The Interstate of Science: Merging Neuroscience and AI
Enter a world of possibilities: generate ideas, get answers, have conversations, create endless images. Often, artificial intelligence or large language models, like ChatGPT, Gemini, or Claude, can seem almost human-like. But this technology is still far from accurately representing the human brain. No matter how sophisticated the tools seem, AI does not think for itself, nor are its connections as deeply complex as the brain. However, AI is in an important feedback loop with Neuroscience. Advancing the questions being asked, accelerating technology, and fostering discovery.
Building Brain Tools
Our senses, emotions, and memories make us uniquely human. How we take in information from the outside world and process it through our eyes, ears, nose, mouth, and fingertips is still very much a mystery that neuroscientists and other researchers are trying to solve, and as we learn more, better model systems can be built to answer new questions at an accelerated rate.
Discovering how neurons in the auditory cortex respond to and process sound to build better computational models of speech aligns with individual neuron research being conducted by Samuel Norman-Haignere, PhD, an associate professor of Neuroscience and Biostatistics at the University of Rochester. Norman-Haignere’s research is leveraging techniques from AI to build better computational models that can predict how the human brain codes complex sounds such as speech and music. His lab collects precise, largescale data from the human brain to train better computational models, and also conducts experiments on models to generate predictions for the next generation of auditory neuroscience experiments, while testing whether the computations of existing models align with the brain.
The scientific loop between computation and model systems is exemplified in the FlyWire Connectome. It is a map of every neuron and synaptic connection in the central brain of Drosophila Melanogaster, or the fruit fly. Assistant Professor of Biomedical Genetics and Neuroscience, Gabriella Sterne, PhD, contributed to this research and has since used the neural connections mapped in the connectome to accelerate findings in the living fruit fly brain. For instance, she was part of a team that found there’s an overlap in circuitry for different tastes, something that would have taken years, if not an entire career, to discover if a computational model based on the connectome had not pointed researchers in the right direction. “The connectome could not be completed without machine learning,” said Sterne. “These findings showed that connectome-based models can predict features of circuits that are non-intuitive, which can then be confirmed experimentally.”
The Gears of Collaboration
Krishnan Padmanabhan, PhD, an associate professor of Neuroscience, conducts fundamental research on the relationship between the olfactory system and memory. It has applications in understanding several brain diseases, as disruptions in smell are often a symptom of neurodegenerative disorders. He and Associate Professor of Neuroscience Julian Meeks, PhD, use machine learning to sift through large data analysis to better understand this relationship.
Padmanabhan also collaborates with Gourab Ghoshal, PhD, professor of Physics and computer science, to develop new ways to explore what computation in the brain means and how it may provide insight into how the brain works. “Rather than studying a specific brain region, my group focuses on how patterns of activity in large networks encode and transmit meaningful information, and how these dynamics support coherent behavior,” Ghoshal said. “This has relevance both for understanding biological brains and for developing new computational principles that may inform future AI systems. The broader field is seeing a lot of crosspollination with AI. Tools from machine learning increasingly shape how we analyze neural data, while ideas from neuroscience, such as recurrent dynamics, modular organization, and energy-based structures, continue to influence AI architectures.”
That deep understanding can be modeled in a tool outside of the brain, and recognizing patterns based on previously shared information is a strength in current AI systems; however, despite being able to recognize patterns, AI does not have the ability to know where the patterns came from. Adam Snyder, PhD, an assistant professor of Neuroscience and Brain and Cognitive Sciences, studies cellular relationships in the biological brain. He and Ralf Haefner, PhD, associate professor of Brain and Cognitive Sciences, are investigating how learning progresses in the brain and how understanding is shared across many neurons. “This suggests that future systems may benefit from moving beyond recognition alone, toward internal models that can explain, predict, and reason about their inputs,” Snyder said.
The AI of today is far from this point. But as Haefner points out, their research is a step toward changing that. “Even with all the advances in experimental techniques, it will be a long time until we have enough data to constrain all the parameters that such a model requires. But our research points to a crucial role of feedback connections, which are mostly missing in modern AI systems, including models of the brain.”
Machine Learning of Yesterday and Today
Machine learning’s roots can be traced back to the 1940s, when Warren McCulloch and Walter Pitts introduced the first mathematical model of a neural network. It has transformed over the last 80 years, and over the past several decades, it has accelerated into a tool that is rapidly
advancing science.
In recent history, researchers known as AI’s “godfathers,” Geoff Hinton, Yoshua Bengio, and Yann LeCun, took much inspiration from neuroscience to create AI models. “Many of the biggest advancements in AI have been inspired by the brain. In my opinion, the biggest challenge is that many researchers today in AI are not interested in neuroscience and think it has nothing left to give; however, I strongly disagree,” said Christopher Kanan, PhD, associate professor of Computer Science and associate director for AI Strategy, Goergen Institute for Data Science and Artificial Intelligence.
Kanan’s research aims to enable AI models to learn over time—mimicking the brains of humans and other animals. “I take a lot of inspiration from the memory consolidation mechanisms that happen during sleep, and specifically the role of the hippocampus during NREM (non-rapid
eye movement) sleep and the impact of REM (rapid eye movement) on improving neural representations.” In the visual system, Kanan’s research incorporates mechanisms related to the prefrontal cortex into deep neural networks. He draws inspiration from neural architecture to overcome the limitations of today’s large language and vision-languagemodels, such as ChatGPT.
In 2025, the University joined the Empire AI Consortium, a group of other public and private research institutions in New York State that aims to bridge the gap between researchers, public interest organizations, and small companies to accelerate the development of artificial intelligence centered in public interest.
AI-ccelerating Science
In 2012, Maiken Needergard, MD, DMSc, co-director of the University’s Center for Translational Neuromedicine, discovered the glymphatic system. The discovery of this system, which clears waste away from the brain as we sleep, has accelerated our understanding of fluid flow, sleep, and their relationship to human health and disease. It has also created a need in the neuroscience community to dive deeper into how it and other fluids move through the brain.
Moving Research to Clinic
Advancements in machine learning to analyze large datasets are accelerating research by Frank Garcea, PhD, assistant professor of Neurosurgery, and Michelle Janelsins, PhD, MPH, the Joan and Gary Morrow Endowed Distinguished Professor of Supportive Care in Cancer at the Wilmot Cancer Institute. Working with cancer patients through the Program for Translational Brain Mapping, Garcea, Janelsins, and MD/PhD student Emma Strawderman have discovered that brain networks in the right hemisphere rewire in response to a tumor in the left hemisphere—and that patterns of rewiring before surgery can predict whether a patient will have a deficit in fluent speech after surgery.
Garcea sees this technology as a promising clinical tool, but he emphasizes the need for caution. “There are real risks if we change how we use this technology to predict outcomes,” he said. “For example, not all rewiring matters the same—how the right hemisphere vision network rewires before surgery does not predict whether a patient will have a deficit in fluent speech after surgery, but how the right hemisphere language network rewires does.”
He emphasizes that with each technological advancement comes new risks to consider in his work, and across science. “Careful evaluation of training data and prediction accuracy is essential, underscoring the importance of keeping human expertise central to the use of these tools in both research and clinical decision making.”
