, would be like working backwards, “reverse engineering,” from the dish to determine the recipe.

At CNSI, Billinge and his colleagues want to capture all of that scientific “cooking,” from ingredients to recipe to feast and back again, in large databases, and then develop autonomous machine-learning systems that can be used to discover the atomic recipes.

LLMs: What Does “Intelligence” Mean? 

During his talk, Billinge pivoted toward broader questions around AI, especially the big one, asking “Do LLMs have intelligence, and will they take over the world and enslave or destroy us?” While he said he couldn't answer that question, he talked through some of his thinking about this issue. 

To start, he said, defining intelligence can be challenging. Currently, the main definition for machine intelligence comes from the Turing test. As Billinge explained it, the Turing test states that if an examiner cannot tell the difference between the responses of a machine and a human to a written test, the machine is considered intelligent. When mathematician and computer scientist Alan Turing came up with the test, first called the imitation game, in 1949, Billinge said, “I bet he thought this would never happen.”  

But starting with ChatGPT-3.5, LLMs began passing the Turing test. Yet there are still significant differences between these machines and human intelligence. For instance, Billinge explained, children learn using all of their senses and, over time, can understand and distinguish between concepts. “LLMs are not even trying to understand concepts.” Rather, they are ingesting enormous quantities of text and then “just doing statistics,” he said. “They're predicting the most probable next word, given the words that came before.

“When someone who doesn't understand the concept is explaining it to you, you are being bullshi**ed,  he added. “I don’t think that LLMs are the problem. I think that bull**it is the problem.” If an LLM can pass the Turing test, he said, that means that humans aren’t discerning enough. “If we can’t detect bull**it, it doesn't matter whether it's a machine or a human behind the curtain, because either way we are being bullshi**ed. This is what we should work on.”

AI and the Maternal Brain

Miolane, co-director of AI for the nationwide Ann S. Bowers Women's Brain Health Initiative, headquartered at UCSB, harnesses AI in a different way: to build models of the maternal brain. The geometric models Miolane and her team have developed, called digital twins, are a critical part of Miolane’s efforts to understand how women’s brains change throughout pregnancy and motherhood — research which has received a $1 million grant from the Chan Zuckerberg Foundation.

The idea behind Miolane’s approach to AI came from what she called a radically simple idea. “Reality is not made of text,” she said. So instead of predicting the next word, what if AI was used to predict the next step in a process? 

In her work, the next step means the next physical state of the brain. To understand this, she and her team create computer digital-twin models that are replicas of the human brain; the researchers then use AI with these models to predict changes in the brain over time. But to get the predictive power that Miolane wants, using current AI models, would likely require billions of images. “In the brain sciences, we have at best thousands of brain images for a particular condition,” she said, “so, most of our work at REAL AI for Science is to rethink AI to make it more efficient.”

One approach to boost efficiency is to use multimodal AI — an effort that ECE assistant professor Yao Qin is leading for REAL AI. Multimodal AI would take advantage of brain images combined with other data, such as medical records, patient notes, and text associated with the images, that could provide more context. In addition, Miolane plans to use generative AI for well-understood aspects of the brain — such as the patterns of atrophy in Alzheimer’s disease — to develop synthetic data that can then be modeled; the lead on these generative approaches is ECE assistant professor Haewon Jeong. As the director and principal investigator of the UCSB Geometric Intelligence Lab, Miolane is working to redesign the building blocks of AI models to take advantage of the structure of data, as well as the amount of it.  

Miolane is already combining these strategies to study the brain of pregnant and post-partum women, research that she started pursuing as she wondered how changing hormones in pregnancy affected the brain. “I was in my pregnancy, waiting for this tsunami of hormones, and also waiting for the hormonal crash that follows birth,” she said. “And it’s not only a question for me, but for the approximately 200 million women who become pregnant each year, one in five of whom suffer from a post-partum disorder.”

Now, she’s developed an app that can chart what’s happening in the pregnant brain during and after pregnancy. When the app is released in 2027, users will be able to enter the gestational week of their pregnancy, and view the anatomical changes happening in the brain at that time for an average pregnancy. A user who enters additional data might some day see changes unfold that more closely approximate how their own brain is changing throughout pregnancy. 

Looking at the 3D model of the brain, a user who is fourteen weeks pregnant and experiencing brain fog may see, for example, that the hippocampus is losing some volume. Miolane and other researchers may someday be able to use this information to better anticipate how to support people during and after pregnancy. The app may also support people right now, providing  comfort to know that the symptoms they’re feeling aren’t imagined, Miolane said, but are, quite literally, physically, “in your head.” 

Miolane’s digital twins have applications beyond pregnancy, allowing people to make more informed decisions about their health. She noted that running scenarios of such outcomes is the norm when people are making financial decisions, such as buying a house. But when it comes to health care, you might get some prescription options from your doctor, try a particular medication, and then wait to see how it works. “You’re experimenting with yourself,” she said, “which is not very efficient, but also probably not very healthy.”