Cristina Savin’s Artificial & Biological Computation Lab and the Quest to Bridge Neural Science and AI
Unveiling the Brain’s Secrets and the Path to Next-Generation AI
In the quest to unravel the complexities of the human brain, a tool older and more sophisticated than any technology at our disposal, Cristina Savin’s Artificial & Biological Computation lab at NYU stands as a beacon of innovation. This unique intersection of the Center for Data Science (CDS) and the Center for Neural Science is where the future of artificial intelligence (AI) begins to take shape, informed by the intricate workings of neural circuits. Savin, a CDS Assistant Professor in Neural Science and Data Science, leads her team with a clear vision: to bridge the vast expanse between the computational principles of the brain and the burgeoning field of AI.
The lab’s mission is underpinned by a belief that the keys to the next leap in AI’s evolution lie within our own biology. “In very generic terms, my research focuses on learning and memory at the level of neural circuits in the brain,” Savin said in a recent interview. Her approach, she said, is twofold: on one hand, delving into theoretical neuroscience to build probabilistic models of neural computation; and on the other, constructing statistical models to analyze the joint activity of neurons. This holistic approach seeks not just to mimic but to understand — to grasp the essence of adaptability and learning that the brain exhibits.
At the heart of their investigation is last year’s “Catalyzing Next-Generation Artificial Intelligence through NeuroAI,” a manifesto-like paper that argues for a concerted investment in what the paper’s 27 authors call “NeuroAI.” This paper, co-authored by Savin, as well as CDS members Yann Lecun and Eero Simoncelli, advocates for an embodied Turing test, challenging AI systems to interact with the world with the same dexterity and adaptability as living organisms. This notion extends AI’s ambition beyond human-centric tasks, focusing instead on the broad spectrum of capabilities honed by a broader range of animals through millions of years of evolution. “Building models that can pass the embodied Turing test will provide a roadmap for the next generation of AI,” the paper’s authors posited, setting the stage for a radical reimagining of AI’s potential.
A Workshop at the Frontier of Neuroscience and Machine Learning
Complementing their groundbreaking research, Savin’s lab is poised to host the “I Can’t Believe It’s Not Better! COSYNE 2024 Edition” workshop at the Cosyne conference in Lisbon this weekend. This initiative was designed to foster a critical dialogue on the limitations and future prospects of machine learning tools in neuroscience. The workshop aims to challenge the status quo, pushing for a collective reevaluation of current methodologies and a search for innovative approaches that could more accurately model brain function.
Inspired by the broader “I Can’t Believe It’s Not Better” (ICBINB) initiative, the workshop seeks to highlight the importance of understanding and embracing the limitations of current machine learning tools. “There has been a rapid proliferation of data-driven models in our community over the past 5–10 years,” Savin and her co-organizers observed, noting, however, the growing concern over potential stagnation in the efficacy of these tools within neuroscience. The workshop discussions are poised to interrogate these fears head-on, questioning whether the community is constrained by the tools, hypotheses, or data available.
Bridging Theory and Experiment: A Synergistic Approach
The pursuit within Savin’s “ABC Lab,” as they call it, is deeply rooted in a symbiotic relationship between theory and experimental data. This interplay is crucial, as it allows the team to test the waters of their computational models against the complex realities of neural activity. Through partnerships with experimental labs, both within NYU and globally, Savin’s team has embarked on a journey to understand how brains learn and adapt, leveraging massive datasets that depict neural activity in unprecedented detail.
This rigorous approach is exemplified in their work on developing new statistical methodologies to analyze neural data. “We ended up developing quite a bit of statistical methodology for neural data analysis as a necessary step towards achieving our scientific goals,” Savin said. One of their notable projects, “A Probabilistic Framework for Task-Aligned Intra- and Inter-Area Neural Manifold Estimation,” published last year, introduces a novel tool for dissecting the complex interplay between different brain regions during task performance. This work, emblematic of the lab’s commitment to innovation, allows researchers to partition neural variability into components that are either shared across areas or unique to each, offering new insights into how the brain integrates information.
Practical Applications and the Quest for Understanding
While the lab’s research is driven by fundamental questions about the nature of intelligence and learning, it also harbors potential for real-world applications. The development of statistical tools for brain-machine interfaces represents a direct avenue through which their work could revolutionize clinical practices. “Some of the statistical data analysis tools that we’re developing have a potential clinical application in prosthetics, and other brain computer interface applications,” Savin shared.
Moreover, the lab’s endeavors in understanding the neural underpinnings of sensory information processing and decision-making have profound implications. For instance, their work in “Transformation of acoustic information to sensory decision variables in the parietal cortex” — another 2023 paper — reveals the intricate mechanisms through which the brain interprets and acts upon sensory data. Such insights not only advance our understanding of the brain’s functionality but also inform the development of AI systems capable of more nuanced interactions with their environment.
The Vision Forward: A Convergence of Disciplines
Cristina Savin and her team at NYU’s Artificial & Biological Computation lab are not just conducting research; they are building bridges. Between the realms of the known and the unknown, between theoretical models and experimental realities, and between the capabilities of the human brain and the potential of artificial intelligence, they are laying the groundwork for a future where the mysteries of the brain inform the next generation of AI. This is the vision that guides their work — a vision that promises to redefine our understanding of both intelligence and the machines we build to mimic it.
By Stephen Thomas
